Anthropic: Tokens, Value Share & the Path to $2T

Download the model

Download Anthropic Hybrid Valuation Model v1.3.3 (.xlsx) · 312 KB · Model as of November 2, 2026

Read the Read Me tab first. This is an independent illustrative scenario model, not Anthropic guidance, an IPO price target or an investment recommendation.

Tokens and subscriptions support about $1.23 trillion

Under Base assumptions and the Q3 2026 results in the November 2 S-1/A, Anthropic’s core business of tokens and subscriptions supports about $1.23 trillion of equity value, or $68.66 per economic share. That is well below the $2 trillion reference valuation.

The range around that number, from the workbook’s stress cases:

If every committed compute dollar through 2030 is spent: about $1.10 trillion under central timing. Earlier timing gives $1.26 trillion; later timing gives $644 billion, because more spending lands in 2030 and lowers the margin the exit value is based on.
If Q3’s compute spike was one-off: about $1.37 trillion, when the forward compute base is normalized and reported costs and cash stay as filed.

Every case sits below $2T. Reaching $2T in the Base case requires roughly 3.75x the assumed value-share adoption, about 56% of the revenue base by 2030, for a business line the S-1/A does not disclose today.

What the November 2 amendment changed

Q3 2026 revenue was $16.96 billion, in line with the model. Compute was not: it ran at 58% of revenue, against 48% in Q2 and the 45% the model had assumed, and included $5.0 billion of short-term capacity. Because later compute spending grows from the 2026 base, that quarter explains nearly all of the move from $1.37 trillion in the prior release to $1.23 trillion now. Version 1.3.3 incorporates the Q3 actuals, refreshes the Q4 forecasting basis and updates the provisional share bridge. Long-term Bear, Base and Bull scenario assumptions are unchanged.

The filing reports $895.4B of aggregate commitments, including $447.5B expected before the end of 2030. The workbook combines the partial dated obligations table with short-term commitments and an assumed allocation of the undated remainder. These are illustrative cash-cost stresses, with no additional revenue attributed to the capacity.

S8 central: $71.36 hybrid / $61.25 core per share ($1.28T / $1.10T). Compute spending from Q4 2026 through 2030 is $447.5B.
S8E earlier: $79.15 hybrid / $70.03 core ($1.42T / $1.26T). An alternative floor scenario; total compute is $460.71B because spending is the greater of the own-schedule path and the commitment floor.
S8L later: $47.34 hybrid / $35.90 core ($849B / $644B). Total compute is $447.5B. Timing changes 2030 margins and therefore exit value.
S9 proportional contingent haircut: $78.64 hybrid / $67.21 core. This assumes a proportional allocation; it does not identify and exclude contingent payments by year.
S10 forward normalization: $87.53 hybrid / $76.10 core ($1.57T / $1.37T). Hypothetically removes $2.233B from the forward compute starting base while preserving reported costs and historical cash.

Commitments continue after 2030. Their annual timing and relationship to ongoing compute costs remain uncertain; the exit valuation assumes the modeled 2030 earnings and cash flow are representative.

Base-case outputs, under the stated assumptions

Core equity value: $1.232 trillion. Tokens and subscriptions under the model’s Base assumptions.
Hybrid equity value: $1.436 trillion. Adds hypothetical value-share partnerships, approximately 17% above core.
Modeled value per economic share: $68.66 core; $80.08 hybrid. The share bridge is provisional and excludes unknown IPO shares and proceeds.
Compute stress: $61.25 per share (hybrid). Total compute spending is 20% higher than Base from 2027 onward, with 2026 unchanged.

These are model outputs, not observed market values. Bear, Base and Bull cases use different operating assumptions. The probability-weighted hybrid result is $125.67 per share; approximately 66% of that weighted result comes from Bull, despite a 25% Bull probability. Read the individual cases before relying on the weighted figure.

