The AI IPO race — OpenAI and Anthropic.
Two of the largest IPOs in history are forming at once, at valuations set in private and about to meet public scrutiny for the first time. The SpaceX question — what are we actually pricing? — returns, now for revenue that is real, growing, and deeply unprofitable.
Cabier Intelligence · 11 June 2026 · ~10 min read
Executive summary
OpenAI and Anthropic have filed within weeks of each other for what would be two of the largest IPOs ever brought to market — reported valuations near $730–850bn and $965bn respectively, against annualised revenue in the $20–30bn range and material losses funded by hyperscaler capex. The institutional question is the same one SpaceX posed: how much of the ask is disclosed fundamentals, how much is discounted optionality, and how much is narrative premium named honestly.
The differences from SPCX matter. Revenue is real and scaling at unprecedented rates. Customers are enterprise, not retail. Optionality is concentrated in frontier capability rather than in a planetary settlement programme. And the governance instrument is the public-benefit corporation, with reserved authorities held by non-profits and Long-Term Benefit Trusts rather than by super-voting founder shares.
OpenAI — reported figures
- IPO status
- Confidential filing; target ~Sep 2026
- S-1 not yet public; figures from reported guidance.
- Reported valuation
- ~$730–850bn (some refs >$1tn)
- Range reflects underwriter and secondary-market marks.
- Raise (reported)
- ≥ ~$60bn
- Would rank among the largest US offerings on record.
- Revenue (ARR)
- ~$25–30bn early-2026
- Up from ~$20bn at end-2025.
- Profitability
- Projected ~$14bn 2026 loss
- ~$44bn cumulative to ~2029 per reported guidance.
- Structure
- Public-benefit corporation
- Microsoft retains ~27% economic interest.
Anthropic — reported figures
- IPO status
- Confidential filing ~1 Jun 2026
- Filing order suggests a Q4 2026 / Q1 2027 listing window.
- Reported valuation
- ~$965bn
- Per most-cited secondary-market and underwriter marks.
- Raise (reported)
- Not yet reported
- Will be disclosed in the public S-1.
- Revenue (ARR)
- 2026 target ~$20–26bn
- Enterprise API and Claude commercial deployments.
- Profitability
- Loss-making, magnitude undisclosed
- Burn driven by training and inference capex.
- Structure
- Public-benefit corporation
- Long-Term Benefit Trust holds reserve voting authority.
The setup — two record-scale listings, weeks apart
Anthropic filed confidentially around 1 June 2026. OpenAI's filing window is targeted at September. Both issuers would, on reported marks, rank among the largest public companies in the world on day one of trading. The compression matters: allocator attention, underwriter syndicate capacity, index-inclusion mechanics and the infrastructure read-through all interact when two listings of this scale form inside a single quarter.
The market context is the SPCX precedent. Whether SpaceX's offering mechanic — fixed price, thin float, outsized retail allocation — is referenced as template or as cautionary tale will shape the offering documents and roadshow materials of both AI issuers. Risk committees should expect the parallels to be drawn explicitly.
Applying the three-layer framework
Cabier's standing framework — fundamentals, discounted optionality, named narrative premium — applies here with a different weighting than at SPCX. Layer 1 is materially larger: enterprise API and product revenue is contracted, growing, and increasingly concentrated in disciplined enterprise procurement processes rather than in consumer-cycle subscription churn. Layer 2 is the AGI / frontier-capability bet, which a disciplined valuation method treats as a portfolio of real options on capability unlocks, each with its own probability and payoff. Layer 3 is the residual — the part of the ask not defended by either fundamentals or a defensibly discounted option. The discipline is to name it, not to absorb it into a multiple.
The honest answer for both issuers is that, on reported marks, Layer 3 is material. Allocators who underwrite the ask underwrite a view on capability trajectory; those who do not should size accordingly.
OpenAI — revenue, burn, and the path to profit
The most cited figures are an ARR step-up from roughly $20bn at the end of 2025 to $25–30bn early in 2026, a projected 2026 loss near $14bn, and cumulative losses to ~2029 of about $44bn before a credible break-even. Microsoft's ~27% economic interest, the partnership architecture, and the exclusivity terms on compute and distribution will require granular S-1 disclosure. The path-to-profit narrative depends on inference unit economics improving faster than training capex scales — a claim that is testable but, for now, asserted rather than demonstrated.
