← PART TWO: REBATES, ROCKETS, AND ROBOTS

THE OIL TRACKER · UPDATED NOVEMBER 16, 2026

The oil tracker: are the bets moving?

Part Two of the oil essay scored twenty hypotheses on October 31, 2026. This page keeps scoring them as new numbers come out: who is ahead today, which leading indicators are moving, and whether the trend is toward the author’s call or away from it.

22 HYPOTHESES11 TOWARD · 9 FLAT · 1 AWAY · 1 TOO EARLYREFRESHED WEEKLYNOT INVESTMENT ADVICE

A verdict is a snapshot. A bet is about direction. This tracker follows the bets.

The premise

Part Two’s synthesis found a split. The hypotheses about direction mostly held: orbit matters, robots are the likeliest gasoline moment, Meta was underweighted, the chain shifts, breadth protects. The hypotheses about who is ahead right now mostly didn’t. Waymo, not Tesla, leads robotaxis. Google, not OpenAI, sells the ads. Nvidia and the clouds, not the volume-controlling labs, keep the margin.

It found a third thing too: the riskiest place in the chain is debt-financed infrastructure with one big customer, and that now includes the supplier balance sheets, Nvidia’s above all, that have guaranteed it.

The tracker takes each hypothesis, states it as a bet on direction, and follows the leading indicators that would show it coming true or failing. Each card says who leads today, lists dated and sourced indicators, and makes a trend call. The call is relative to the author’s bet, not to the market’s mood.

THE AUTHOR’S POSITION

The author’s view is that the “who is ahead” hypotheses were bets on where things are headed, not claims about today. He thinks the trend is moving his way: Tesla has a structural cost advantage over Waymo, and OpenAI is structurally better placed than Google to sell search ads in the future.

That is a fair thing to test, and a fair thing to be wrong about. So each card scores trajectory, not just the snapshot. Where the indicators move away from the call, the card says so.

All bets at a glance

TRACKER SUMMARYTrend relative to the author’s call · as of Nov 16, 2026
HypothesisToday’s leaderTrendLast updated
1. Orbital data centers deserve to be taken seriouslySpaceX (on paper)MOVING TOWARDNov 9, 2026
2. Launch becomes the new railroad, with an Nvidia-like moatSpaceXMOVING TOWARDNov 16, 2026
3. A lunar mass driver is the pipeline of orbital computeNo one; there is no lunar industry yetTOO EARLYNov 2, 2026
4. Robots are AI’s gasoline momentWaymo (robotaxis); Chinese makers (humanoids)MOVING TOWARDNov 16, 2026
5. Robots are underweighted outside a small groupCapital is not underweightingMOVING AWAYNov 2, 2026
6. Tesla overtakes Waymo in robotaxisWaymoFLAT / MIXEDNov 16, 2026
7. Google is Standard Oil after its primeGoogleFLAT / MIXEDNov 2, 2026
8. OpenAI wins AI advertisingGoogleFLAT / MIXEDNov 16, 2026
9. Amazon wins agentic shoppingAmazonMOVING TOWARDNov 16, 2026
10. Meta was underweightedMetaMOVING TOWARDNov 16, 2026
11. The interface is more Neuralink than glassesGlasses (Meta / EssilorLuxottica)FLAT / MIXEDNov 2, 2026
12. Meta’s hardware record is weaker than Musk’sMusk’s companiesFLAT / MIXEDNov 2, 2026
13. Volume-controlling labs are Standard; the clouds are railroadsNvidia and the cloudsFLAT / MIXEDNov 16, 2026
14. Someone consolidates the neocloudsNo consolidator yet (lenders and Nvidia hold the leverage)MOVING TOWARDNov 16, 2026
15. Oracle is the Pennsylvania Railroad overplaying its handOverextension fits; the Empire move is others’MOVING TOWARDNov 16, 2026
16. Elon is RockefellerIntegration growing; as a landlordFLAT / MIXEDNov 2, 2026
17. Nvidia moves across the stackNvidiaMOVING TOWARDNov 16, 2026
18. Revenue breadth is survivalThe diversified majorsMOVING TOWARDNov 16, 2026
19. The riskiest link: debt with one big customer, now guaranteed by suppliersStress is where predictedMOVING TOWARDNov 16, 2026
20. Pure token production loses value fastSplit: commodity tokens fall, frontier tokens holdFLAT / MIXEDNov 16, 2026
21. AI runs oil-style boom-and-bust cyclesUpswing: memory and power are the scarce inputsMOVING TOWARDNov 16, 2026
22. Consumer AI settles at $50–100 a person a monthSubscriptions, ads and commerce togetherFLAT / MIXEDNov 16, 2026

The two big bets

CHART 1 · ROBOTAXIS

Robotaxi rides: Waymo vs Tesla

Top: Waymo paid rides per week, linear scale. Middle: cumulative driverless miles, log scale (each gridline is 10×). Bottom: Apollo Go fully driverless rides per quarter, linear scale. Each panel has one y-axis; dates share one time axis.

  • Official (company / filing)
  • Secondhand or estimate

Waymo paid rides per week

As announced by Waymo / Alphabet; dated by announcement

0200K400K600K2023202420252026Tesla has not disclosed a comparableweekly paid-ride number.Waymo · 2024-08-20 · 100K · Secondhand · TechCrunch, “Waymo is now giving 100,000 robotaxi rides a week” (Aug 20, 2024) · Tekedra Mawakana post on X, as reportedWaymo · 2024-10-29 · 150K · Official · Sundar Pichai, Alphabet Q3 2024 earnings remarks (Oct 29, 2024)Waymo · 2024-12-18 · 150K · Official · Waymo, 2024 Year in Review (Dec 18, 2024)Waymo · 2025-02-27 · 200K · Secondhand · TechCrunch, “Waymo has doubled its weekly robotaxi rides in less than a year” (Feb 27, 2025) · Sundar Pichai post on X, as reportedWaymo · 2025-04-24 · 250K · Official · Sundar Pichai, Alphabet Q1 2025 earnings remarks (Apr 24, 2025) · “over a quarter of a million” paid rides per weekWaymo · 2026-02-02 · 400K · Official · Waymo, $16B investment round announcement (Feb 2, 2026) · “over 400,000”; repeated in Pichai’s Q4 2025 remarks, Feb 4, 2026Waymo · 2026-03-27 · 500K · Secondhand · TechCrunch, “Waymo’s skyrocketing ridership in one chart” (Mar 27, 2026) · Waymo post on X (“this week”), as reported; exact post date not confirmedWaymo · 2026-04-29 · 500K · Official · Sundar Pichai, Alphabet Q1 2026 earnings remarks (Apr 29, 2026) · “over 500,000”. No weekly figure in the Q2 (Jul 22) or Q3 (Oct 28) 2026 remarks. Waymo’s 1M/week by end-2026 is a target, not plotted.Waymo 500K

Cumulative miles, log scale

Waymo rider-only miles vs Tesla robotaxi miles (Tesla’s only consistent disclosure)

  • Waymo autonomous
  • Tesla paid
  • Tesla unsupervised
100K1M10M100M1B2023202420252026Waymo autonomous · 2023-10-31 · 7.1M · Official · Waymo, 7.1 million rider-only miles safety study (Dec 2023) · “7.14 million fully autonomous miles … through the end of October 2023”Waymo autonomous · 2025-07-23 · 100M · Official · Sundar Pichai, Alphabet Q2 2025 earnings remarks (Jul 23, 2025) · “over 100 million” fully autonomous milesWaymo autonomous · 2026-02-02 · 127M · Official · Waymo, $16B investment round announcement (Feb 2, 2026) · 127M miles as stated in the post; may lag the live total (a >200M figure on Feb 24, 2026 is search-snippet only and not plotted)Waymo autonomous · 2026-06-30 · 271M · Official · Waymo Safety Impact hub (data through June 2026) · Rider-only miles through June 2026; hub release dated Oct 2, 2026Waymo autonomous · 2026-10-28 · 300M · Official · Sundar Pichai, Alphabet Q3 2026 earnings remarks (Oct 28, 2026) · “over 300 million” autonomous miles, 15 US citiesWaymo autonomous 300MTesla paid · 2025-12-31 · 610K · Secondhand · Drive Tesla Canada, “Tesla Robotaxi usage surges in Q1 2026 as paid miles nearly triple” (Apr 22, 2026) · ~610K read off the Q4 2025 deck chart (no printed value in Tesla’s deck)Tesla paid · 2026-03-31 · 1.7M · Secondhand · Drive Tesla Canada, “Tesla Robotaxi usage surges in Q1 2026 as paid miles nearly triple” (Apr 22, 2026) · ~1.7M read off the Q1 2026 deck chart; Tesla’s text says paid miles “nearly doubled sequentially”Tesla paid · 2026-06-30 · 2.4M · Secondhand · EVwire, “Tesla robotaxi: 2.4 million paid miles, six metros” (Jul 23, 2026) · ~2.4M read off the Q2 2026 deck chart. The Q3 2026 deck gives no paid-miles figure.Tesla paid 2.4MTesla unsupervised · 2026-09-30 · 2M · Official · Tesla Q3 2026 Update (SEC exhibit, Oct 21, 2026) · “nearly two million” cumulative unsupervised Robotaxi miles, 8× QoQ; plotted at 2MTesla unsupervised · 2026-10-21 · 3.1M · Secondhand · TeslaNorth: Tesla’s new Robotaxi Safety Report (Oct 21, 2026) · Tesla Robotaxi Safety Report (first published Oct 21): 3.1M driverless miles through Oct 18, counting only miles with no safety monitor inside (incl. remote-assisted trips), as reported by TeslaNorth; tesla.com blocks our fetcher. Same concept as the Q3 deck’s unsupervised miles is likely but not confirmed.Tesla unsupervised · 2026-11-09 · 4.8M · Estimate · robotaxistats: Tesla robotaxi safety figures (Tesla live counter read Nov 9, 2026) · Tesla’s live safety-report counter, which Tesla says estimates miles at the fleet’s current pace; read by robotaxistats Nov 9 ~3:20 AM PT. Company estimate, not a reported total.Tesla unsupervised · 2026-11-14 · 5.2M · Estimate · robotaxistats: Tesla robotaxi safety figures (Tesla live counter read Nov 14, 2026) · Tesla’s live safety-report counter, which Tesla says estimates miles at the fleet’s current pace; read by robotaxistats Nov 14 ~10:14 AM PT. Company estimate, not a reported total.Tesla unsupervised 5.2M

Apollo Go fully driverless rides per quarter

Baidu quarterly results; rides per quarter, not per week

01M2M3M4M2023202420252026Apollo Go · 2024-03-31 · 826K · Official · Baidu Q1 2024 results, 6-K EX-99.1 (May 16, 2024) · Total rides (not only fully driverless) before Q2 2025Apollo Go · 2024-06-30 · 899K · Official · Baidu Q2 2024 results, 6-K EX-99.1 (Aug 22, 2024) · Total rides (not only fully driverless) before Q2 2025Apollo Go · 2024-09-30 · 988K · Official · Baidu Q3 2024 results, 6-K EX-99.1 (Nov 21, 2024) · Total rides; Baidu said over 70% were fully driverlessApollo Go · 2024-12-31 · 1.1M · Official · Baidu Q4 2024 results, 6-K EX-99.1 (Feb 18, 2025) · “over 1.1 million” total ridesApollo Go · 2025-03-31 · 1.4M · Official · Baidu Q1 2025 results, 6-K EX-99.1 (May 21, 2025) · “over 1.4 million” rides; 100% fully driverless in China since Feb 2025 per BaiduApollo Go · 2025-06-30 · 2.2M · Official · Baidu Q2 2025 results, 6-K EX-99.1 (Aug 20, 2025) · “over 2.2 million” fully driverless ridesApollo Go · 2025-09-30 · 3.1M · Official · Baidu Q3 2025 results, 6-K EX-99.1 (Nov 18, 2025) · Fully driverless ridesApollo Go · 2025-12-31 · 3.4M · Official · Baidu Q4 2025 results, 6-K EX-99.1 (Feb 26, 2026) · Fully driverless rides; weekly peak over 300KApollo Go · 2026-03-31 · 3.2M · Official · Baidu Q1 2026 results (HKEX filing, May 18, 2026) · Fully driverless rides; weekly peak over 350K in March. Baidu gave no Q2 2026 quarterly figure (H1 2026: over 4.1M), so no Q2 point is plotted.Apollo Go 3.2M

LAST POINT: NOV 14, 2026HOVER OR TAP A DOT FOR ITS SOURCE

Waymo is the only operator that publishes a weekly paid-ride count, so it stands alone in the top panel; no Tesla line is drawn there because no Tesla number exists. Miles are the one measure both companies report, and they differ by orders of magnitude, hence the log scale. Lines join only announced points; nothing between dots is data.

CHART 1 DATAEvery plotted point, newest first
DateSeriesValueTypeSource
2026-11-14Tesla unsupervised5.2M miEstimaterobotaxistats: Tesla robotaxi safety figures (Tesla live counter read Nov 14, 2026) ↗Tesla’s live safety-report counter, which Tesla says estimates miles at the fleet’s current pace; read by robotaxistats Nov 14 ~10:14 AM PT. Company estimate, not a reported total.
2026-11-09Tesla unsupervised4.8M miEstimaterobotaxistats: Tesla robotaxi safety figures (Tesla live counter read Nov 9, 2026) ↗Tesla’s live safety-report counter, which Tesla says estimates miles at the fleet’s current pace; read by robotaxistats Nov 9 ~3:20 AM PT. Company estimate, not a reported total.
2026-10-28Waymo autonomous300M miOfficialSundar Pichai, Alphabet Q3 2026 earnings remarks (Oct 28, 2026) ↗“over 300 million” autonomous miles, 15 US cities
2026-10-21Tesla unsupervised3.1M miSecondhandTeslaNorth: Tesla’s new Robotaxi Safety Report (Oct 21, 2026) ↗Tesla Robotaxi Safety Report (first published Oct 21): 3.1M driverless miles through Oct 18, counting only miles with no safety monitor inside (incl. remote-assisted trips), as reported by TeslaNorth; tesla.com blocks our fetcher. Same concept as the Q3 deck’s unsupervised miles is likely but not confirmed.
2026-09-30Tesla unsupervised2M miOfficialTesla Q3 2026 Update (SEC exhibit, Oct 21, 2026) ↗“nearly two million” cumulative unsupervised Robotaxi miles, 8× QoQ; plotted at 2M
2026-06-30Tesla paid2.4M miSecondhandEVwire, “Tesla robotaxi: 2.4 million paid miles, six metros” (Jul 23, 2026) ↗~2.4M read off the Q2 2026 deck chart. The Q3 2026 deck gives no paid-miles figure.
2026-06-30Waymo autonomous271M miOfficialWaymo Safety Impact hub (data through June 2026) ↗Rider-only miles through June 2026; hub release dated Oct 2, 2026
2026-04-29Waymo500K paid rides/wkOfficialSundar Pichai, Alphabet Q1 2026 earnings remarks (Apr 29, 2026) ↗“over 500,000”. No weekly figure in the Q2 (Jul 22) or Q3 (Oct 28) 2026 remarks. Waymo’s 1M/week by end-2026 is a target, not plotted.
2026-03-31Apollo Go3.2M rides/qtrOfficialBaidu Q1 2026 results (HKEX filing, May 18, 2026) ↗Fully driverless rides; weekly peak over 350K in March. Baidu gave no Q2 2026 quarterly figure (H1 2026: over 4.1M), so no Q2 point is plotted.
2026-03-31Tesla paid1.7M miSecondhandDrive Tesla Canada, “Tesla Robotaxi usage surges in Q1 2026 as paid miles nearly triple” (Apr 22, 2026) ↗~1.7M read off the Q1 2026 deck chart; Tesla’s text says paid miles “nearly doubled sequentially”
2026-03-27Waymo500K paid rides/wkSecondhandTechCrunch, “Waymo’s skyrocketing ridership in one chart” (Mar 27, 2026) ↗Waymo post on X (“this week”), as reported; exact post date not confirmed
2026-02-02Waymo400K paid rides/wkOfficialWaymo, $16B investment round announcement (Feb 2, 2026) ↗“over 400,000”; repeated in Pichai’s Q4 2025 remarks, Feb 4, 2026
2026-02-02Waymo autonomous127M miOfficialWaymo, $16B investment round announcement (Feb 2, 2026) ↗127M miles as stated in the post; may lag the live total (a >200M figure on Feb 24, 2026 is search-snippet only and not plotted)
2025-12-31Apollo Go3.4M rides/qtrOfficialBaidu Q4 2025 results, 6-K EX-99.1 (Feb 26, 2026) ↗Fully driverless rides; weekly peak over 300K
2025-12-31Tesla paid610K miSecondhandDrive Tesla Canada, “Tesla Robotaxi usage surges in Q1 2026 as paid miles nearly triple” (Apr 22, 2026) ↗~610K read off the Q4 2025 deck chart (no printed value in Tesla’s deck)
2025-09-30Apollo Go3.1M rides/qtrOfficialBaidu Q3 2025 results, 6-K EX-99.1 (Nov 18, 2025) ↗Fully driverless rides
2025-07-23Waymo autonomous100M miOfficialSundar Pichai, Alphabet Q2 2025 earnings remarks (Jul 23, 2025) ↗“over 100 million” fully autonomous miles
2025-06-30Apollo Go2.2M rides/qtrOfficialBaidu Q2 2025 results, 6-K EX-99.1 (Aug 20, 2025) ↗“over 2.2 million” fully driverless rides
2025-04-24Waymo250K paid rides/wkOfficialSundar Pichai, Alphabet Q1 2025 earnings remarks (Apr 24, 2025) ↗“over a quarter of a million” paid rides per week
2025-03-31Apollo Go1.4M rides/qtrOfficialBaidu Q1 2025 results, 6-K EX-99.1 (May 21, 2025) ↗“over 1.4 million” rides; 100% fully driverless in China since Feb 2025 per Baidu
2025-02-27Waymo200K paid rides/wkSecondhandTechCrunch, “Waymo has doubled its weekly robotaxi rides in less than a year” (Feb 27, 2025) ↗Sundar Pichai post on X, as reported
2024-12-31Apollo Go1.1M rides/qtrOfficialBaidu Q4 2024 results, 6-K EX-99.1 (Feb 18, 2025) ↗“over 1.1 million” total rides
2024-12-18Waymo150K paid rides/wkOfficialWaymo, 2024 Year in Review (Dec 18, 2024) ↗
2024-10-29Waymo150K paid rides/wkOfficialSundar Pichai, Alphabet Q3 2024 earnings remarks (Oct 29, 2024) ↗
2024-09-30Apollo Go988K rides/qtrOfficialBaidu Q3 2024 results, 6-K EX-99.1 (Nov 21, 2024) ↗Total rides; Baidu said over 70% were fully driverless
2024-08-20Waymo100K paid rides/wkSecondhandTechCrunch, “Waymo is now giving 100,000 robotaxi rides a week” (Aug 20, 2024) ↗Tekedra Mawakana post on X, as reported
2024-06-30Apollo Go899K rides/qtrOfficialBaidu Q2 2024 results, 6-K EX-99.1 (Aug 22, 2024) ↗Total rides (not only fully driverless) before Q2 2025
2024-03-31Apollo Go826K rides/qtrOfficialBaidu Q1 2024 results, 6-K EX-99.1 (May 16, 2024) ↗Total rides (not only fully driverless) before Q2 2025
2023-10-31Waymo autonomous7.1M miOfficialWaymo, 7.1 million rider-only miles safety study (Dec 2023) ↗“7.14 million fully autonomous miles … through the end of October 2023”
CHART 2 · AI ADS

Ad revenue: OpenAI vs Google Search

Top: annualized US$, log scale (each gridline is 10×). Google = quarterly “Search & other” revenue × 4; OpenAI = reported annualized ad run-rate. Bottom: OpenAI’s run-rate as a % of Google’s latest annualized quarter, linear scale.

