DEMAND RUNWAY · EVERCORE BASE CASE (DIGITIZED)

Token demand
goes agentic.

Hyperscaler CapEx only “funds itself” if token (and lab revenue) offtake shows up. Evercore ISI’s base case sketches the demand runway: ~235Q tokens in 2026E → ~4,075Q by 2030E, with the agentic wedge as the late-decade story (15% → 78% of the stack). Digitized here so Bubble Burster can put offtake next to the Plan Timeline’s Gerstner hurdles — not as our forecast.

01 · BASE CASE STACK

2024A → 2030E on one chart.

Evercore ISI Research · Figure 2: BASE CASE Token Consumption (CY24–CY30E) / Annual Token Usage (Base Case). As-of ~Sep 2026. Y-axis 0–4,500Q.

AGENTIC SHARE · 2026E

15%of 235Q total

AGENTIC SHARE · 2030E

78%of 4,075Q · the late-decade wedge
Base-case token consumption — stacked bars + total line (digitized)05001,0001,5002,0002,5003,0003,5004,0004,500QUADRILLIONS / YEAR2024A2025E≈100Q PROSE2026E2027E2028E2029E≈4Qe PROSE2030E
— Total (line)

2030E · Consumer 380Q · Enterprise 290Q · Agentic 3,180Q · Frontier 200Q · SLM 25Q · Total 4,075Q · Agentic share 78%

Digitized approximation from a publicly circulated chart — not an official Evercore data extract. Segment heights aimed at figure geometry; totals may differ ±rounding from the printed note.

02 · WHY IT MATTERS ON BUBBLE BURSTER

Offtake is the other side of CapEx.

The Plan Timeline and Horsemen Index already track Brad Gerstner / Altimeter’s Flight Path checkpoints — takeoff around ~$8B/mo AI-lab offtake, stall near ~$4B/mo, with a collective YE path that needs to clear a much higher bar. Token consumption is the physical demand story underneath those dollar hurdles: if agentic workloads do not scale, the revenue (and power) story behind hyperscaler spend weakens.

This page redigests a sell-side base case so we can reason about offtake geometry next to the dating board — and, later, next to CapEx vs offtake on a future /infra surface. Cross-check the Plan Timeline and Horsemen Index for the dollar side of the same bet.

03 · ASSUMPTIONS (EVERCORE’S)

BB did not invent this model.

  • Users ~1.5B → ~3.7B consumer AI users by 2030
  • Agents ~50M → ~800M AI agents by 2030
  • CAGR ~150% 4-year CAGR 2026 → 2030 (~100Q → ~4 quintillion / 4000Q)
  • Power ~250 GW of data-center compute capacity cited alongside the token path
  • Shape Agentic becomes the dominant stack by 2030E in this base case

Summarized from Evercore ISI base-case prose as circulated publicly. Bubble Burster is visualizing, not endorsing, the forecast.

04 · SOURCES / METHODOLOGY

Digitized, disclosed, not official.

Evercore prose anchors: ~100 quadrillion tokens in 2026 → ~4 quintillion by 2030 (~150% 4-yr CAGR); agentic workloads become the dominant stack late-decade.

  • Primary figure Evercore ISI Research — Figure 2: BASE CASE Token Consumption (CY24–CY30E) / Annual Token Usage (Base Case) (as-of ~Sep 2026)
  • Public discussion Public discussion of Evercore ISI base-case token demand (Syz / public note) ↗
  • Circulation Figure also circulated publicly around Sep 10, 2026 (e.g. Puru Saxena / AlphaTarget discussion).
  • Method Digitized approximation from a publicly circulated chart — not an official Evercore data extract. Segment heights aimed at figure geometry; totals may differ ±rounding from the printed note.
  • Disclaimer Sell-side estimate — not Bubble Burster’s forecast and not investment advice.