How the model works

The workbook projects core and hybrid operations through 2030 across Bear, Base and Bull. It blends a discounted earnings-multiple method with a discounted cash-flow-multiple method. Both methods use the same operating forecasts, so agreement between them is not independent confirmation. Roughly 92–93% of Base operating value comes from the exit value, making terminal assumptions decisive.

The model distinguishes earnings, actual cash flow for liquidity and economic cash flow for valuation. Stock compensation remains an economic expense; liquidity adds non-cash compensation back. Cash-basis compute is a simplifying convention. Owned-capacity depreciation is tracked by vintage for earnings and taxes. Explicit federal and state NOL schedules replace a simple tax ramp. State apportionment of 72% and a 6% state rate are assumptions.

Value share is a hypothesis to test

Value-share partnerships are a hypothetical commercial design in this workbook, not an announced Anthropic business line. The model balances revenue cuts against waived token revenue, extra compute and delivery costs, overhead, capex, taxes and collection timing. It keeps operating, cash-flow and valuation break-even tests separate: a positive operating contribution need not create cash or equity value.

The Base adoption multiplier that reaches $2T is approximately 3.75x, subject to the workbook’s live validity conditions. It is a conditional solve, not a forecast of adoption.

Stress the cash path before trusting the valuation

The stress tests cover flat revenue, higher compute, slower growth, lower exit multiples, a higher discount rate, combined downside, synthetic financing, alternative commitment floors, a proportional contingent-commitment haircut and forward compute normalization. Cases with invalid terminal economics remain unresolved; no cash-only liquidation recovery is modeled.

Flat revenue leaves an unfunded cash deficit of approximately $16.3B at the end of 2030 and a peak funding need of $21.32B against the assumed $5B cash buffer. Terminal valuation remains unresolved. A funding requirement does not establish that financing is available.

S7 illustrates a hypothetical $200B start-of-2027 compute prepayment, not a disclosed payment forecast. With an assumed $5B operating cash buffer, it needs $67.34B of net funding. Modeled funded value is $76.80, $76.16 and $75.26 per share at 0%, 15% and 30% issuance discounts, with 3% fees. The operating buffer is separate from the disclosed quarter-end liquidity covenant; using cash as a proxy does not establish covenant compliance.

Use the workbook

Read the Read Me, then inspect Inputs and the provisional share bridge.
Compare core and hybrid on each case tab before using the weighted output.
Change one material assumption at a time, then test the combined downside.
Review Stress Tests for cash lows, funding, dilution and validity flags.
Use Summary and Change Log to trace the result and its limitations.

The judgment behind the model

A valuation claim needs an operating path, a cash path and a credible claim on the resulting value. Revenue growth alone does not answer who pays for compute, how much value customers retain, or what current holders own after financing and dilution.

Keep three decisions separate: whether a revenue stream contributes operating profit, whether it produces cash and whether it creates incremental equity value. Review the downside before treating a headline valuation as a destination.

Known limits

Offering size, price range, IPO shares and proceeds remain unspecified. The definition of fully diluted shares, future Public Benefit issuance, compute commitment payment timing and lease treatment, RSU tax deductions and withholding cash, state apportionment and revenue by model tier remain uncertain or assumption-dependent. The model does not supply a liquidation recovery or a complete contractual obligation schedule.

Use the workbook’s disclosed limitations and update the inputs when offering terms and additional disclosures are available.

Sources and version history

Disclosure inputs: Anthropic PBC S-1/A, November 2, 2026, alongside the October 26, 2026 S-1. Forecasts, probabilities, terminal assumptions and value-share economics are independent model assumptions.

v1.3.3 (November 2, 2026): S-1/A actuals updated; Q4 forecasting basis refreshed; long-term scenario assumptions unchanged. Reconciled the commitment floor across Q4 2026–2030, added earlier/later floor cases and forward compute normalization, relabeled the contingent haircut, updated the RSU share bridge and added commitment input checks. v1.3.1 is preserved as the prior published release. See the workbook Change Log for full history. The next offering update follows price-range and share disclosures.