  • Official (company / filing)
  • Secondhand or estimate

Annualized ad revenue, log scale

Annualized to annualized: Google quarter × 4 vs OpenAI run-rate

  • Google Search ×4
  • OpenAI ads
$100M$1B$10B$100B$1T2023202420252026Google Search ×4 · 2022-03-31 · $158B · Official · Alphabet Q1 2022 earnings release (Apr 26, 2022)Google Search ×4 · 2022-06-30 · $163B · Official · Alphabet Q2 2022 earnings release (Jul 26, 2022)Google Search ×4 · 2022-09-30 · $158B · Official · Alphabet Q3 2022 earnings release (Oct 25, 2022)Google Search ×4 · 2022-12-31 · $170B · Official · Alphabet Q4 2022 earnings release (Feb 2, 2023)Google Search ×4 · 2023-03-31 · $161B · Official · Alphabet Q1 2023 earnings release (Apr 25, 2023)Google Search ×4 · 2023-06-30 · $171B · Official · Alphabet Q2 2023 earnings release (Jul 25, 2023)Google Search ×4 · 2023-09-30 · $176B · Official · Alphabet Q3 2023 earnings release (Oct 24, 2023)Google Search ×4 · 2023-12-31 · $192B · Official · Alphabet Q4 2023 earnings release (Jan 30, 2024)Google Search ×4 · 2024-03-31 · $185B · Official · Alphabet Q1 2025 earnings release (Apr 24, 2025) · Q1 2024 value from the prior-year column of the Q1 2025 releaseGoogle Search ×4 · 2024-06-30 · $194B · Official · Alphabet Q2 2024 earnings release (Jul 23, 2024)Google Search ×4 · 2024-09-30 · $198B · Official · Alphabet Q3 2024 earnings release (Oct 29, 2024)Google Search ×4 · 2024-12-31 · $216B · Official · Alphabet Q4 2024 earnings release (Feb 4, 2025)Google Search ×4 · 2025-03-31 · $203B · Official · Alphabet Q1 2025 earnings release (Apr 24, 2025)Google Search ×4 · 2025-06-30 · $217B · Official · Alphabet Q2 2025 earnings release (Jul 23, 2025)Google Search ×4 · 2025-09-30 · $226B · Official · Alphabet Q3 2025 earnings release (Oct 29, 2025)Google Search ×4 · 2025-12-31 · $252B · Official · Alphabet Q4 2025 earnings release (Feb 4, 2026)Google Search ×4 · 2026-03-31 · $242B · Official · Alphabet Q1 2026 earnings release (Apr 29, 2026)Google Search ×4 · 2026-06-30 · $253B · Official · Alphabet Q2 2026 earnings release (Jul 22, 2026)Google Search ×4 · 2026-09-30 · $253B · Official · Alphabet Q3 2026 earnings release (Oct 28, 2026)Google Search ×4 $253BOpenAI ads · 2026-03-27 · $100M · Secondhand · LetsDataScience, “OpenAI reaches $100M annualized ChatGPT ads” (Mar 27, 2026) · OpenAI spokesperson figure as reported; US pilot ~6 weeks after the Feb 9 launch. No OpenAI primary post found.OpenAI ads · 2026-08-31 · $1B · Secondhand · Shopifreaks, “OpenAI says its advertising business has reached a $1 billion annualized revenue run rate” (Aug 31, 2026) · OpenAI claim as reported (~200 days after launch). The Information’s Oct 2026 “roughly $1B” repeats this reading and is not plotted separately.OpenAI ads $1B

OpenAI ads as % of Google Search

OpenAI run-rate ÷ (latest reported Google quarter × 4)

0%0.1%0.2%0.3%0.4%Apr 26Jul 26Oct 26OpenAI ÷ Google · 2026-03-27 · 0.04% · Secondhand · LetsDataScience, “OpenAI reaches $100M annualized ChatGPT ads” (Mar 27, 2026) · Derived: $100M ÷ (4 × Google Search & other $63B for the quarter ended Dec 31, 2025).OpenAI ÷ Google · 2026-08-31 · 0.4% · Secondhand · Shopifreaks, “OpenAI says its advertising business has reached a $1 billion annualized revenue run rate” (Aug 31, 2026) · Derived: $1B ÷ (4 × Google Search & other $63B for the quarter ended Jun 30, 2026).OpenAI ÷ Google 0.4%

LAST POINT: SEP 30, 2026HOVER OR TAP A DOT FOR ITS SOURCE

On a shared linear axis OpenAI’s line would sit on the floor, so the top panel uses a log scale and the bottom panel shows the gap as a ratio. Google’s figures are from Alphabet’s quarterly results; OpenAI’s are company claims relayed by the press with no primary OpenAI post found, so they are drawn hollow; the ratio panel is our arithmetic on those points. Run-rates annualize a recent month or week and can overstate a full year.

CHART 2 DATAEvery plotted point, newest first
DateSeriesValueTypeSource
2026-09-30Google Search ×4$63B quarter (×4 = $253B annualized)OfficialAlphabet Q3 2026 earnings release (Oct 28, 2026) ↗
2026-08-31OpenAI ÷ Google0.4%SecondhandShopifreaks, “OpenAI says its advertising business has reached a $1 billion annualized revenue run rate” (Aug 31, 2026) ↗Derived: $1B ÷ (4 × Google Search & other $63B for the quarter ended Jun 30, 2026).
2026-08-31OpenAI ads$1B annualizedSecondhandShopifreaks, “OpenAI says its advertising business has reached a $1 billion annualized revenue run rate” (Aug 31, 2026) ↗OpenAI claim as reported (~200 days after launch). The Information’s Oct 2026 “roughly $1B” repeats this reading and is not plotted separately.
2026-06-30Google Search ×4$63B quarter (×4 = $253B annualized)OfficialAlphabet Q2 2026 earnings release (Jul 22, 2026) ↗
2026-03-31Google Search ×4$60B quarter (×4 = $242B annualized)OfficialAlphabet Q1 2026 earnings release (Apr 29, 2026) ↗
2026-03-27OpenAI ÷ Google0.04%SecondhandLetsDataScience, “OpenAI reaches $100M annualized ChatGPT ads” (Mar 27, 2026) ↗Derived: $100M ÷ (4 × Google Search & other $63B for the quarter ended Dec 31, 2025).
2026-03-27OpenAI ads$100M annualizedSecondhandLetsDataScience, “OpenAI reaches $100M annualized ChatGPT ads” (Mar 27, 2026) ↗OpenAI spokesperson figure as reported; US pilot ~6 weeks after the Feb 9 launch. No OpenAI primary post found.
2025-12-31Google Search ×4$63B quarter (×4 = $252B annualized)OfficialAlphabet Q4 2025 earnings release (Feb 4, 2026) ↗
2025-09-30Google Search ×4$57B quarter (×4 = $226B annualized)OfficialAlphabet Q3 2025 earnings release (Oct 29, 2025) ↗
2025-06-30Google Search ×4$54B quarter (×4 = $217B annualized)OfficialAlphabet Q2 2025 earnings release (Jul 23, 2025) ↗
2025-03-31Google Search ×4$51B quarter (×4 = $203B annualized)OfficialAlphabet Q1 2025 earnings release (Apr 24, 2025) ↗
2024-12-31Google Search ×4$54B quarter (×4 = $216B annualized)OfficialAlphabet Q4 2024 earnings release (Feb 4, 2025) ↗
2024-09-30Google Search ×4$49B quarter (×4 = $198B annualized)OfficialAlphabet Q3 2024 earnings release (Oct 29, 2024) ↗
2024-06-30Google Search ×4$49B quarter (×4 = $194B annualized)OfficialAlphabet Q2 2024 earnings release (Jul 23, 2024) ↗
2024-03-31Google Search ×4$46B quarter (×4 = $185B annualized)OfficialAlphabet Q1 2025 earnings release (Apr 24, 2025) ↗Q1 2024 value from the prior-year column of the Q1 2025 release
2023-12-31Google Search ×4$48B quarter (×4 = $192B annualized)OfficialAlphabet Q4 2023 earnings release (Jan 30, 2024) ↗
2023-09-30Google Search ×4$44B quarter (×4 = $176B annualized)OfficialAlphabet Q3 2023 earnings release (Oct 24, 2023) ↗
2023-06-30Google Search ×4$43B quarter (×4 = $171B annualized)OfficialAlphabet Q2 2023 earnings release (Jul 25, 2023) ↗
2023-03-31Google Search ×4$40B quarter (×4 = $161B annualized)OfficialAlphabet Q1 2023 earnings release (Apr 25, 2023) ↗
2022-12-31Google Search ×4$43B quarter (×4 = $170B annualized)OfficialAlphabet Q4 2022 earnings release (Feb 2, 2023) ↗
2022-09-30Google Search ×4$40B quarter (×4 = $158B annualized)OfficialAlphabet Q3 2022 earnings release (Oct 25, 2022) ↗
2022-06-30Google Search ×4$41B quarter (×4 = $163B annualized)OfficialAlphabet Q2 2022 earnings release (Jul 26, 2022) ↗
2022-03-31Google Search ×4$40B quarter (×4 = $158B annualized)OfficialAlphabet Q1 2022 earnings release (Apr 26, 2022) ↗
ORBIT & LAUNCHOrbit and launch

Is launch becoming the railroad, and does orbit become a place to compute? Plus a new, much longer bet on the Moon.

BET 01 · PART TWO VERDICT: HOLDSMOVING TOWARD

Orbital data centers deserve to be taken seriously

THE BET

Space data centers are a serious potential outcome for AI compute, not a stunt. Part One undersold them.

WHERE IT STANDS TODAY · LEADER: SPACEX (ON PAPER)

No one runs orbital compute at scale. SpaceX’s application for up to one million data-center satellites (filed Jan 30, 2026) is still before the FCC. Shotwell says Starmind compute satellites fly in 2027; the S-1/A says “as early as 2028.” Rivals are flying prototypes: Google’s Suncatcher (four TPUs) on Oct 1, 2026 and eight Chinese computing satellites on Oct 14. An independent estimate (Feb 2026) puts orbital compute at about three times the cost of ground. On the ground, the wall is getting political: after the midterms, Goldman Sachs estimates moratoriums could affect about 45% of planned US data-center capacity.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
FCC ruling on SpaceX’s 1M-satellite applicationStill under reviewas of Oct 1, 2026Filed; public notice issuedFeb 4, 2026● FLATWilmerHale, “The legal reality of orbital data centers” (Oct 1, 2026) ↗No grant found as of Nov 1
First Starmind compute-satellite launch (target)2027 (Shotwell); “as early as 2028” (S-1/A)EST.as of Oct 20, 2026“Next year” (Musk, Q2 call)Aug 4, 2026● FLATSpaceX Q3 2026 earnings call transcript (Motley Fool) ↗Company targets
Rival compute prototypes in orbitGoogle Suncatcher (4 TPUs); China’s second batch (8 computing satellites)as of Oct 14, 2026Starcloud-1 (one H100); China’s first batch of 12Nov 2, 2025▲ TOWARDCGTN: China launches eight satellites for space-based computing (Oct 14, 2026) ↗Suncatcher: GPB/AP, Oct 1, 2026
Orbital vs ground cost, 1 GW over 5 years~3x ($51B vs $16B)EST.as of Feb 26, 2026—▼ AWAYIEEE Spectrum on the McCalip orbital-vs-ground calculator (Feb 26, 2026) ↗McCalip calculator estimate
Google’s own Suncatcher lead on cost“I don’t see this being something where it’s cheaper to do this in the next five years”as of Oct 1, 2026—▼ AWAYGPB/AP: Google launches Project Suncatcher prototype (Oct 1, 2026) ↗Travis Beals, quoted by AP
Texas large-load interconnection requests (the ground-side power wall)~474 GW, ~90% data centersas of Jun 2026—▲ TOWARDPOWER: Abbott orders audit of Texas data-center interconnection queue (Aug 4, 2026) ↗ERCOT testimony; ERCOT’s September report shows only 9,456 MW approved to energize
US data-center capacity at risk from construction pauses~45% of planned US capacity (Goldman Sachs); ~30% of 2028 additions (UBS)EST.as of Nov 6, 2026—▲ TOWARDCNBC: The AI ‘credit card’ is getting costlier as CoreWeave bond yield tops 16% (Nov 6, 2026) ↗Bank estimates after the midterm results. Goldman via CNBC (Nov 6), “with a large amount of that in Texas”; UBS via Reuters (Nov 5). Texas governor-elect Gina Hinojosa wants a moratorium until the legislature passes protections

TREND CALLToward on seriousness, not economics. More players are putting real hardware in orbit, and the ground-side power wall is measured in hundreds of gigawatts. The cost math hasn’t moved: Google’s own project lead doesn’t expect orbit to be cheaper within five years, and the FCC hasn’t ruled. Stays “toward” while prototypes keep flying; moves to “away” if Starmind slips past 2028.

NOT YET SOURCEDAny measured cost per kW of compute in orbit

UPDATED NOV 9, 2026Part Two: Musk’s case →Part Two: The case against →Part Two: Six ways it could go →
BET 02 · PART TWO VERDICT: PARTLYMOVING TOWARD

Launch becomes the new railroad, with an Nvidia-like moat

THE BET

Launch is the railroad of orbital compute, and SpaceX’s lead will last much longer than most AI positions, more like Nvidia’s than like the OpenAI-versus-Anthropic race.

WHERE IT STANDS TODAY · LEADER: SPACEX

SpaceX says it has launched more than 80% of the world’s mass to orbit every year since 2023 (Musk said almost 90% on Oct 20); BryceTech’s spacecraft-upmass data give about 76% for H1 2026. Crowd-sourced counts have SpaceX flying about 122 of roughly 254 orbital launches worldwide through October. Starship first reached orbit on Sep 28, 2026. On Flight 15 (Nov 13–14) it reached orbit again, deployed 26 Starlink V3 satellites and returned to Starbase, where the tower caught the ship for the first time; the Super Heavy booster aborted its catch, diverted offshore and was destroyed. Propellant transfer has not been attempted. Launch still costs an estimated $1,500–2,500 per kilogram (May 2026) against the under-$100 Starship target.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
SpaceX share of spacecraft upmass (BryceTech)67.0% (Q2 2026)EST.as of Jun 30, 202685.9% (Q1 2026)Mar 31, 2026▼ AWAYBryceTech, Bryce Briefing Q2 2026 (PDF) ↗Derived from Bryce upmass; H1 ≈ 76%. Musk claims ~90% of total mass (Oct 20). Quarterly swings are large
SpaceX mass to orbit per year2,213 t (2025)as of Dec 31, 20251,699 t (2024)Dec 31, 2024▲ TOWARDSpaceX S-1/A (SEC, Jun 3, 2026) ↗
SpaceX share of 2026 orbital launches~122 of ~254 (~48%)EST.as of Oct 31, 202651% (165 of 325, 2025, BryceTech)Dec 31, 2025● FLATWikipedia: 2026 in spaceflight (crowd-sourced tally, through Oct 31) ↗Crowd-sourced tally; counts differ by source (global 253–256)
Starship statusFlight 15: orbit, 26 Starlink V3 deployed, 14 orbits, ship returned and caughtas of Nov 14, 2026Reached orbit on Flight 14; 26 Starlink V3 deployedSep 28, 2026▲ TOWARDSpaceX: Starship Flight 15 (Nov 13–14, 2026; page read Nov 16) ↗Lifted off Nov 13, caught Nov 14 at 3:20 p.m. CST (Ship 42, Booster 22); orbit count and catch time from Spaceflight Now (Nov 13)
Ship catchShip caught by the tower on Flight 15, the first upper-stage catch; Super Heavy aborted its catch, diverted offshore and was destroyedas of Nov 14, 2026Not yet; Flight 15 to attempt the first catch of both Starship and Super Heavy, NET Nov 13Nov 9, 2026▲ TOWARDSpaceX: Starship Flight 15 (Nov 13–14, 2026; page read Nov 16) ↗Spaceflight Now: “It was the first time the upper stage of an orbital launch vehicle has returned to Earth and a crucial step to making Starship a fully reusable rocket.” The booster’s boostback burn ended early on low fuel mass; root cause under investigation. Three earlier boosters were caught, but no V3 Super Heavy yet
On-orbit propellant transferNot yet attempted; attempt planned “early next year”as of Oct 20, 2026—● FLATSpaceX Q3 2026 earnings call transcript (Motley Fool) ↗
Launch cost per kg today~$1,500/kg (Forethought) to ~$2,500/kg (Dwarkesh)EST.as of May 22, 2026—● FLATForethought, “Will we really put data centers in space?” (May 22, 2026) ↗Estimates; Starship target <$100/kg; rideshare list ~$7,000/kg
Nearest heavy-lift rivalNew Glenn grounded after May 28 hot-fire anomaly; return targeted by year-endas of Oct 11, 2026—▲ TOWARDNew Space Economy: Blue Moon and New Glenn status (Oct 11, 2026) ↗

TREND CALLThe railroad half is moving toward the call: SpaceX keeps growing tonnage, Starship reached orbit, and the nearest rival is grounded. Cautions: BryceTech’s upmass share fell sharply in Q2, and independent counts run well below Musk’s 90%. The orbital-compute half needs a roughly tenfold cut in cost per kilogram, which needs full Starship reuse and refueling. Flight 15’s ship catch was the next step toward; the booster catch failed, and refueling hasn’t been tried. Next: a V3 booster catch and the propellant-transfer attempt planned for early 2027.

NOT YET SOURCEDBryceTech Q3 2026 briefing (not yet published) · Independent 2026 mass-to-orbit share for all payloads · Starship V3 payload mass demonstrated

UPDATED NOV 16, 2026Part Two: How long would it last? →Part One: The railroads →
BET 03 · NEW HYPOTHESISTOO EARLY

A lunar mass driver is the pipeline of orbital compute

THE BET

Rockets are the railroad; a mass driver on the Moon is the pipeline. Pipelines displaced railroads for crude because they were fixed infrastructure with near-zero marginal cost per barrel (the Tidewater line, 1879). An electromagnetic launcher on the Moon would do the same for the orbital-compute supply chain, launching satellites built from lunar material without rockets.

WHERE IT STANDS TODAY · LEADER: NO ONE; THERE IS NO LUNAR INDUSTRY YET

This is SpaceX’s stated plan, not just a podcast idea. Its S-1/A (Jun 3, 2026) defines a “lunar mass driver” as “a launch system that we intend to build on the Moon’s surface” and says lunar satellite production plus a mass driver could enable “a petawatt-scale AI constellation.” No design, site, contract or hardware has been disclosed. The prerequisites are early: Starship first reached orbit on Sep 28, 2026; propellant transfer is first attempted in early 2027; NASA’s first crewed landing is now Artemis IV in 2028; and NASA’s 100-kilowatt lunar reactor is a 2030 requirement with no flight contract.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
SpaceX filings on a lunar mass driverS-1/A defines it and states the intent to build oneas of Jun 3, 2026—▲ TOWARDSpaceX S-1/A (SEC, Jun 3, 2026) ↗Stated intent; S-1/A also says these efforts “may not achieve commercial viability”
Musk statements“We’re going to build the factories on the moon.”as of Aug 4, 2026“You want to have a mass driver on the moon”Feb 5, 2026▲ TOWARDFortune: Musk details plan to make the Moon a manufacturing site (Aug 6, 2026) ↗Q2 call, quoted by Fortune; prior from the Dwarkesh podcast
Mass-driver design, site, contracts or hardwareNone disclosedas of Nov 1, 2026—● FLATNot a Tesla App: Understanding SpaceX’s lunar mass driver (Mar 24, 2026) ↗
Starship to orbit (prerequisite)First orbital flightas of Sep 28, 2026—▲ TOWARDAstronomy: Starship reaches orbit on 14th test flight (Sep 30, 2026) ↗
On-orbit propellant transfer (prerequisite)Not yet demonstrated; attempt early 2027as of Oct 20, 2026—● FLATSpaceX Q3 2026 earnings call transcript (Motley Fool) ↗
First crewed lunar landing (NASA)Artemis IV, 2028; Artemis III recast as an Earth-orbit docking test, summer 2027EST.as of Oct 23, 2026Artemis III planned as the landingJan 2026▼ AWAYSpectrum News: NASA gives update to Artemis III mission (Oct 23, 2026) ↗NASA targets; Artemis II flew Apr 1–10, 2026
Lunar surface powerNASA requirement: ≥100 kWe reactor ready for launch by 2030; no flight-hardware contract foundas of Feb 27, 2026—● FLATNEI Magazine: NASA Act mandates Moon reactor (Feb 27, 2026) ↗
Lunar resource extraction (ISRU)Interlune helium-3/hydrogen extraction payload, ready for launch in 2028 ($6.9M NASA contract)as of May 4, 2026—● FLATInterlune: $6.9M NASA contract for lunar resource extraction (May 4, 2026) ↗No on-Moon oxygen-extraction demo found for 2026
China’s lunar programChang’e-7 launch postponed; no new dateas of Aug 23, 2026—● FLATAP via News4Jax: China postpones Chang’e-7 launch (Aug 23, 2026) ↗

TREND CALLToo early, said plainly. The analogy is apt: Tidewater was fixed plant that made each extra barrel almost free to move. But it was built because the Bradford field was already pumping crude in volume. A mass driver pays only once there is lunar mining and manufacturing to ship, so it comes after Starship works, not instead of it. Musk himself says Earth launch can reach about a terawatt a year before the Moon is needed. What can be tracked today are the prerequisites (refueling, landings, surface power, resource extraction) and whether SpaceX moves from statements to hardware.

NOT YET SOURCEDAny engineering estimate of a lunar mass driver’s cost or throughput · NASA HLS propellant-transfer milestone date from a primary NASA document

UPDATED NOV 2, 2026Part One: Pipelines: routing around the rails →Part Two: How long would it last? →
ROBOTS & ROBOTAXISRobots and robotaxis

The gasoline-moment candidates, and the head-to-head between Waymo’s lead and Tesla’s cost structure.

BET 04 · PART TWO VERDICT: PLAUSIBLEMOVING TOWARD

Robots are AI’s gasoline moment

THE BET

Robotaxis and robots, not chat, are the new use that turns out much bigger than the original, the way the car made gasoline matter more than kerosene.

WHERE IT STANDS TODAY · LEADER: WAYMO (ROBOTAXIS); CHINESE MAKERS (HUMANOIDS)

Waymo reached 500,000 paid rides a week in March 2026, double its April 2025 rate, and runs in 15 US metros. IDC estimates nearly 25,000 humanoid robots shipped worldwide in H1 2026, up 432%, and Smart Analytics Global estimates about 16,500 in Q3, up 347%. Unit costs are falling: Morgan Stanley puts Waymo’s Ojai at about $125,000 all-in (May 2026) versus about $200,000 for the Jaguar it replaces. China’s leader, Apollo Go, shrank in Q2 2026 after a permit freeze; Baidu’s Q3 release (Nov 16) puts it in 30 cities, with Shenzhen driverless rides up about 170% on the quarter, but gives no total ride count.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Waymo paid rides per week500,000+as of Mar 26, 2026250,000Apr 2025▲ TOWARDInsideEVs: Waymo doubles to 500,000 paid rides a week (Mar 26, 2026) ↗No newer company figure as of Nov 1
Global humanoid shipments, H1 2026~25,000 (+432% y/y)EST.as of Jun 30, 2026—▲ TOWARDIDC via Webull: H1 2026 humanoid shipments near 25,000 (Sep 29, 2026) ↗IDC estimate; Counterpoint >22,000, SAG ~19,100
Global humanoid shipments, Q3 2026 (SAG)~16,500 (+347% y/y)EST.as of Sep 30, 2026—▲ TOWARDSmart Analytics Global: Q3 2026 humanoid robot shipments (Nov 8, 2026) ↗Smart Analytics Global estimate, a different firm from IDC’s H1 figure, so the two can’t be added. AGIBOT 4,900, UBTech 4,500, Unitree ~4,400; Chinese vendors ~97%. SAG cut its 2026 unit forecast (~54,000) citing US restrictions, while raising its revenue outlook (~US$1.9B)
Robotaxi vehicle cost (Waymo, all-in)~$125,000 (6th-gen Ojai)EST.as of May 28, 2026~$200,000 (Jaguar I-PACE)May 28, 2026▲ TOWARDThe Next Web: Waymo’s Ojai robotaxi is cheaper to build (May 28, 2026) ↗Morgan Stanley estimate
China robotaxi cost target (Pony.ai Gen-7, 2027)Below RMB 230,000 (~$32k)EST.as of Apr 24, 2026—▲ TOWARDPony.ai release via AAP: Gen-7 cost targets (Apr 24, 2026) ↗Company target
Apollo Go fully driverless rides4.1M+ in H1 2026 (Q2 ~0.9M, derived)as of Jun 30, 20262.2M in Q2 2025Jun 30, 2025▼ AWAYBaidu 2026 Interim Report (SEC 6-K exhibit) ↗Derived: H1 >4.1M − Q1 3.2M; Baidu did not state Q2. Q2 hit by China’s permit freeze after the Wuhan outage
Apollo Go Q3 2026 operations30 cities (Oct); ~34M autonomous km in Q3; Shenzhen driverless rides ~+170% q/qas of Sep 30, 2026—▲ TOWARDBaidu Q3 2026 results (HKEX announcement, Nov 16, 2026) ↗Baidu did not give a Q3 total ride count, so the Q2 drop can’t yet be called reversed. More than 390M cumulative autonomous km at Oct 31
China rare-earth control pauseSuspension formally extended from Nov 11, 2026 to Jan 10, 2027 (MOFCOM/GACC Announcement No. 58)as of Nov 9, 2026Politically extended to Jan 10, 2027; no formal MOFCOM notice yetNov 1, 2026● FLATMOFCOM/GACC Announcement No. 58 of 2026 (Nov 9, 2026) ↗Formal notice, but still a temporary pause; prior from China Prospect

TREND CALLVolumes and unit costs are moving the right way on wheels and legs. Two cautions: humanoid shipment estimates differ by about 30% between research firms, and Apollo Go’s Q2 drop shows how fast permission, not technology, can cap the market. The rare-earth pause is now formal but runs only to Jan 10, 2027, so it stays the near-term chokepoint, and SAG has already cut its 2026 unit forecast citing US restrictions. Baidu’s Q3 release shows Apollo Go growing in cities and kilometres but gives no total ride count; Pony.ai (Nov 18) and WeRide (Nov 20) are the next readings.

NOT YET SOURCEDIDC and Counterpoint Q3 2026 humanoid shipment estimates · Apollo Go Q3 2026 total ride count (not in Baidu’s Nov 16 release) · Pony.ai (Nov 18) and WeRide (Nov 20) Q3 2026 ride counts

UPDATED NOV 16, 2026Part Two: Robot economics →Part One: The gasoline moment →Part One: Gasoline, 1911 →
BET 05 · PART TWO VERDICT: HALFMOVING AWAY

Robots are underweighted outside a small group

THE BET

Everyone except a small group of technologists is underweighting robots and robotaxis.

WHERE IT STANDS TODAY · LEADER: CAPITAL IS NOT UNDERWEIGHTING

Public attention may lag, but capital has arrived. Waymo raised $16 billion at a $126 billion valuation in February 2026; Unitree opened 629% above its IPO price on Aug 19, 2026; Figure’s last round (Sep 2025) valued it at $39 billion. Tesla’s CEO pay plan is built around 1 million robotaxis and 1 million Optimus robots. IDC sizes the whole H1 2026 humanoid market at just over $740 million of revenue.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Waymo valuation$126Bas of Feb 3, 2026—▼ AWAYelectrive: Waymo raises $16 billion (Feb 3, 2026) ↗
Unitree Shanghai debut+629% at open (¥1,100 vs ¥150.8 IPO)as of Aug 19, 2026—▼ AWAYCaixin: Unitree soars 629% in Shanghai debut (Aug 19, 2026) ↗
Figure AI last priced round$39B post-moneyas of Sep 2025—▼ AWAYValue Add VC: Figure AI valuation 2026 ↗
Humanoid market revenue, H1 2026>$740M (+322.7%)EST.as of Jun 30, 2026—▼ AWAYIDC via Webull: H1 2026 humanoid shipments near 25,000 (Sep 29, 2026) ↗IDC estimate; valuations are far ahead of revenue

TREND CALLAway. Investors no longer underweight robots; valuations run well ahead of revenue (a $126 billion Waymo and a Unitree priced like a megacap, against an industry selling under a billion dollars of humanoids in six months). The direction call in the card above can still be right while this one is wrong. It would move back if robot valuations fell hard while revenue kept growing.

NOT YET SOURCEDA consistent measure of public attention (e.g. survey or search-interest series)

UPDATED NOV 2, 2026Part Two: China →Part Two: The West’s humanoid makers →
BET 06 · PART TWO VERDICT: DOESN’T HOLD TODAYFLAT / MIXED

Tesla overtakes Waymo in robotaxis

THE BET

Tesla (with SpaceX) is best placed in robotaxis. Waymo has the technology lead today, but Tesla’s structural cost advantage, cheap camera-based cars built in its own factories at scale, will let it out-scale Waymo.

WHERE IT STANDS TODAY · LEADER: WAYMO

Waymo leads on everything Tesla doesn’t disclose: 500,000+ paid rides a week (March 2026), more than 4,000 driverless cars, 15 driverless metros and more than 300 million fully autonomous miles. Tesla’s safety report counted 3.1 million driverless miles through Oct 18 (nearly 2 million at the end of Q3). San Antonio became its eighth unsupervised city on Nov 9, and on Nov 12 Tesla said rides now run 24/7 in all eight. On Nov 13 it opened invite-only pre-orders for a personal-use Cybercab at $29,990 plus $299 a month, with delivery estimated for December 2026. On the Texas automated-vehicle roster Tesla had 1,306 robotaxis registered on Nov 14 against Waymo’s 1,321; Tesla briefly led on Nov 10, but Waymo added more over the week. Registrations are not cars carrying riders. Nationally, a community tracker counts 666 Tesla driverless cars against Waymo’s 4,000+, and an independent app check found Tesla’s bookable area is 59–78% of what it advertises. Tesla still discloses neither paid rides nor fleet size, and its CFO said on Oct 21 it is not yet at its 30-cent-a-mile goal. Waymo is cutting its own vehicle cost by an estimated third with the Ojai, and faces a new NHTSA query into 2,895 vehicles over school-bus violations.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Cost per vehicle: Waymo~$125,000 (Ojai, all-in)EST.as of May 28, 2026~$195,000–200,000 (Jaguar I-PACE)Apr 20, 2026▼ AWAYThe Next Web: Waymo’s Ojai robotaxi is cheaper to build (May 28, 2026) ↗Morgan Stanley estimate; prior is Sherwood’s characterization. Waymo’s cost falling narrows Tesla’s edge
Cost per vehicle: Tesla Cybercab$29,990 list price + $299/month autonomy fee (invite-only personal-use pre-orders; delivery est. Dec 2026)as of Nov 13, 2026$30,000 or less (sale price promised before 2027)Feb 17, 2026▲ TOWARDTechCrunch: Tesla opens Cybercab pre-orders ‘for personal use’ with a $299 monthly fee (Nov 14, 2026) ↗A sale price, not Tesla’s build cost; no deliveries yet. Runs only inside the Robotaxi geofences (Ashok Elluswamy). Musk calls the fee “too high” and says it should be “more like ~$115/month”. Compare Waymo’s Ojai at ~$125,000 all-in, a Morgan Stanley cost estimate, not a price. Prior: Musk on X (CleanTechnica)
Operating cost per mileWaymo ~$1.36 vs Tesla ~$0.74EST.as of Apr 20, 2026—▲ TOWARDSherwood: Robotaxis are the new millennial lifestyle subsidy (Apr 20, 2026) ↗Morgan Stanley estimate; excludes Tesla safety monitors and AV development. Another summary gives Tesla ~$0.81
Tesla progress to $0.30/mile goal“Not at our goal of 30 cents”; “path” to itas of Oct 21, 2026—● FLATShacknews: Tesla Q3 2026 earnings call notes (Oct 21, 2026) ↗
Paid rides per weekWaymo 500,000+; Tesla not disclosedas of Mar 26, 2026—▼ AWAYInsideEVs: Waymo doubles to 500,000 paid rides a week (Mar 26, 2026) ↗Tesla reported ~2.4M cumulative paid miles through June 2026
Driverless citiesWaymo 15 metros; Tesla 8 unsupervised cities (San Antonio opened Nov 9; + Bay Area with safety driver)as of Nov 9, 2026Waymo 15 metros; Tesla 7 citiesNov 2, 2026● FLATElectrek: Tesla launches Robotaxi in San Antonio with a 274 sq mi geofence on paper (Nov 9, 2026) ↗Tesla’s @robotaxi account on X announced San Antonio from 5 a.m. CT Nov 9 (Electrek; Not a Tesla App). Advertised San Antonio area ~274 sq mi vs Waymo’s 59 sq mi there (Robotaxi Tracker); an independent app check (Koby Pierce, via Electrek) found Tesla’s bookable area is 59–78% of advertised across six cities (Austin ~59%). Waymo count from Alphabet’s Q3 call; Bloomberg (Nov 4) says 14 cities
Service hours: Tesla24/7 in all 8 unsupervised cities (Bay Area still supervised)as of Nov 12, 20266 a.m.–2 a.m. in AustinOct 29, 2026▲ TOWARDNot a Tesla App: Tesla Robotaxi now runs 24/7 in every city (Nov 12, 2026) ↗Announced by Tesla’s @robotaxi account on X. More hours per car is the main lever on cost per mile. Miami’s service area nearly doubled to 59 sq mi (Robotaxi Tracker, via TeslaNorth, Nov 16). Prior: robotaxistats changelog
Driverless fleet, national: Waymo vs TeslaWaymo 4,000+ vs Tesla 666EST.as of Nov 9, 2026Waymo ~4,000 (Waymo-reported); Tesla not disclosedSep 2026▼ AWAYElectrek: Tesla launches Robotaxi in San Antonio with a 274 sq mi geofence on paper (Nov 9, 2026) ↗Robotaxi Tracker community count, quoted by Electrek; roughly 6:1. Bloomberg (Nov 4) restates Waymo’s “roughly 4,000 vehicles”
Texas AV registry: Tesla vs WaymoTesla 1,306 (671 Model Y, 573 Cybercab, 62 unlabeled) vs Waymo 1,321EST.as of Nov 14, 2026Tesla 1,282 vs Waymo 1,285Nov 8, 2026● FLATrobotaxistats: Texas robotaxi services and TxDMV registry (Tesla and Waymo re-read Nov 14, 2026) ↗Community read of the TxDMV automated-vehicle registry (robotaxistats). Since Nov 8: Tesla +24, Waymo +36. Tesla briefly led on Nov 10 (1,302 vs 1,297). Registrations, not cars carrying paying riders; Tesla has not published a fleet number
Unsupervised / driverless milesTesla 3.1M driverless (through Oct 18); Waymo 300M+EST.as of Oct 18, 2026Tesla ~2M cumulative unsupervised (Q3 deck)Sep 30, 2026▲ TOWARDTeslaNorth: Tesla’s new Robotaxi Safety Report (Oct 21, 2026) ↗Tesla Robotaxi Safety Report (first published Oct 21), as reported by TeslaNorth: counts only miles with no safety monitor in the car, incl. remote-assisted trips. Tesla’s live counter, which Tesla says estimates miles at the fleet’s current pace, read 5.23M on Nov 14 (robotaxistats safety page; 4.81M on Nov 9). Same concept as the Q3 deck’s figure is likely but not confirmed
Safety-driver status: TeslaAustin rides appear fully driverless; Dallas 608,130 no-monitor miles; Bay Area still with driveras of Oct 18, 2026—▲ TOWARDTesla North: Q3 call takeaways — Cybercab to Dallas (Oct 21, 2026) ↗Austin from a community tracker (Aug 2026); Dallas from Tesla’s safety report
Regulatory scrutiny: Tesla and WaymoWaymo: NHTSA recall query into 2,895 vehicles over school-bus violations. Tesla: school-zone speeding review (~1,000 vehicles) + Cybercab self-certification probeas of Nov 9, 2026Tesla only: school-zone speeding review + Cybercab self-certification probeOct 13, 2026▲ TOWARDReuters via Investing.com: US opens probe into Waymo vehicles over school bus safety violations (Nov 9, 2026) ↗Waymo query follows its Dec 2025 software recall. Tesla items: Reuters via The Star (Oct 13). Both leaders now face school-safety reviews

TREND CALLFlat on trajectory, because the decisive number is still missing. Two things moved toward the call this week. The cost edge got its first real price point: a $29,990 Cybercab, against an estimated ~$125,000 all-in for Waymo’s Ojai. And Tesla’s main utilization lever moved: rides now run 24/7 in all eight unsupervised cities. Against that: the Cybercab figure is a sale price, not a build cost, and no customer has one yet; Waymo out-registered Tesla in Texas this week (+36 vs +24); a community tracker still counts 666 Tesla driverless cars nationally against Waymo’s 4,000+; and Tesla still discloses no paid rides, fleet or cost per mile. Moves to “toward” when Tesla discloses paid rides or fleet size showing it closing that gap; to “away” if new cost estimates show Waymo’s per-mile cost catching Tesla’s.

NOT YET SOURCEDTesla paid rides per week (never disclosed) · Tesla official fleet size · Cybercab deliveries and production rate (deliveries estimated Dec 2026) · Waymo paid rides per week after March 2026 · Waymo official fleet total for Oct 2026 · Waymo Hyundai Ioniq 5 per-vehicle cost · A Sep–Nov 2026 analyst cost-per-mile comparison · Cost per vehicle of Waymo’s Toyota bZ4X robotaxi (Oct 30, 2026 announcement gave no cost or volumes) · Tesla Robotaxi active fleet count from Tesla (only state registrations available)

UPDATED NOV 16, 2026Part Two: Waymo’s lead →Part Two: Tesla’s bet →Part Two: Robot economics →
ADS & DISTRIBUTIONAds and distribution

Who sells the ads and the shopping when the front door to the internet becomes a conversation.

BET 07 · PART TWO VERDICT: PARTLYFLAT / MIXED

Google is Standard Oil after its prime

THE BET

Google is still the dominant integrator, but its core search business is past its peak and new suppliers of answers will erode it, the way Standard’s share fell once new fields and new refiners appeared.

WHERE IT STANDS TODAY · LEADER: GOOGLE

Search & other revenue grew 12% in Q3 2026 to $63.35 billion, down from 19% in Q1 and 17% in Q2; paid clicks rose 8% and cost per click 4%. But Alphabet grew 25% overall, Cloud grew 103%, the Gemini app passed one billion users (Oct 28), and management says query growth is the fastest in five years. StatCounter has Google at 91.05% of worldwide search in October 2026.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Google Search & other revenue growth (y/y)+12% ($63.35B, Q3 2026)as of Sep 30, 2026+17% (Q2 2026); +19% (Q1)Jun 30, 2026▲ TOWARDAlphabet Q3 2026 results, EX-99.1 (SEC) ↗CFO named FX as the single largest driver of the slowdown
Search paid clicks (y/y)+8% (Q3 2026)as of Sep 30, 2026+11% (first nine months)Sep 30, 2026▲ TOWARDPPC Land: Google ad clicks gain 8% in Q3 (10-Q, Oct 2026) ↗From the 10-Q; cost per click +4%
Google share of worldwide search (StatCounter)91.05%EST.as of Oct 202689.94%Sep 2026● FLATStatCounter: worldwide search engine market share ↗Page-view sample; ~1pt monthly swings are normal
Google share of US search (StatCounter)85.45% (Bing 9.78%)EST.as of Oct 2026—● FLATStatCounter: US search engine market share ↗
Search query growth“Highest it’s been in the last five years”as of Oct 28, 2026—▼ AWAYGoogle: Q3 2026 earnings call, remarks from our CEO (Oct 28, 2026) ↗Company statement; no number given
Google Cloud revenue growth+103% ($30.7B)as of Sep 30, 2026+82% (Q2 2026)Jun 30, 2026▼ AWAYAlphabet Q3 2026 results, EX-99.1 (SEC) ↗
Gemini app users1B+as of Oct 28, 2026950M (Q2 2026)Jul 2026▼ AWAYGoogle: Q3 2026 earnings call, remarks from our CEO (Oct 28, 2026) ↗

TREND CALLMixed, as Part Two found. Search revenue and clicks are decelerating, which is what “after its prime” predicts. But Google is building new engines faster than the old one slows: Cloud doubled, Gemini keeps adding users, and query volume is growing. Standard in 1901 still had a decade of dominance left. Moves to “toward” if Q4 search growth slows further (management guided to a few more points of pressure) while share slips.

NOT YET SOURCEDAI Mode and AI Overviews user counts after May 2026 · AI Mode ad pricing vs classic search

UPDATED NOV 2, 2026Part Two: Standard after its prime? →Part One: The Trust and the breakup →
BET 08 · PART TWO VERDICT: DOESN’T HOLD (YET)FLAT / MIXED

OpenAI wins AI advertising

THE BET

OpenAI is structurally better placed than Google to sell search-style ads in the future: it owns the conversation where intent is expressed, has a billion-plus weekly users, and has no legacy search business to cannibalize.

WHERE IT STANDS TODAY · LEADER: GOOGLE

OpenAI says ChatGPT ads passed a $1 billion annualized run rate less than 200 days after launch, and on Oct 21, 2026 said more than 100,000 advertisers have run ads since launch. The run rate has sat at about $1 billion since Sep 1, and The Information reported on Oct 18 that a $2.4 billion 2026 target is likely to be missed. Google Search alone brought in $63.35 billion in Q3 2026 (about $253 billion annualized), and Google is scaling ads inside AI Overviews and AI Mode, and on Nov 9 said AI Mode ads are expanding to more than 10 new countries, with multiturn ads in the US. OpenAI’s consumer lead said on Nov 9 that ChatGPT’s search-like queries are now roughly a fifth of Google’s volume, counting every question that searches the web. Advertiser returns are weak so far: a working paper reported by Digiday (Nov 5) finds paid ChatGPT visits earn US advertisers $0.89 a session, against $2.45 for Google paid search, and Adthena data reported by Business Insider (Nov 11) found 52% of 30 enterprise brands struggled to recoup their ChatGPT ad spend in July–September.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
ChatGPT ads annualized run rateMore than $1B (company statement)EST.as of Oct 21, 2026$100MMar 2026● FLATOpenAI: How we built ChatGPT’s novel ad stack for user trust and relevance (Oct 21, 2026) ↗Company run-rate claim. Flat: ~$1B since Sep 1; The Information’s Oct 18 “roughly $1B” repeated that reading. The Information’s “roughly 60,000 advertisers” counts something different from OpenAI’s cumulative “more than 100,000 advertisers” since launch. Prior came ~6 weeks after the Feb 9 launch
OpenAI 2026 ad revenue target$2.4B, reportedly likely to be missedEST.as of Oct 18, 2026$2.5BApr 9, 2026▼ AWAYDealroom summary of The Information: OpenAI ad run rate ~$1B, chasing $2.4B target (Oct 18, 2026) ↗Internal target, reported; prior from Axios via Reuters
ChatGPT click price vs Google~$5 vs $30–$50EST.as of Oct 18, 2026—▼ AWAYDealroom summary of The Information: OpenAI ad run rate ~$1B, chasing $2.4B target (Oct 18, 2026) ↗Reported
ChatGPT weekly users~1.2Bas of Sep 29, 2026900MFeb 27, 2026▲ TOWARDTechnologyChecker: ChatGPT statistics Q3 2026 (citing Altman at DevDay) ↗Altman at DevDay, reported second-hand; prior from TechCrunch
ChatGPT share of AI chatbot referrals (StatCounter)51.58% (Gemini 28.86%)EST.as of Oct 2026—▲ TOWARDStatCounter: AI chatbot market share worldwide ↗
OpenAI ad business as a share of Google Search revenue~0.4%EST.as of Sep 30, 2026—▼ AWAYAlphabet Q3 2026 results, EX-99.1 (SEC) ↗Derived: $1B ÷ (Q3 Search $63.35B × 4)
ChatGPT search-like queries vs Google~one-fifth of Google’s volume (OpenAI claim)EST.as of Nov 9, 2026—▲ TOWARDCMSWire (Alex Kantrowitz): ChatGPT search volume gains on Google, OpenAI’s consumer lead says (Nov 9, 2026) ↗Company claim by Adam Fry, ChatGPT’s consumer lead: OpenAI counts every question that searches the web as a search; no Google denominator given
Google ads inside AI answersAI Mode ads expanding to 10+ more English-language countries; multiturn ads live in the USas of Nov 9, 2026Scaling ads in AI Overviews and AI Mode globally; multi-turn ads in the USOct 28, 2026▼ AWAYSearch Engine Journal: Google Ads now runs in AI Mode across 10 more countries (Nov 10, 2026) ↗Announced on Google’s official Google Ads X account Nov 9. Brazil, Mexico and the UK planned for Q4 2026. AI Mode results are not reported separately to advertisers. Prior: TechCrunch (Oct 28)
Amazon Ads × ChatGPT pilotPilot with a group of US advertisersas of Sep 10, 2026—▲ TOWARDBloomberg Government: Amazon, OpenAI in ChatGPT advertising partnership (Sep 10, 2026) ↗
Advertiser outcomes, ChatGPT vs Google paid search (working paper)$0.89 vs $2.45 revenue per session; 0.37% vs 1.17% conversionEST.as of Nov 5, 2026—▼ AWAYDigiday: Six thousand brands later, ChatGPT ads still have a conversion problem (Nov 5, 2026) ↗Kaiser & Schulze working paper (under review), US advertisers, as reported by Digiday. Similarweb/Graphite: fewer than half of ad-eligible ChatGPT users see ads (US 43.4%, UK 45.7%), so there is headroom
Advertisers recouping ChatGPT ad spend (Adthena, 30 enterprise brands)52% struggled to recoup; 26% matched or beat Google adsEST.as of Sep 30, 2026—▼ AWAYBusiness Insider: ChatGPT Ads got brands to bite. Now comes the hard part. (Nov 11, 2026) ↗Adthena data for July–September, reported by Business Insider; small sample. BI also reports more than 100,000 advertisers have tested ChatGPT Ads

TREND CALLPushback, judged on trajectory. The structural argument got a little stronger: ChatGPT has the users and the most-used chatbot, and OpenAI now claims its search-like queries are about a fifth of Google’s (its own count, with a broad definition of a search). But the ad trajectory has stalled: the run rate has sat near $1 billion since Sep 1 (OpenAI’s Oct 21 “more than $1 billion” is the same level, and its consumer lead gave no newer number on Nov 9), the 2026 target is reportedly out of reach, clicks sell for a fraction of Google’s, and Google is now selling multiturn ads inside AI Mode in more markets. Outside evidence on ad returns is weak: a working paper finds paid ChatGPT visits convert at 0.37%, against 1.17% for Google paid search, and Adthena found 52% of 30 enterprise brands struggled to recoup their spend. Google added about $6.8 billion of Search revenue year over year in Q3 alone, several times OpenAI’s entire run rate, so the gap is widening in dollars. Moves to “toward” on a run rate or booked revenue well above $1 billion, or Google search growth falling further; to “away” if the run rate is still near $1 billion when Q4 numbers arrive.

NOT YET SOURCEDShare of ChatGPT messages that are commercial or shopping queries (the NBER paper’s “purchasable products” share is only in a chart image) · Similarweb visits, chatgpt.com vs google.com, Oct 2026 · Primary OpenAI source for 1.2B weekly users

UPDATED NOV 16, 2026Part Two: OpenAI and AI ads →Part Two: Google →
BET 09 · PART TWO VERDICT: LEANS YESMOVING TOWARD

Amazon wins agentic shopping

THE BET

Amazon can win agentic shopping if it gets its act together: own the shopping agent, monetize it with ads, and keep outside agents from becoming the front door to its store.

WHERE IT STANDS TODAY · LEADER: AMAZON

Amazon says more than 400 million customers used Alexa for Shopping (Rufus and Alexa+ combined) in the 12 months to Q3 2026, with interactions up 570%; Jassy calls it “the primary shopping agent customers use to shop Amazon.” Its robots.txt blocked 99 named AI agents on Oct 31, and it blocked Meta’s Muse agent in September. On Nov 13 it made Alexa for Shopping available in full in 13 countries and in beta in two more; the same day OpenAI launched ChatGPT Wallet, which lets its agents buy for US Pro users, though Amazon blocks them from buying on its site. The legal fence is weaker: the Ninth Circuit vacated Amazon’s injunction against Perplexity’s Comet on Aug 4, 2026.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Alexa for Shopping users (past 12 months)400M+ customersas of Sep 30, 2026Rufus: 300M+ customers in 2025Dec 31, 2025▲ TOWARDAmazon Q3 2026 results, EX-99.1 (SEC) ↗Company figure; prior is Rufus alone
Alexa for Shopping interactions (y/y)+570%as of Sep 30, 2026—▲ TOWARDAmazon Q3 2026 results, EX-99.1 (SEC) ↗
Spend per order, agent users vs others (US)+50%as of Oct 29, 2026“over 40%”Jun 30, 2026▲ TOWARDAmazon Q3 2026 earnings call transcript (Singju Post) ↗Company claim
Outside AI agents blocked in robots.txt99 named agentsas of Oct 31, 2026—▲ TOWARDamazon.com/robots.txt (checked Oct 31, 2026) ↗
Legal fence against outside agentsInjunction vacated; case back in district courtas of Aug 4, 2026—▼ AWAYEngadget: Ninth Circuit vacates Amazon’s injunction against Perplexity (Aug 4, 2026) ↗
Rival agents shoppingOver a quarter of active Meta Muse users shop with itas of Oct 21, 2026—▼ AWAYMarketBeat: Meta Q3 2026 earnings call highlights (Oct 21, 2026) ↗Meta claim
Alexa for Shopping reachFull in 13 countries + beta in 2; shopping by text open to all usersas of Nov 13, 2026—▲ TOWARDPYMNTS: Amazon expands Alexa for Shopping to 15 countries and lets anyone shop by text (Nov 13, 2026) ↗
Rival agent checkoutOpenAI’s ChatGPT Wallet lets its dots agents buy for US Pro users; Amazon blocks such agents from buying on its siteas of Nov 13, 2026—▼ AWAYCNBC: OpenAI launches ChatGPT Wallet, allowing its ‘dots’ agents to make purchases (Nov 13, 2026) ↗OpenAI takes no cut of transactions to start; built with Amex, Mastercard and Visa
Jassy on outside agents“Few customers are doing much shopping this way yet”as of Oct 29, 2026—▲ TOWARDAmazon Q3 2026 earnings call transcript (Singju Post) ↗
Amazon advertising revenue$19.6B, +11% (+12% ex-FX)as of Sep 30, 2026$19.8B, +26% (Q2)Jun 30, 2026● FLATAmazon Q3 2026 results, EX-99.1 (SEC) ↗Q3 depressed by the Prime Day shift into Q2

TREND CALLAmazon is executing the plan: its own agent runs at enormous scale, it sells ads inside it, and outside agents are still small. Against it: the courts have so far declined to let Amazon fence agents out by law, and rival agents (Meta’s Muse above all) are already used to shop. Amazon also told investors it expects to work with outside agents, a different strategy from the fence. Moves to “flat” if outside agents start taking a measurable share of Amazon orders.

NOT YET SOURCEDBuy for Me sales or usage in 2026 · Alexa+ user count · Agent-driven share of Amazon orders

UPDATED NOV 16, 2026Part Two: Amazon’s shopping test →Part One: Pipelines →
META & INTERFACESMeta and the interface

The integrated major nobody cast in Part One, and what people will wear, or implant, to talk to AI.

BET 10 · PART TWO VERDICT: HOLDSMOVING TOWARD

Meta was underweighted

THE BET

Meta is the integrated major Part One barely cast: it owns its infrastructure and models, runs the clearest AI-for-ads flywheel, and Zuckerberg has navigated every platform shift in two decades.

WHERE IT STANDS TODAY · LEADER: META

Meta’s Q3 2026 ad revenue was $63.06 billion, up 26%, with price per ad up 14% on 10% more impressions. Its Muse agent passed 10 million accounts in about six weeks (Oct 21). The bill is showing: 2026 capex guidance is $135–145 billion, Q3 free cash flow was about –$2.1 billion, and on Oct 26 Meta announced an equity raise of up to $95 billion, pricing $37.5 billion of it the same day ($25 billion of common stock via forward sales, $12.5 billion of mandatory convertible preferred). Debt followed in November: $15 billion of dollar notes on Nov 5 and €9 billion of euro notes, its first euro deal, on Nov 12; S&P revised its outlook to negative on Oct 30.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Advertising revenue growth (y/y)+26% ($63.06B)as of Sep 30, 2026—▲ TOWARDMeta Q3 2026 results, EX-99.1 (SEC) ↗
Average price per ad (y/y)+14%as of Sep 30, 2026+12% (Q2 2026)Jun 30, 2026▲ TOWARDPPC Land: Meta ad revenue +26%, $9.89B settlement charge (Oct 2026) ↗Rising ad prices are the AI-targeting signal
Ad impressions (y/y)+10%as of Sep 30, 2026+14% (Q2 2026)Jun 30, 2026● FLATPPC Land: Meta ad revenue +26%, $9.89B settlement charge (Oct 2026) ↗
Muse agent accounts (US and Canada)10M+ in ~6 weeks; ~25% use it dailyas of Oct 21, 2026—▲ TOWARDMarketBeat: Meta Q3 2026 earnings call highlights (Oct 21, 2026) ↗Company claim
2026 capex guidance$135–145BEST.as of Oct 21, 2026$115–135BJan 2026● FLATMeta Q3 2026 results, EX-99.1 (SEC) ↗Company guidance
Quarterly free cash flow–$2.12Bas of Sep 30, 2026—▼ AWAYMeta Q3 2026 results, EX-99.1 (SEC) ↗Reported as Meta’s first negative quarter in more than a decade
Equity raise$37.5B priced: $25.0B common (upsized from $22.5B, forward sale at $625/share) + $12.5B 6.50% mandatory convertible preferredas of Oct 26, 2026Up to $95B announced (~$35B underwritten + up to $60B ATM)Oct 26, 2026▼ AWAYMeta: pricing of depositary shares and Class A common stock offerings (Oct 26, 2026) ↗Company release. Common-stock cash arrives as forward sales settle, on or before Nov 19, 2027; underwriter options could add up to $3.75B common and 37.5M more depositary shares. The at-the-market portion is not part of this pricing
Debt raised in November$15B dollar notes (Nov 5) + €9B euro notes (Nov 12; debut euro deal)as of Nov 12, 2026—▼ AWAYPPC Land: Meta borrows €9bn in euros, at up to 6.4% until 2065 (Nov 14, 2026) ↗From Meta 8-Ks as reported by PPC Land. Euro notes in six tranches maturing 2031–2065, coupons 4.300%–6.400%; Reuters: a record €53B of orders. S&P revised its outlook to negative, affirming AA- (Oct 30). ~$62.6B raised in under a month including equity (PPC Land calculation)

TREND CALLThe AI-for-ads flywheel is working: ad prices are rising faster than impressions, and Muse is growing fast. The new risk is the bill: after free cash flow went negative, equity and debt are now arriving together (about $62.6 billion in under a month, by PPC Land’s count), the equity in non-voting shares so Zuckerberg keeps control, and S&P has turned its outlook negative. “Underweighted” is no longer true of the market; “integrated and executing” still is. Moves to “flat” if ad-price growth stalls while free cash flow stays negative.

NOT YET SOURCEDMeta AI monthly users (no 2026 figure disclosed) · 2027 capex guidance (not yet given) · Size and pace of the at-the-market portion of Meta’s raise

UPDATED NOV 16, 2026Part Two: The survivor →Part One: Casting the characters →
BET 11 · PART TWO VERDICT: NOT THIS DECADEFLAT / MIXED
THE BET

The long-run interface to AI may be closer to a brain-computer interface like Neuralink than to Meta’s glasses.

WHERE IT STANDS TODAY · LEADER: GLASSES (META / ESSILORLUXOTTICA)

EssilorLuxottica sold more than 7 million AI glasses in 2025. Neuralink had implanted roughly 27 people by July 2026, by an unofficial count, and its device is still investigational. Glasses growth is cooling: EssilorLuxottica’s wearables grew 20% in Q3 2026 after nearly 100% in Q2.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Neuralink implants (cumulative)~27EST.as of Jul 8, 202621Jan 2026● FLATDiario Bitcoin: Neuralink at ~27 implanted patients (Jul 11, 2026) ↗Unofficial count from a patient video; prior from Wikipedia
AI glasses sold by EssilorLuxottica7M+ in 2025as of Dec 31, 2025~2M across 2023–24Dec 31, 2024▼ AWAYMezha (citing CNBC): EssilorLuxottica sold over 7M AI glasses in 2025 (Feb 2026) ↗
EssilorLuxottica wearables growth+20% (€486M, Q3 2026)as of Sep 30, 2026nearly +100% (Q2 2026)Jun 30, 2026▲ TOWARDChannelNews: EssilorLuxottica Q3 2026 wearables (Oct 20, 2026) ↗
XR headset + glasses shipments (IDC)+35.3% y/y (Q2 2026)EST.as of Jun 30, 2026+130.2% (Q1 2026)Mar 31, 2026▲ TOWARDIDC: Smart glasses growth cools to 35% (Sep 16, 2026) ↗

TREND CALLGlasses are cooling from a hot start, which nudges toward the claim, but the gap is still on the order of 250,000 to one in units, and no implant is approved for healthy users. Nothing in the data suggests the order changes this decade. Meta’s EMG wristband, shipping with its Display glasses, is the likelier middle path. Moves to “toward” only if an implant is cleared beyond trials.

NOT YET SOURCEDOfficial Neuralink implant count for 2026 · AI glasses unit sales for 2026

UPDATED NOV 2, 2026Part Two: Glasses or Neuralink →
BET 12 · PART TWO VERDICT: HOLDSFLAT / MIXED

Meta’s hardware record is weaker than Musk’s

THE BET

Meta’s record in hardware (Portal, Quest, the metaverse) is weaker than Tesla’s and SpaceX’s manufacturing record, so bets that depend on building physical things favor Musk.

WHERE IT STANDS TODAY · LEADER: MUSK’S COMPANIES

Reality Labs lost $4.13 billion in Q3 2026 and roughly $96 billion since 2020 (our arithmetic). Tesla delivered 486,532 vehicles in Q3 2026, down 2%, and SpaceX flew 35 launches in the quarter. Meta’s one hardware hit, the Ray-Ban glasses, is built by a partner.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Reality Labs operating loss (quarter)$4.13B (Q3 2026)as of Sep 30, 2026$4.43B (Q3 2025)Sep 30, 2025● FLATMeta Q3 2026 results, EX-99.1 (SEC) ↗
Reality Labs cumulative losses since 2020~$96BEST.as of Sep 30, 2026—▲ TOWARDOutlook Respawn: Reality Labs losses $83.6B through 2025 (Jan 31, 2026) ↗Derived: $83.6B through 2025 + $12.78B in 9M 2026
Reality Labs loss guidance~$15B in 2027EST.as of Oct 21, 2026$18–19B in 2026Oct 21, 2026▼ AWAYPPC Land: Meta ad revenue +26%, $9.89B settlement charge (Oct 2026) ↗Company guidance; narrowing as spend shifts to AI
Tesla quarterly deliveries486,532 (–2% y/y)as of Sep 30, 2026—▼ AWAYDrive Tesla Canada: Tesla Q3 2026 results (Oct 21, 2026) ↗
SpaceX launches (quarter)35, including 2 Starship flightsas of Sep 30, 2026—▲ TOWARDSpaceX Q3 2026 earnings call transcript (Motley Fool) ↗

TREND CALLStill holds as a description of the record, and nothing this month changes it: Meta is narrowing hardware losses by spending less, not by shipping a hit, while Tesla’s car volumes dipped and SpaceX’s launch machine keeps compounding. The next test is whether Optimus reaches outside customers.

UPDATED NOV 2, 2026Part Two: Glasses or Neuralink →Part Two: Tesla’s bet →
MARGINS & THE RAILSMargins and the rails

Where the margin sits in the chain, who gets the rebates, and who is overplaying their hand.

BET 13 · PART TWO VERDICT: PARTLYFLAT / MIXED

Volume-controlling labs are Standard; the clouds are railroads

THE BET

The map is upside down: hyperscalers and neoclouds are the railroads that crave guaranteed volume, and the labs that control the volume are Standard Oil. The labs will end up with the rebates and, in time, the margin.

WHERE IT STANDS TODAY · LEADER: NVIDIA AND THE CLOUDS

Margin still sits upstream. Nvidia’s gross margin was 75.0% in its July 2026 quarter, and the clouds’ operating margins widened sharply: AWS 38.7% and Google Cloud 36.2% in Q3 2026, against 34.6% and 23.7% a year earlier. The leading lab’s progress reversed in Q3: Anthropic’s S-1 filings show compute falling from 251% of revenue in Q1 2025 to 48% in Q2 2026, then rising to 58% in Q3 on $17.0 billion of revenue as it bought short-term compute, turning a $974 million Q2 operating profit into a $567 million loss. OpenAI’s gross margin was reported at 39% in Q1 2026, with a $9.3 billion operating loss; Bloomberg reports its adjusted operating margin improved from –150% in Q2 to –109% in Q3, on more than $10 billion of revenue. Compute sellers are raising prices: CoreWeave says Ampere and Hopper prices rose 30–50% in Q3.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Nvidia gross margin75.0% (GAAP and non-GAAP, Q2 FY27)as of Jul 26, 202672.4% (Q2 FY26)Jul 27, 2025▼ AWAYNvidia: Q2 fiscal 2027 results (Aug 26, 2026) ↗Q3 guide 74.0% ±50bp; results Nov 17
AWS operating margin38.7%EST.as of Sep 30, 202634.6%Sep 30, 2025▼ AWAYAmazon Q3 2026 results, EX-99.1 (SEC) ↗Derived: $17.77B ÷ $45.98B; prior $11.43B ÷ $33.01B
Google Cloud operating margin36.2%EST.as of Sep 30, 202623.7%Sep 30, 2025▼ AWAYAlphabet Q3 2026 results, EX-99.1 (SEC) ↗Derived: $11.12B ÷ $30.71B; prior $3.59B ÷ $15.16B
Microsoft Agents and Infra operating margin52.8%EST.as of Sep 30, 202653.1% (recast)Sep 30, 2025● FLATMicrosoft FY27 Q1 press release, segment results (Microsoft IR) ↗Derived; segment mixes Azure with software, not an Azure-only margin
Anthropic compute as % of revenue58% (Q3 2026)as of Sep 30, 202648% (Q2 2026)Jun 30, 2026▼ AWAYAnthropic, PBC Form S-1/A, Amendment No. 1 (SEC, Nov 2, 2026) ↗Includes $5.0B of compute under short-term commitments, which “currently costs significantly more” than multi-year compute. 251% in Q1 2025
Anthropic operating income–$567M (–3% of revenue, Q3 2026)as of Sep 30, 2026+$974M (8% of revenue, Q2 2026)Jun 30, 2026▼ AWAYAnthropic, PBC Form S-1/A, Amendment No. 1 (SEC, Nov 2, 2026) ↗Driven by the short-term compute and $2.0B of non-cash donation-matching charges. Company-adjusted operating margin 19% → 15% (non-GAAP)
Anthropic implied gross margin≈85.5% (9M 2026)EST.as of Sep 30, 202685.4% (H1 2026)Jun 30, 2026● FLATAnthropic, PBC Form S-1/A, Amendment No. 1 (SEC, Nov 2, 2026) ↗Derived: 1 − $4,812M cost of revenue ÷ $33,242M revenue. In H1 only $2.38B of $8.95B compute was in cost of revenue
OpenAI gross margin39% (Q1 2026); operating loss $9.3BEST.as of Mar 31, 202633% (2025)Dec 31, 2025▲ TOWARDThe Decoder (citing The Information): OpenAI Q1 2026 revenue, burn and margin (Jun 20, 2026) ↗Reported from shareholder documents (The Information)
OpenAI adjusted operating margin–109% (Q3 2026, revenue >$10B)EST.as of Sep 30, 2026–150% (Q2 2026)Jun 30, 2026▲ TOWARDTechCrunch (citing Bloomberg): OpenAI’s quarterly revenue reportedly crosses $10B for the first time (Nov 11, 2026) ↗Reported by Bloomberg from anonymous sources; company-adjusted, not comparable with the Q1 gross margin
Compute seller pricing (CoreWeave)>$40M per MW a year short-term vs $20–25M on new 4–5-year contracts; Ampere/Hopper prices +30–50% q/qas of Nov 10, 2026—▼ AWAYCoreWeave Q3 2026 earnings call transcript (Motley Fool, Nov 10, 2026) ↗Company statements on the Q3 call; CoreWeave raised prices across SKUs by 25% in July and another 10% later in the quarter

TREND CALLFlat, but the newest lab reading moved away. Anthropic’s profit lasted one quarter: buying short-term compute that “currently costs significantly more” pushed compute to 58% of revenue and the company back to a $567 million operating loss in Q3. That is evidence the compute sellers still hold pricing power, and CoreWeave confirmed it this week: Ampere and Hopper prices up 30–50% in Q3, and short-term capacity selling for more than $40 million per megawatt a year, against $20–25 million on new four- and five-year contracts. Nvidia and the clouds kept widening their margins. On the other side, OpenAI’s reported adjusted operating margin narrowed from –150% to –109%. It is one quarter each way, so the call holds; Nvidia’s Nov 17 results (Q3 guide 74.0%), its first under interim CEO Colette Kress, are the next test. Moves to “away” if Nvidia’s margin holds near 75% while lab compute shares keep rising; to “toward” if Nvidia’s margin eases while lab margins recover.

NOT YET SOURCEDAzure-only operating margin (not disclosed) · Oracle OCI realized margin (not disclosed) · OpenAI Q2/Q3 2026 gross margin (only an adjusted operating margin was reported)

UPDATED NOV 16, 2026Part Two: Volume and rebates →Part Two: Three maps →Part One: The railroads →
BET 14 · PART TWO VERDICT: LIKELY, NOT YETMOVING TOWARD

Someone consolidates the neoclouds

THE BET

Someone will roll up the neoclouds the way Standard bought out Cleveland’s refiners in 1872: numerous, leveraged, and worth more inside a larger system than alone.

WHERE IT STANDS TODAY · LEADER: NO CONSOLIDATOR YET (LENDERS AND NVIDIA HOLD THE LEVERAGE)

No neocloud merger has completed in 2026 by any source we could verify; CoreWeave’s bid for Core Scientific failed in October 2025. The stress that precedes a roll-up has arrived: after the midterm selloff, CoreWeave’s five-year CDS hit a record 1,267.9bp on Nov 5, up from 843bp at the end of September, and the yield on its 2032 bond topped 16%. Its Q3 results (Nov 10) eased that: revenue rose 168% to $3.66 billion, backlog reached about $132.5 billion, the shares jumped about 25% on Nov 11 and the CDS fell to about 1,152bp. But CoreWeave says it won’t issue parent-level unsecured debt through 2027 at current yields, and Nebius cut its 2026 revenue forecast on capacity delays (Nov 12). A neocloud has failed to raise in public markets: Firmus pulled its ~A$7 billion Australian IPO in October, and Nscale hasn’t updated its prospectus since September. The biggest customer is consolidating: Microsoft is reportedly shrinking its roster of neocloud partners. Nscale’s S-1 shows a –350% operating margin for H1 2026. Private capital is still there: Lambda is raising up to $4 billion (Oct 6) and Crusoe raised $3.9 billion at $30.9 billion (September).

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Completed neocloud mergers in 2026None verifiedas of Nov 1, 2026CoreWeave–Core Scientific terminatedOct 30, 2025● FLATCoreWeave: comment on Core Scientific stockholder vote (Oct 30, 2025) ↗
CoreWeave 5-year CDS~1,152bpas of Nov 10, 20261,267.9bp (record)Nov 5, 2026▼ AWAYGlitch in the Credit newsletter: CoreWeave jumps 26% on $132.5B backlog (Nov 11, 2026) ↗Newsletter-reported level after Q3 results; vendor not named (it gives a 1,278bp Nov 5 close vs S&P Global’s 1,267.9bp). The 9.75% 2031 bonds yielded 15.5%, down from 16.7%. Still far above 843bp at Sep 30 (S&P Global). Prior: Reuters
CoreWeave share price$61.20 (+25.6% on the day)as of Nov 11, 2026$54.87Oct 30, 2026▼ AWAYGlitch in the Credit newsletter: CoreWeave jumps 26% on $132.5B backlog (Nov 11, 2026) ↗CNBC gives +24.71%. Still above the $40 IPO price; a less cheap target. Prior: StockAnalysis
CoreWeave funding planNo parent-level unsecured debt through 2027 at current yields; investment-grade asset-level loan at ~7–7.5% planned; ~$9B prepayments due in 3 monthsas of Nov 10, 2026—▼ AWAYCoreWeave: Q3 2026 results press release (Nov 10, 2026) ↗Loan and debt plans from the Q3 call (Motley Fool transcript); prepayments from the release. Q3 revenue $3.66B (+168%); backlog ~$132.5B
Nebius Q3Cut 2026 revenue guide to $3.0–3.2B (top was $3.4B) on third-party capacity delays; shares –16%as of Nov 12, 2026—▲ TOWARDReuters: Nebius pares back 2026 revenue forecast on capacity delays, shares slump (Nov 12, 2026) ↗Plans to fund 75–85% of $40–45B of 2027 capex with customer prepayments and asset-backed financing. Dropped a planned high-yield bond in October (Reuters, via Glitch in the Credit)
Nscale operating margin–350% (H1 2026)as of Jun 30, 2026—▲ TOWARDNscale Ltd Form S-1 (SEC, Sep 18, 2026) ↗From the S-1
Private capital for independentsLambda raising up to $4B at $14.5B pre-money; Crusoe $3.9B at $30.9Bas of Oct 6, 2026—▼ AWAYTechCrunch: Lambda to raise $4B ahead of planned IPO (Oct 6, 2026) ↗Crusoe: Zawya, Sep 2026
Debt-market access for neoclouds“Debt markets have almost entirely shut off access to money for any company that doesn’t have substantial revenue independent of AI data centers”EST.as of Oct 25, 2026—▲ TOWARDCNBC: Elon Musk says more than 200 neoclouds face ‘guppy-level death’ (Oct 26, 2026) ↗Musk on X (Oct 25), quoted by CNBC; he predicts “guppy-level death and consolidation” for “200+ neoclouds”. Claim from a competitor (SpaceX sells compute). Nvidia’s Huang called neocloud demand “off the charts”
Public-market raisesFirmus pulled its ~A$7B ASX IPO; Nscale hasn’t updated its prospectus since Septemberas of Nov 6, 2026Nscale still targeting ~$30B IPO; not yet pricedOct 29, 2026▲ TOWARDABC News (Australia): How Firmus’s plans for the biggest IPO in three decades fell over (Oct 10, 2026) ↗Firmus (Nvidia-backed, Australia) pulled the listing Oct 9–10 at a planned A$43.7B market cap; founders hope to raise privately and try Nasdaq next year (ABC News). Nscale status: CNBC (Nov 6). Microsoft declined an option to more than double its Nscale Texas deal (Reuters, Nov 5). Prior: Bloomberg via Moneycontrol
Biggest customer consolidating suppliersMicrosoft shrinking its roster of neocloud partnersas of Nov 6, 2026—▲ TOWARDCNBC: Microsoft plans to use fewer neoclouds as AI buildout shows signs of strain (Nov 6, 2026) ↗Reported, people familiar; Microsoft “hasn't decided which neoclouds it will stop using”. Partners named: CoreWeave, Nebius, Chronoscale, IREN, Lambda, Nscale. IREN: “Capital markets have become more selective in the last few weeks”

TREND CALLToward, but less firmly. The move-back trigger (easy refinancing at normal spreads) was only half met: CoreWeave’s results calmed its credit and it can still borrow secured at investment-grade rates, but both CoreWeave and Nebius are avoiding unsecured markets, and Nebius cut its guidance. Firmus’s pulled IPO and Microsoft’s roster cuts still stand. There is still no merger, so this is pressure toward a roll-up, not the roll-up itself. Moves to “flat” if CoreWeave’s investment-grade loan prices at 7–7.5% and its CDS keeps falling toward September’s 843bp.

NOT YET SOURCEDNeocloud M&A reported only in search snippets (e.g. Brookfield Radiant–Ori) — not verified

UPDATED NOV 16, 2026Part Two: The neocloud roll-up →Part One: The chokepoint was the refinery →
BET 15 · PART TWO VERDICT: PARTLYMOVING TOWARD

Oracle is the Pennsylvania Railroad overplaying its hand

THE BET

Oracle is the Pennsylvania Railroad of 1877: an infrastructure owner overextending on borrowed money for a single dominant customer, and likely to be forced to retreat.

WHERE IT STANDS TODAY · LEADER: OVEREXTENSION FITS; THE EMPIRE MOVE IS OTHERS’

Oracle’s backlog reached $664 billion at the end of August 2026, roughly half tied to OpenAI, with free cash flow of –$5.4 billion in the quarter. S&P cut it to BBB- on Jul 9, 2026. Its five-year CDS closed at 298.75bp on Nov 12, a little below 305bp on Nov 5 but up from 242.5bp at Sep 30 and roughly double January’s 144bp; the stock closed at $129.86 on Nov 11, about 62% below its record. At its Oct 28 investor day Oracle capped external financing for AI projects at $20 billion through fiscal 2030; on Nov 2 it filed to raise all of it, $10 billion of stock and $10 billion of senior notes, and its shares fell 8.8% that day. Total debt was $125.34 billion at Aug 31, as reported by CNBC. Bloomberg reports Microsoft is winding down an arrangement to rent AI computing from Oracle (Nov 6).

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Remaining performance obligations$664B (~half OpenAI)as of Aug 31, 2026$638BMay 31, 2026▲ TOWARDERP Today: Oracle Q1 FY27 results, $664B backlog (Sep 11, 2026) ↗
Oracle 5-year CDS298.75bpas of Nov 12, 2026305bpNov 5, 2026● FLATInvesting.com: Oracle 5Y CDS daily history ↗242.5bp on Sep 30; 144bp on Jan 2, 2026; 2026 high close 334.25bp (Oct 20)
S&P ratingBBB-as of Jul 9, 2026—▲ TOWARDPPC Land: S&P cuts Oracle to BBB- (Jul 2026) ↗
Free cash flow (quarter)–$5.4B (Q1 FY27)as of Aug 31, 2026—▲ TOWARDERP Today: Oracle Q1 FY27 results, $664B backlog (Sep 11, 2026) ↗
Share price vs record$129.86, ~62% below $345.69EST.as of Nov 11, 2026$130.67Oct 30, 2026● FLATGlitch in the Credit newsletter: CoreWeave jumps 26% on $132.5B backlog (Nov 11, 2026) ↗Close, +10.1% on the day. Drawdown derived: 129.86 ÷ 345.69 − 1 ≈ −62% (the newsletter’s “48% below its high” uses a different high). Prior: StockAnalysis
FinancingFiled to sell $10B common stock + $10B senior notes, the full $20B cap; shares –8.8% that dayas of Nov 2, 2026External financing for AI projects capped at $20B through FY30Oct 28, 2026▲ TOWARDCNBC: Oracle files to sell $20 billion in new stock, bonds to finance AI buildout (Nov 2, 2026) ↗CNBC; Oracle expects positive free cash flow “in the second half of 2028”. Prior: investor-day target (Crypto Briefing). Two people told CNBC Microsoft “opted not to renew a deal with Oracle” (Nov 6); Bloomberg via Reuters (Nov 6) says Microsoft “is winding down an arrangement to rent AI computing from Oracle” on older Nvidia chips, while Oracle says demand for its AI infrastructure “remains strong”
Total debt$125.34Bas of Aug 31, 2026—▲ TOWARDCNBC: Oracle files to sell $20 billion in new stock, bonds to finance AI buildout (Nov 2, 2026) ↗As reported by CNBC (Nov 2)

TREND CALLThe overextension half keeps fitting: more backlog tied to one customer, a junk-adjacent rating, CDS well above September’s level and doubled this year (it eased slightly this week). The investor-day cap on outside financing looked like the first sign of the retreat the analogy predicts; filing to raise the whole cap within a week, as the stock fell, fits the overextension. The other half, the Empire move into the customer’s business, is being made by stronger players (Google, SpaceX), not Oracle. Moves to “flat” if Oracle’s CDS falls back toward January levels without a retreat.

UPDATED NOV 16, 2026Part Two: Oracle and Empire →Part One: When the railroad tried to refine →
BET 16 · PART TWO VERDICT: AMBITION, NOT POSITIONFLAT / MIXED

Elon is Rockefeller

THE BET

Musk is the Rockefeller of AI: integrating launch, compute, models, chips and distribution into one low-cost machine.

WHERE IT STANDS TODAY · LEADER: INTEGRATION GROWING; AS A LANDLORD

SpaceX’s Super Intelligence segment brought in $7.56 billion in Q3 2026, nearly triple Q2, and 52% of revenue. But $7.02 billion of it was “solutions & infrastructure,” mostly compute rented to Anthropic, Google and Reflection AI. Nameplate compute reached 1.7 GW at Sep 30 (1.4 GW in June), the Cursor acquisition closed in August, and Terafab’s first test chip is targeted for mid-2027.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
SpaceX Super Intelligence revenue$7.56B (Q3 2026)as of Sep 30, 2026$2.56B (Q2 2026)Jun 30, 2026▲ TOWARDSpaceX Q3 2026 results, EX-99.1 (SEC) ↗
Share of SI revenue from renting infrastructure~93% ($7.02B of $7.56B)EST.as of Sep 30, 2026—▼ AWAYSpaceX Q3 2026 results, EX-99.1 (SEC) ↗Derived; a landlord’s revenue, not a refiner’s
Nameplate compute1.7 GWas of Sep 30, 20261.4 GWJun 30, 2026▲ TOWARDSpaceX Q3 2026 results, EX-99.1 (SEC) ↗Target 2.5–3 GW by year-end
Terafab chip programTest chip mid-2027; volume around Q3 2028EST.as of Oct 22, 2026—● FLATTesla Q3 2026 earnings call transcript (Singju Post) ↗Company targets

TREND CALLMusk is integrating faster than anyone, but mostly as Vanderbilt: SpaceX earns its AI money renting compute to the labs it competes with, not as the low-cost refiner of models. A strong business, just not the Rockefeller position. Moves to “toward” if SpaceX’s own models and products (Grok, Cursor, X) become most of the segment.

NOT YET SOURCEDGrok usage or model market share from a consistent source

UPDATED NOV 2, 2026Part One: Casting the characters →Part Two: Three maps →
BET 17 · PART TWO VERDICT: HOLDS, AS FINANCIERMOVING TOWARD

Nvidia moves across the stack

THE BET

Nvidia, secure in chip design, can move across the stack: systems, networking, models, clouds and finance.

WHERE IT STANDS TODAY · LEADER: NVIDIA

Nvidia is moving across the stack mainly as a financier. Its 10-Q for the quarter ended Jul 26, 2026 shows $279 billion of supply and capacity commitments (up from $119 billion a quarter earlier), guarantees of up to $105 billion on OpenAI-affiliated data-center leases plus $3.5 billion for AI cloud partners, and $25 billion of equity commitments to model makers. On Aug 10 it signed non-binding agreements with six asset managers to mobilize over $500 billion for its customers, and on Oct 26 it committed $3 billion to Meta’s equity raise. On Nov 11 its board accepted the resignation of CEO Jensen Huang, effective immediately; CFO Colette Kress is interim CEO. Q3 results on Nov 17 are the first without him.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Supply and capacity commitments$279Bas of Jul 26, 2026$119BApr 26, 2026▲ TOWARDNvidia 10-Q, quarter ended Jul 26, 2026 (SEC) ↗
Lease guaranteesUp to $105B (SB Energy / OpenAI) + $3.5B (AI clouds)as of Jul 26, 2026—▲ TOWARDNvidia 10-Q, quarter ended Jul 26, 2026 (SEC) ↗
Equity commitments to model makers$25Bas of Jul 26, 2026—▲ TOWARDNvidia 10-Q, quarter ended Jul 26, 2026 (SEC) ↗
Third-party financing platforms>$500B targeted with Apollo, BlackRock, Blackstone, Brookfield, Goldman, KKREST.as of Aug 10, 2026—▲ TOWARDNvidia: AI compute financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman, KKR (Aug 10, 2026) ↗Non-binding target
Stakes in customers$3B in Meta’s equity raiseas of Oct 26, 2026—▲ TOWARDFree Press Journal: Nvidia, AMD to invest $4B in Meta (Oct 28, 2026) ↗
The financier’s architectCEO Jensen Huang resigned Nov 11; CFO Colette Kress interim CEO; Harvey C. Jones board chairas of Nov 11, 2026—▼ AWAYNvidia Form 8-K: CEO transition (SEC, filed Nov 12, 2026) ↗Reuters: “Huang’s departure threatens to disrupt an AI ecosystem that he helped shape by pouring Nvidia’s swelling cash pile back into cloud providers, suppliers and labs.” Traders said the exit could curb Nvidia’s borrowing (Bloomberg; ICE CDS 139bp, Nov 12). Shares fell 4.3% on Nov 12 (Reuters). Q3 results Nov 17

TREND CALLStill toward on the disclosed record, but the risk has changed. Nvidia is not becoming a rival cloud or lab; it is becoming the bank that lets its customers keep buying. That keeps demand up and gross margin near 75% (tracked on the margins card above), and moves risk onto Nvidia’s own balance sheet (see the debt card below). The person who built that financier strategy is now gone. The Nov 17 results and Q3 10-Q, and the new CEO’s stance on guarantees, decide the next reading; it moves to “flat” if commitments and guarantees stop growing or the board signals a pullback.

UPDATED NOV 16, 2026Part Two: Nvidia across the stack →Part One: Vanderbilt →
RISK & FINANCINGRisk and financing

Breadth as survival, and Part Two’s third finding: the fragile layer is debt with one big customer.

BET 18 · PART TWO VERDICT: HOLDSMOVING TOWARD

Revenue breadth is survival

THE BET

Companies with many revenue streams outlast single-source players. Debt-financed, single-revenue infrastructure is the most fragile part of the chain.

WHERE IT STANDS TODAY · LEADER: THE DIVERSIFIED MAJORS

Credit markets price it that way. On Nov 12, 2026, five-year CDS were 298.75bp for Oracle against 93.8bp for Amazon and 49.6bp for Microsoft. CoreWeave’s was about 1,152bp on Nov 10 after its Q3 results, down from a record 1,267.9bp on Nov 5, the day of the post-midterm selloff. The Oracle–Microsoft gap has widened from about 191bp at the end of September to about 249bp. CoreWeave’s interest expense was about a quarter of its revenue in Q2. But protection is relative: Amazon’s CDS is still more than double January’s 37.3bp and Microsoft’s is well above January’s 35.4bp, as the whole sector borrows to build.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
CDS gap: Oracle minus Microsoft~249bpEST.as of Nov 12, 2026~250bpNov 5, 2026● FLATInvesting.com: Oracle 5Y CDS daily history ↗Derived: 298.75 − 49.6 vs 305.0 − 54.7. ~191bp on Sep 30; ~109bp on Jan 2
CoreWeave 5-year CDS~1,152bpas of Nov 10, 20261,267.9bp (record)Nov 5, 2026▼ AWAYGlitch in the Credit newsletter: CoreWeave jumps 26% on $132.5B backlog (Nov 11, 2026) ↗Newsletter-reported level after Q3 results; vendor not named (it gives a 1,278bp Nov 5 close vs S&P Global’s 1,267.9bp). Prior: S&P Global data via Reuters (843bp on Sep 30)
Microsoft 5-year CDS49.6bpas of Nov 12, 202654.7bpNov 5, 2026● FLATInvesting.com: Microsoft 5Y CDS daily history ↗Peaked at 58.7bp on Oct 30. 35.4bp on Jan 1, 2026
Amazon 5-year CDS93.8bpas of Nov 12, 2026100.0bpNov 5, 2026▲ TOWARDInvesting.com: Amazon 5Y CDS daily history ↗52-week high 101.5bp; 37.3bp on Jan 12, 2026. A diversified name tightening
CoreWeave interest expense / revenue~25% (Q2 2026)EST.as of Jun 30, 2026—▲ TOWARDCoreWeave Q2 2026 press release (Seeking Alpha) ↗Derived: $640M ÷ $2,575M
Meta advertising share of revenue97.6%EST.as of Sep 30, 2026—● FLATMeta Q3 2026 results, EX-99.1 (SEC) ↗Derived; the one-stream major, now funding with equity

TREND CALLThe spread between fragile and diversified names widened into the post-midterm selloff, which is what the hypothesis predicts. This week was mixed: CoreWeave’s gap to Microsoft narrowed by about 110bp after its results (derived; CoreWeave read Nov 10, Microsoft Nov 12), while Oracle’s held near 249bp. That is one week, and only one of the two gaps, so the trigger isn’t met. The pushback: breadth buys relative safety, not immunity. Every hyperscaler is far cheaper to insure than CoreWeave, but all cost more than in January. Moves to “flat” if the Oracle–Microsoft and CoreWeave gaps both keep narrowing.

UPDATED NOV 16, 2026Part Two: Revenue breadth →Part One: The railroad bubble →
BET 19 · PART TWO VERDICT: SYNTHESIS FINDINGMOVING TOWARD

The riskiest link: debt with one big customer, now guaranteed by suppliers

THE BET

Part Two’s third finding: the riskiest place in the chain is debt-financed infrastructure with one big customer, and it now includes the supplier balance sheets, Nvidia’s above all, that have guaranteed it.

WHERE IT STANDS TODAY · LEADER: STRESS IS WHERE PREDICTED

The stress shows exactly there. CoreWeave’s CDS rose from 843bp at Sep 30 to a record 1,267.9bp on Nov 5, before easing to about 1,152bp after its Q3 results (Nov 10). The guarantor has been repriced too: Nvidia’s five-year CDS was about 139bp on Nov 6, up 76% in a month, and 139bp on Nov 12, the day after CEO Jensen Huang resigned. The biggest customer is pulling back: Microsoft is reportedly shrinking its roster of neocloud partners and declined an option to more than double its Nscale Texas deal. Oracle’s backlog is about half OpenAI; Nscale’s largest customer was 52% of its H1 2026 revenue. The common customer is the weak point: OpenAI’s run rate was about $50 billion at end-September (reported Oct 8), below the ~$70 billion widely reported, and it forecasts about $278 billion of cash burn from 2026 to 2030. Bloomberg reports its Q3 revenue topped $10 billion, up from about $6.7 billion in Q2, with an adjusted operating margin of –109%. Behind it stands Nvidia, with up to $105 billion of guarantees on OpenAI-affiliated leases (Jul 26). Two things are easing: concentration ratios are falling, and private-credit redemptions slowed in Q3.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
CoreWeave largest customer, share of revenue36% (Q2 2026)as of Jun 30, 202645% (Q1 2026)Mar 31, 2026▼ AWAYCoreWeave 10-Q, quarter ended Jun 30, 2026 (SEC) ↗Filing warns letters may refer to different customers across periods
Nscale largest customer, share of revenue52% (H1 2026)as of Jun 30, 202673% (FY2025)Dec 31, 2025▼ AWAYNscale Ltd Form S-1 (SEC, Sep 18, 2026) ↗
Oracle backlog tied to OpenAI~half of $664Bas of Aug 31, 2026—▲ TOWARDERP Today: Oracle Q1 FY27 results, $664B backlog (Sep 11, 2026) ↗
OpenAI run rate vs reported~$50B vs ~$70B widely reportedas of Oct 8, 2026—▲ TOWARDCNBC: OpenAI revenue below reported figure; Oracle, CoreWeave fall (Oct 8, 2026) ↗Oracle fell ~6% and CoreWeave ~8% on the news
OpenAI cash burn forecast, 2026–2030~$278BEST.as of Sep 19, 2026—▲ TOWARDThe Star (Reuters/FT): OpenAI forecasts cash burn near $280B by 2030 (Sep 19, 2026) ↗Company forecast, reported
Common customer: OpenAI quarterly revenue>$10B in Q3 (vs ~$6.7B in Q2); ~$50B run rateEST.as of Sep 30, 2026—▼ AWAYTechCrunch (citing Bloomberg): OpenAI’s quarterly revenue reportedly crosses $10B for the first time (Nov 11, 2026) ↗Reported by Bloomberg from anonymous sources; a stronger customer. Adjusted operating margin still –109%
Nvidia guarantees of customer leasesUp to $105B + $3.5Bas of Jul 26, 2026—▲ TOWARDNvidia 10-Q, quarter ended Jul 26, 2026 (SEC) ↗
CoreWeave 5-year CDS~1,152bpas of Nov 10, 20261,267.9bp (record)Nov 5, 2026▼ AWAYGlitch in the Credit newsletter: CoreWeave jumps 26% on $132.5B backlog (Nov 11, 2026) ↗Newsletter-reported level after Q3 results; vendor not named (it gives a 1,278bp Nov 5 close vs S&P Global’s 1,267.9bp). Still far above 843bp on Sep 30. Prior: S&P Global data via Reuters
Guarantor credit: Nvidia 5-year CDS139bp (ICE)as of Nov 12, 2026~139bp, up 76% in a monthNov 6, 2026● FLATBloomberg Tax: Nvidia default-risk gauge improves following CEO Huang’s ouster (Nov 12, 2026) ↗ICE Data Services, intraday 1:30 p.m. ET, down 2.25bp on the day after Huang’s resignation; traders said his exit could curb Nvidia’s borrowing. Prior: CNBC
Largest customer pulling backMicrosoft shrinking its neocloud roster; declined option to more than double its Nscale Texas dealas of Nov 6, 2026—▲ TOWARDCNBC: Microsoft plans to use fewer neoclouds as AI buildout shows signs of strain (Nov 6, 2026) ↗Reported, people familiar. Nscale: Reuters (Nov 5); existing contract unaffected, decision made months ago. IREN says ~48% of its backlog is tied to Microsoft (CNBC, Nov 9). Microsoft is also winding down its Oracle rental deal (Bloomberg via Reuters, Nov 6)
Blue Owl non-traded fund withdrawal requests$4.2B (Q3 2026)as of Oct 5, 2026$4.7B (Q2 2026)Jul 2026▼ AWAYInvesting.com: Blue Owl redemptions ease (Oct 5, 2026) ↗Private-credit stress easing

TREND CALLToward: AI credit stress is concentrated where Part Two said it would be, the guarantor layer (Nvidia, and Google’s lease backstops) keeps growing, and Nvidia’s own CDS has widened this autumn. Huang’s resignation adds a new question about that layer: the executive who shaped Nvidia’s customer-financing strategy has gone, and traders expect Nvidia may borrow less. The honest counterweights: CoreWeave’s spread eased after its results, the common customer is growing (OpenAI’s reported Q3 revenue topped $10 billion), the neoclouds are diversifying (CoreWeave’s top customer fell from 72% of revenue in H1 2025 to 36% in Q2 2026), and private-credit redemptions are easing. Moves to “flat” if CoreWeave’s spread keeps falling toward September’s 843bp; Nvidia’s Nov 17 results and 10-Q are the next readings.

NOT YET SOURCEDOpenAI Q2/Q3 2026 cash burn

UPDATED NOV 16, 2026Part Two: Hypotheses, judged →Part Two: The neocloud roll-up →Part One: The railroad bubble →
CYCLES & PRICESCycles and prices

Token prices, GPU rents, memory and power: whether AI is running oil’s boom-and-bust cobweb.

BET 20 · PART TWO VERDICT: HALFFLAT / MIXED

Pure token production loses value fast

THE BET

Selling tokens is a commodity business: prices fall fast and the margin goes to whoever owns something scarcer.

WHERE IT STANDS TODAY · LEADER: SPLIT: COMMODITY TOKENS FALL, FRONTIER TOKENS HOLD

Commodity token prices keep falling: Ornn’s DeepSeek index fell about 39% in the month to Nov 15, 2026; YipitData sees about a 50% annual decline (Oct 22); and Epoch estimates the price of a given level of performance falls about 47% a quarter. Closed-lab prices are stickier: Ornn’s OpenAI and Google indices rose over the same month, though Anthropic’s S-1/A says price cuts in new model versions and contractual discounts offset about $600 million of its Q2-to-Q3 revenue growth. Open-weight models handled 71.7% of OpenRouter’s top-50 tokens in September. Frontier list prices fell far less, though Anthropic cut its top price 20% on Nov 9: Claude Fable 5.2 is $8/$40 per million tokens (Fable 5.1 was $10/$50) and Opus 5.5 $4/$20, against GPT-4’s $30/$60 at launch in 2023, and Anthropic’s implied gross margin was about 85% in H1 2026.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
Ornn token indices, 1-month changeOpenAI +11.2% ($0.44/M); Anthropic –1.1% ($0.85/M); Google +31.7% ($0.80/M); DeepSeek –38.9% ($0.03/M)EST.as of Nov 15, 2026OpenAI +18.6% ($0.43/M); Anthropic –3.9% ($0.87/M); Google +10.7% ($0.63/M); DeepSeek –35.5% ($0.03/M)Nov 8, 2026▼ AWAYOrnn Token Price Index (settled Nov 15, 2026) ↗Index methodology is Ornn’s. Closed-lab blended prices rose over the month; only open-weight DeepSeek kept deflating fast
Anthropic price reductions, Q2→Q3 2026~$600M of revenue growth offset by price cutsas of Sep 30, 2026—▲ TOWARDAnthropic, PBC Form S-1/A, Amendment No. 1 (SEC, Nov 2, 2026) ↗S-1/A: “price reductions introduced as part of new model versions and contractual discounts”; Q3 revenue $17.0B
Token price decline (YipitData)~50% y/yEST.as of Oct 22, 2026—▲ TOWARDFourWeekMBA citing YipitData: token prices down ~50% y/y (Oct 2026) ↗
Price at fixed performance (Epoch)~–47% per quarter since 2023EST.as of Sep 22, 2026—▲ TOWARDEpoch AI, “The plunging price of thought” (Sep 22, 2026), via Pasquale Pillitteri ↗
Open-weight share of OpenRouter top-50 tokens71.7%EST.as of Sep 202677.5% (peak)Jul 2026● FLATOpenRouter: open-weight vs closed-weight models ↗Developer-routing sample, not the whole market
Frontier list prices (per M tokens, in/out)Fable 5.2 $8/$40 (Fable 5.1 was $10/$50); Opus 5.5 $4/$20as of Nov 9, 2026Fable 5.1 $10/$50; Opus 5.5 $4/$20Oct 31, 2026▲ TOWARDheise online: Anthropic cuts Claude Fable 5.2 prices (Nov 10, 2026) ↗A 20% cut; optional fast mode costs twice as much. claude.com lists Fable 5.1 as legacy at $10/$50 (read Nov 16). GPT-4 was $30/$60 at launch (Mar 14, 2023). Prior: claude.com/pricing
Anthropic implied gross margin85.4% (H1 2026)EST.as of Jun 30, 2026—▼ AWAYAnthropic, PBC Form S-1 (SEC, Oct 26, 2026) ↗Derived; frontier production is high-margin

TREND CALLStill half right. Commodity tokens are deflating fast, and open weights carry most developer volume. Frontier tokens are the exception: their prices fell only a few-fold in three years while capability jumped, and the leading lab books high gross margins on them. This week’s readings again point both ways: Anthropic cut its top list price 20% with Fable 5.2, while Ornn’s OpenAI and Google indices still rose over the month. One 20% cut is not commodity speed. Turns “toward” if frontier list prices start falling at commodity speed.

NOT YET SOURCEDOpenRouter open-weight share for October 2026 (chart values not published as text)

UPDATED NOV 16, 2026Part Two: Boom, bust, repeat →Part One: Jevons for intelligence →Part One: Too much oil, again →
BET 21 · PART TWO VERDICT: HOLDSMOVING TOWARD

AI runs oil-style boom-and-bust cycles

THE BET

AI follows oil’s cobweb: breakthroughs cut prices, cheap prices unlock demand, demand pushes prices back up, and the scarce input keeps moving.

WHERE IT STANDS TODAY · LEADER: UPSWING: MEMORY AND POWER ARE THE SCARCE INPUTS

The upswing is still on. Silicon Data’s H100 rental index read $3.10 an hour on Nov 16, up from $2.00 last December. Ornn’s H100 index fell 9.2% in the week to Nov 15, to $3.06, but is still up 9.3% over 30 days. B200s rent for $6.87 (Silicon Data) to $9.66 (Ornn, Nov 15). Contract prices are rising: CoreWeave says Ampere and Hopper prices rose 30–50% in Q3, and Nebius’s CFO says 2027 prices are going up. Memory is the tightest input: Micron’s gross margin hit 86.8% in its August quarter, and SK hynix has contracts covering more than half of its 2027–28 capacity. Power is next: Texas had about 474 GW of large-load requests in June, and GE Vernova’s gas-turbine backlog and reservations reached 122 GW at Sep 30.

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
H100 rental (Silicon Data index)$3.10/hrEST.as of Nov 16, 2026$3.01/hrNov 9, 2026▲ TOWARDSilicon Data: Silicon Index (read Nov 16, 2026) ↗Index; page undated, read Nov 16 (prior read Nov 9). $2.00/hr on Dec 9, 2025 (Silicon Data blog)
H100 rental (Ornn OCPI-H100)$3.06/hr (7-day –9.2%; 30-day +9.3%)EST.as of Nov 15, 2026$3.37/hrNov 8, 2026▼ AWAYOrnn markets: GPU price indices (settled Nov 15, 2026) ↗Index, settled Nov 15. H200 $6.28 (7-day –7.9%); A100 $1.02
B200 rental (Silicon Data index)$6.87/hrEST.as of Nov 16, 2026$6.84/hrNov 9, 2026● FLATSilicon Data: Silicon Index (read Nov 16, 2026) ↗Page undated, read Nov 16. Ornn’s B200 index settled at $9.66 on Nov 15 (7-day –2.8%, 30-day +16.4%); the two indices still disagree
Contract compute pricing (CoreWeave)>$40M per MW a year short-term; Ampere/Hopper prices +30–50% q/qas of Nov 10, 2026—▲ TOWARDCoreWeave Q3 2026 earnings call transcript (Motley Fool, Nov 10, 2026) ↗Company statements on the Q3 call. Nebius’s CFO says “prices for 2027 are going up” (Reuters, Nov 12)
Micron gross margin86.8% (FQ4 2026)as of Sep 3, 2026—▲ TOWARDMicron FQ4 2026 results (SEC) ↗
Memory contractsSK hynix LTAs cover >half of 2027–28 capacity; shortage expected through 2030EST.as of Oct 28, 2026—▲ TOWARDSovereign Magazine: How long will the memory shortage last? (Oct 2026) ↗Company outlook
GE Vernova gas backlog + slot reservations122 GWas of Sep 30, 2026116 GWJun 30, 2026▲ TOWARDGE Vernova Q3 2026 results (SEC) ↗
Texas large loads: requested vs energized~474 GW requested; 9,456 MW approved to energize; 4,132 MW peak drawas of Sep 30, 2026—▲ TOWARDERCOT Monthly Operational Overview, September 2026 (PDF, Oct 16, 2026) ↗Request total is ERCOT testimony (Jun 2026); energized figures from ERCOT’s September report

TREND CALLToward: prices for the newest chips and for memory are still rising, and the scarce input has moved to memory and power, the cobweb Part Two described. The first sign of the next glut would be H100 rents rolling over while Blackwell and Rubin ramp. Ornn’s H100 index fell 9.2% in the week to Nov 15, which is worth watching, but Silicon Data’s H100 index rose and CoreWeave says contract prices are rising. One week isn’t a rollover.

NOT YET SOURCEDERCOT large-load request total newer than June 2026 · TrendForce Q4 2026 DRAM contract price (forecasts conflict)

UPDATED NOV 16, 2026Part Two: Boom, bust, repeat →Part One: Titusville, 1859 →Part One: East Texas →
BET 22 · PART TWO VERDICT: AS TOTAL MONETIZATION, NOT PRICEFLAT / MIXED

Consumer AI settles at $50–100 a person a month

THE BET

Consumer spending on AI settles around $50 to $100 a person a month, like streaming plus broadband.

WHERE IT STANDS TODAY · LEADER: SUBSCRIPTIONS, ADS AND COMMERCE TOGETHER

Few people pay, though more than before: nearly 100 million of ChatGPT’s more than 1.2 billion weekly users were paying at the end of Q3 2026 (about 8%), up from about 5.6% in February. Across all of OpenAI’s revenue, including API and enterprise, that works out to about $2.80 a month per weekly user (our arithmetic). Price points are converging on $100: OpenAI added a $100 Pro tier in April, Google cut AI Ultra from $249.99 to $99.99 in May, and Claude Max is $100. The other layers are arriving: ChatGPT ads at a run rate above $1 billion (OpenAI, Oct 21), over a quarter of active Muse users shopping with it at more than $200 a purchase (Oct 21), and Amazon saying agent users spend 50% more per order (Oct 29).

LEADING INDICATORSDirection is relative to the author’s call
IndicatorLatestPriorDirectionSource
ChatGPT paying share of weekly users~8% (nearly 100M paying of 1.2B+ weekly)EST.as of Sep 30, 2026~5.6% (50M of 900M)Feb 27, 2026▲ TOWARDCNBC: OpenAI launches ChatGPT Wallet, allowing its ‘dots’ agents to make purchases (Nov 13, 2026) ↗Derived: ~100M ÷ 1.2B ≈ 8.3%; CNBC, Nov 13. Prior derived from TechCrunch (Feb 27)
OpenAI revenue per weekly user~$2.80 a month (Q3 2026)EST.as of Sep 30, 2026—▼ AWAYTechCrunch (citing Bloomberg): OpenAI’s quarterly revenue reportedly crosses $10B for the first time (Nov 11, 2026) ↗Derived: >$10B Q3 revenue ÷ 3 months ÷ 1.2B weekly users; includes API and enterprise revenue, so consumer spend is lower still
$100 consumer tiersOpenAI $100 Pro (Apr); Google AI Ultra $99.99 (May); Claude Max $100as of Oct 31, 2026Google AI Ultra $249.99 at launchMay 2025▲ TOWARDGadgetBond: Google AI Ultra drops to $99.99 at I/O 2026 (May 20, 2026) ↗Also TechCrunch (Apr 9) and claude.com/pricing
Ads on free users (ChatGPT)More than $1B run rateEST.as of Oct 21, 2026—▲ TOWARDOpenAI: How we built ChatGPT’s novel ad stack for user trust and relevance (Oct 21, 2026) ↗Company claim; ~$1B since Sep 1
Agent commerceOver a quarter of active Muse users shop with it; average purchase >$200as of Oct 21, 2026—▲ TOWARDMarketBeat: Meta Q3 2026 earnings call highlights (Oct 21, 2026) ↗Meta claim

TREND CALLMixed. The pieces of a $50–100 stack are visible: $100 tiers, ads on free users, commerce through agents. The paying share is up, to about 8% of weekly users. But OpenAI’s total revenue works out to about $2.80 a month per weekly user, and that includes API and enterprise, so consumers are spending far below $50–100. The decisive number (consumer monetization per engaged user) still isn’t disclosed by anyone. Moves when any service discloses revenue per user.

NOT YET SOURCEDAverage revenue per user for any consumer AI service

UPDATED NOV 16, 2026Part Two: $50 to $100 a month? →
METHODHow this is scored

Each hypothesis is restated as the author’s directional bet. The trend call asks one question: over the latest readings, are the leading indicators moving toward that bet, away from it, or neither?

That is different from asking whether the bet is right today. A hypothesis can trail badly and still be moving toward the call, and a hypothesis that holds today can be moving away from it.

  1. Toward. Most of the leading indicators, weighted by how directly they test the claim, moved in the bet’s direction since the prior reading.
  2. Flat / mixed. Indicators moved both ways, or barely moved, or the most important one is undisclosed.
  3. Away. The decisive indicators moved against the bet. The card says so plainly.
  4. Too early. There is no meaningful data yet. This is not a polite “away”; it means the test hasn’t started.
  5. Indicator direction. Each indicator’s arrow compares the latest reading with the prior one, judged against the bet. Where there is no prior reading, it judges the level against the claim.
  6. Estimates. Anything marked EST. is an analyst or third-party estimate, a company forecast or target, or arithmetic on reported figures. Company run-rate claims are labeled as claims.
  7. Sources and dates. Every number is dated and linked. If a figure can’t be sourced, it is not shown; it is listed under “Not yet sourced” on its card.
  8. Cadence. Numbers are refreshed weekly. Once a month, stale indicators are pruned, links re-checked and priors re-baselined. Changes to trend calls are logged in the changelog.
LOGChangelog
  • Updated 37 indicators (16 new) across 15 cards, plus 3 note-only updates: Starship Flight 15 reached orbit and the tower caught the ship for the first time (Nov 14), while the booster aborted its catch; Nvidia CEO Jensen Huang resigned (Nov 11), with CFO Colette Kress as interim CEO; CoreWeave’s Q3 results (revenue +168%, ~$132.5B backlog), its CDS easing to ~1,152bp (Nov 10) and its plan to avoid parent-level unsecured debt through 2027; Nebius cut its 2026 revenue guide; OpenAI’s reported Q3 revenue above $10B (adjusted operating margin –109%) and nearly 100M paying ChatGPT users; Tesla’s $29,990 Cybercab pre-orders, 24/7 service in all eight unsupervised cities, and Texas registry 1,306 vs Waymo 1,321; SAG’s Q3 humanoid estimate (~16,500); Baidu’s Q3 Apollo Go operations (no total ride count); Meta’s $15B dollar and €9B euro notes; Anthropic’s 20% Fable 5.2 price cut; Ornn’s H100 index down 9.2% in a week; Oracle, Microsoft and Amazon CDS through Nov 12. One chart point added: Tesla live counter 5.23M miles (Nov 14), marked estimate. No trend calls changed; tesla-vs-waymo and openai-ads stay flat.
  • Updated 30 indicators (7 new) across 12 cards: CoreWeave CDS a record 1,267.9bp (Nov 5) after the midterm selloff, with Nvidia’s CDS ~139bp, up 76% in a month; Oracle CDS 305bp and a $20B stock-plus-notes filing (Nov 2); Anthropic’s Q3 S-1/A (compute 58% of revenue, –$567M operating loss, ~$600M of price cuts); Firmus’s pulled IPO and Microsoft cutting its neocloud roster; Tesla’s San Antonio launch (8 cities), Texas registry 1,282 vs Waymo 1,285, 666 vs 4,000+ driverless cars nationally, and an NHTSA query into 2,895 Waymo vehicles; MOFCOM’s formal rare-earth pause to Jan 10, 2027; Starship Flight 15 NET Nov 13; H100 rents up 18% in a month; Goldman’s ~45% of US data-center capacity at risk from moratoriums; a working paper on ChatGPT ad conversion. Two Tesla driverless-miles chart points added. Trend changes: neocloud-consolidation flat → toward (Firmus pulled its IPO, record CoreWeave CDS, Microsoft cutting its neocloud roster). No other call changes; tesla-vs-waymo and openai-ads stay flat.
  • Monthly summary for October and early November 2026. Consolidates the launch week: launch on Nov 1 with 22 hypotheses, “The two big bets” charts, and the Nov 2 weekly refresh. Big shifts: Tesla’s Texas robotaxi registrations rose from 578 (Oct 1) to 1,204 (Oct 31) vs Waymo’s 1,259, and Fort Worth became its seventh city (Nov 2). CoreWeave’s CDS went from 843bp (Sep 30) to a record 1,167bp (Oct 22) and stayed above 1,100bp (Oct 30); Musk warned on Oct 25 that debt markets had shut for AI-only neoclouds. Meta priced $37.5B of an up-to-$95B raise (Oct 26) after negative Q3 free cash flow. Google Search growth slowed to 12% in Q3 while Cloud grew 103%. OpenAI’s ad run rate has sat near $1B since Sep 1 (restated as “more than $1B” Oct 21). No trend calls changed: tesla-vs-waymo stays flat (registrations aren’t rides; paid rides and fleet undisclosed); openai-ads stays flat (run rate stalled). Cleanup: 195 links checked, none dead, 3 redirects updated; priors re-baselined to Sep 30; OpenAI run-rate indicators merged; Tesla Austin fleet, Waymo freeway and duplicate Nvidia margin indicators dropped. No hypotheses retired or added.
  • Tracker launched with 22 hypotheses: Part Two’s twenty scorecard verdicts, its third synthesis finding on debt-financed infrastructure, and one new bet on a lunar mass driver. Indicators are dated through Nov 1, 2026.

READ THE ESSAYS BEHIND THE BETSPart One: If AI Is the New Oil, Who Is Rockefeller? →Part Two: Rebates, Rockets, and Robots →

This tracker is analysis and opinion built on dated, sourced figures. Company run-rate claims, forecasts and third-party estimates are labeled as such; where sources disagree, the card says so. Trend calls are the author’s judgment about direction, not predictions or price targets. It is not investment advice; mentions of companies are not recommendations.

DON'T TAKE MY WORD FOR IT

Research trail

WilmerHale, “The legal reality of orbital data centers” (Oct 1, 2026) ↗SpaceX Q3 2026 earnings call transcript (Motley Fool) ↗CGTN: China launches eight satellites for space-based computing (Oct 14, 2026) ↗IEEE Spectrum on the McCalip orbital-vs-ground calculator (Feb 26, 2026) ↗GPB/AP: Google launches Project Suncatcher prototype (Oct 1, 2026) ↗POWER: Abbott orders audit of Texas data-center interconnection queue (Aug 4, 2026) ↗CNBC: The AI ‘credit card’ is getting costlier as CoreWeave bond yield tops 16% (Nov 6, 2026) ↗BryceTech, Bryce Briefing Q2 2026 (PDF) ↗SpaceX S-1/A (SEC, Jun 3, 2026) ↗Wikipedia: 2026 in spaceflight (crowd-sourced tally, through Oct 31) ↗SpaceX: Starship Flight 15 (Nov 13–14, 2026; page read Nov 16) ↗Forethought, “Will we really put data centers in space?” (May 22, 2026) ↗New Space Economy: Blue Moon and New Glenn status (Oct 11, 2026) ↗Fortune: Musk details plan to make the Moon a manufacturing site (Aug 6, 2026) ↗Not a Tesla App: Understanding SpaceX’s lunar mass driver (Mar 24, 2026) ↗Astronomy: Starship reaches orbit on 14th test flight (Sep 30, 2026) ↗Spectrum News: NASA gives update to Artemis III mission (Oct 23, 2026) ↗NEI Magazine: NASA Act mandates Moon reactor (Feb 27, 2026) ↗Interlune: $6.9M NASA contract for lunar resource extraction (May 4, 2026) ↗AP via News4Jax: China postpones Chang’e-7 launch (Aug 23, 2026) ↗InsideEVs: Waymo doubles to 500,000 paid rides a week (Mar 26, 2026) ↗IDC via Webull: H1 2026 humanoid shipments near 25,000 (Sep 29, 2026) ↗Smart Analytics Global: Q3 2026 humanoid robot shipments (Nov 8, 2026) ↗The Next Web: Waymo’s Ojai robotaxi is cheaper to build (May 28, 2026) ↗Pony.ai release via AAP: Gen-7 cost targets (Apr 24, 2026) ↗Baidu 2026 Interim Report (SEC 6-K exhibit) ↗Baidu Q3 2026 results (HKEX announcement, Nov 16, 2026) ↗MOFCOM/GACC Announcement No. 58 of 2026 (Nov 9, 2026) ↗electrive: Waymo raises $16 billion (Feb 3, 2026) ↗Caixin: Unitree soars 629% in Shanghai debut (Aug 19, 2026) ↗Value Add VC: Figure AI valuation 2026 ↗TechCrunch: Tesla opens Cybercab pre-orders ‘for personal use’ with a $299 monthly fee (Nov 14, 2026) ↗Sherwood: Robotaxis are the new millennial lifestyle subsidy (Apr 20, 2026) ↗Shacknews: Tesla Q3 2026 earnings call notes (Oct 21, 2026) ↗Electrek: Tesla launches Robotaxi in San Antonio with a 274 sq mi geofence on paper (Nov 9, 2026) ↗Not a Tesla App: Tesla Robotaxi now runs 24/7 in every city (Nov 12, 2026) ↗robotaxistats: Texas robotaxi services and TxDMV registry (Tesla and Waymo re-read Nov 14, 2026) ↗TeslaNorth: Tesla’s new Robotaxi Safety Report (Oct 21, 2026) ↗Tesla North: Q3 call takeaways — Cybercab to Dallas (Oct 21, 2026) ↗Reuters via Investing.com: US opens probe into Waymo vehicles over school bus safety violations (Nov 9, 2026) ↗Alphabet Q3 2026 results, EX-99.1 (SEC) ↗PPC Land: Google ad clicks gain 8% in Q3 (10-Q, Oct 2026) ↗StatCounter: worldwide search engine market share ↗StatCounter: US search engine market share ↗Google: Q3 2026 earnings call, remarks from our CEO (Oct 28, 2026) ↗OpenAI: How we built ChatGPT’s novel ad stack for user trust and relevance (Oct 21, 2026) ↗Dealroom summary of The Information: OpenAI ad run rate ~$1B, chasing $2.4B target (Oct 18, 2026) ↗TechnologyChecker: ChatGPT statistics Q3 2026 (citing Altman at DevDay) ↗StatCounter: AI chatbot market share worldwide ↗CMSWire (Alex Kantrowitz): ChatGPT search volume gains on Google, OpenAI’s consumer lead says (Nov 9, 2026) ↗Search Engine Journal: Google Ads now runs in AI Mode across 10 more countries (Nov 10, 2026) ↗Bloomberg Government: Amazon, OpenAI in ChatGPT advertising partnership (Sep 10, 2026) ↗Digiday: Six thousand brands later, ChatGPT ads still have a conversion problem (Nov 5, 2026) ↗Business Insider: ChatGPT Ads got brands to bite. Now comes the hard part. (Nov 11, 2026) ↗Amazon Q3 2026 results, EX-99.1 (SEC) ↗Amazon Q3 2026 earnings call transcript (Singju Post) ↗amazon.com/robots.txt (checked Oct 31, 2026) ↗Engadget: Ninth Circuit vacates Amazon’s injunction against Perplexity (Aug 4, 2026) ↗MarketBeat: Meta Q3 2026 earnings call highlights (Oct 21, 2026) ↗PYMNTS: Amazon expands Alexa for Shopping to 15 countries and lets anyone shop by text (Nov 13, 2026) ↗CNBC: OpenAI launches ChatGPT Wallet, allowing its ‘dots’ agents to make purchases (Nov 13, 2026) ↗Meta Q3 2026 results, EX-99.1 (SEC) ↗PPC Land: Meta ad revenue +26%, $9.89B settlement charge (Oct 2026) ↗Meta: pricing of depositary shares and Class A common stock offerings (Oct 26, 2026) ↗PPC Land: Meta borrows €9bn in euros, at up to 6.4% until 2065 (Nov 14, 2026) ↗Diario Bitcoin: Neuralink at ~27 implanted patients (Jul 11, 2026) ↗Mezha (citing CNBC): EssilorLuxottica sold over 7M AI glasses in 2025 (Feb 2026) ↗ChannelNews: EssilorLuxottica Q3 2026 wearables (Oct 20, 2026) ↗IDC: Smart glasses growth cools to 35% (Sep 16, 2026) ↗Outlook Respawn: Reality Labs losses $83.6B through 2025 (Jan 31, 2026) ↗Drive Tesla Canada: Tesla Q3 2026 results (Oct 21, 2026) ↗Nvidia: Q2 fiscal 2027 results (Aug 26, 2026) ↗Microsoft FY27 Q1 press release, segment results (Microsoft IR) ↗Anthropic, PBC Form S-1/A, Amendment No. 1 (SEC, Nov 2, 2026) ↗The Decoder (citing The Information): OpenAI Q1 2026 revenue, burn and margin (Jun 20, 2026) ↗TechCrunch (citing Bloomberg): OpenAI’s quarterly revenue reportedly crosses $10B for the first time (Nov 11, 2026) ↗CoreWeave Q3 2026 earnings call transcript (Motley Fool, Nov 10, 2026) ↗CoreWeave: comment on Core Scientific stockholder vote (Oct 30, 2025) ↗Glitch in the Credit newsletter: CoreWeave jumps 26% on $132.5B backlog (Nov 11, 2026) ↗CoreWeave: Q3 2026 results press release (Nov 10, 2026) ↗Reuters: Nebius pares back 2026 revenue forecast on capacity delays, shares slump (Nov 12, 2026) ↗Nscale Ltd Form S-1 (SEC, Sep 18, 2026) ↗TechCrunch: Lambda to raise $4B ahead of planned IPO (Oct 6, 2026) ↗CNBC: Elon Musk says more than 200 neoclouds face ‘guppy-level death’ (Oct 26, 2026) ↗ABC News (Australia): How Firmus’s plans for the biggest IPO in three decades fell over (Oct 10, 2026) ↗CNBC: Microsoft plans to use fewer neoclouds as AI buildout shows signs of strain (Nov 6, 2026) ↗ERP Today: Oracle Q1 FY27 results, $664B backlog (Sep 11, 2026) ↗Investing.com: Oracle 5Y CDS daily history ↗PPC Land: S&P cuts Oracle to BBB- (Jul 2026) ↗CNBC: Oracle files to sell $20 billion in new stock, bonds to finance AI buildout (Nov 2, 2026) ↗SpaceX Q3 2026 results, EX-99.1 (SEC) ↗Tesla Q3 2026 earnings call transcript (Singju Post) ↗Nvidia 10-Q, quarter ended Jul 26, 2026 (SEC) ↗Nvidia: AI compute financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman, KKR (Aug 10, 2026) ↗Free Press Journal: Nvidia, AMD to invest $4B in Meta (Oct 28, 2026) ↗Nvidia Form 8-K: CEO transition (SEC, filed Nov 12, 2026) ↗Investing.com: Microsoft 5Y CDS daily history ↗Investing.com: Amazon 5Y CDS daily history ↗CoreWeave Q2 2026 press release (Seeking Alpha) ↗CoreWeave 10-Q, quarter ended Jun 30, 2026 (SEC) ↗CNBC: OpenAI revenue below reported figure; Oracle, CoreWeave fall (Oct 8, 2026) ↗The Star (Reuters/FT): OpenAI forecasts cash burn near $280B by 2030 (Sep 19, 2026) ↗Bloomberg Tax: Nvidia default-risk gauge improves following CEO Huang’s ouster (Nov 12, 2026) ↗Investing.com: Blue Owl redemptions ease (Oct 5, 2026) ↗Ornn Token Price Index (settled Nov 15, 2026) ↗FourWeekMBA citing YipitData: token prices down ~50% y/y (Oct 2026) ↗Epoch AI, “The plunging price of thought” (Sep 22, 2026), via Pasquale Pillitteri ↗OpenRouter: open-weight vs closed-weight models ↗heise online: Anthropic cuts Claude Fable 5.2 prices (Nov 10, 2026) ↗Anthropic, PBC Form S-1 (SEC, Oct 26, 2026) ↗Silicon Data: Silicon Index (read Nov 16, 2026) ↗Ornn markets: GPU price indices (settled Nov 15, 2026) ↗Micron FQ4 2026 results (SEC) ↗Sovereign Magazine: How long will the memory shortage last? (Oct 2026) ↗GE Vernova Q3 2026 results (SEC) ↗ERCOT Monthly Operational Overview, September 2026 (PDF, Oct 16, 2026) ↗GadgetBond: Google AI Ultra drops to $99.99 at I/O 2026 (May 20, 2026) ↗