Anthropic — the safety-mission listing
Anthropic's reported $965bn mark sits above OpenAI's mid-range despite a similar revenue trajectory and a smaller commercial footprint by some measures. The premium appears to price safety-research leadership and a perceived governance discipline. The Long-Term Benefit Trust holds reserved voting authority on certain material decisions; the public S-1 will need to disclose the scope and the mechanics of those reserves with enough granularity for an allocator to price the minority position.
Governance — public-benefit corporations and the minority discount
The PBC form legally permits the board to weigh mission against shareholder return. For OpenAI, the operating company sits beneath a non-profit parent with reserved authorities. For Anthropic, the Long-Term Benefit Trust holds analogous reserves. Neither is a defect; both are deliberate. The discipline is to price the minority position for what it is — equity in an entity where some decisions are not made under a shareholder-primacy mandate — and to apply a control discount accordingly.
Infrastructure read-through — Nvidia, Oracle, CoreWeave
A material share of Nvidia GPU allocation, Oracle and CoreWeave capacity, and hyperscaler capex flows toward these two issuers. The infrastructure stack's market capitalisation is partly underwritten by the assumption that AI capex continues to grow at current pace. A re-rating of either issuer transmits into the stack; a re-rating of the stack transmits back into the issuers' cost structures. The concentration is reciprocal and should be modelled symmetrically.
Market-structure risk — two mega-IPOs in one window
The historical pattern when two record-scale offerings form inside a tight window is that the second prices at a discount, or that one is delayed. Underwriter syndicates overlap; allocator budgets are finite; index-inclusion mechanics will compete. Cabier's diligence cadence assumes both deals close in sequence and both deals close at the wider end of the indicated range — the prudent base case, not the roadshow case.
Precedent — what risk committees should test
Three tests, in order. First, the sum-of-the-parts walk: every dollar of the ask allocated explicitly to fundamentals, discounted optionality or named narrative premium. Second, the governance discount: the price an allocator would pay if the PBC and Trust reserves were exercised against the shareholder-return objective in a stress. Third, the suitability defence: whether the retail allocation, if outsized, can be defended on the same standard a regulator would apply to any other listing. The deals are different from SPCX; the standing questions are the same.
References and citations
Primary sources. Positions change; verify at source before relying on any figure or determination.
- 1US Securities and Exchange Commission, Division of Corporation Finance — Form S-1 and Regulation S-K guidance — Disclosure framework applied to the listing-window analysis.Source
- 2US Securities and Exchange Commission, EDGAR filing record, accessed Q3 2026 — Primary source for any filed registration statement referenced.Source
- 3NIST AI Risk Management Framework (AI RMF 1.0) and Generative AI Profile — Control reference for the model-governance layer of the valuation framework.Source
- 4European Union, Regulation (EU) 2024/1689 (AI Act) — general-purpose AI obligations — Regulatory exposure referenced in the risk-factor discussion.Source
- 5Reuters, Financial Times and Bloomberg reporting through Q3 2026 — Secondary reporting used for market context only.
Named sources
- The Statesman, CMC Markets, AI Weekly, Euronews, Reuters / WSJ, Investing.com — May–June 2026 reporting cited above.
- Public S-1 filings — Direct references will be added to this article as the public S-1s are filed and amendments are released.
Frequently asked questions
- Why are OpenAI and Anthropic listing in the same window?
- Both have reached a scale where private capital can no longer fund their next phase of compute and capex without a public-market step-up. The confidential filings landed within weeks of each other because the AI capital cycle is synchronised — the same Nvidia allocation, the same hyperscaler partnerships, the same end-customer demand profile.
- Does ARR of $25–30bn justify a $730–850bn valuation for OpenAI?
- Not on conventional multiples. Even at the high end of reported ARR, the implied price-to-sales ratio is 25–35x against a business projected to lose ~$14bn in 2026 and ~$44bn cumulatively through ~2029. The gap is optionality — frontier capability, agentic platforms, enterprise penetration — and a residual narrative premium. The institutional discipline is to name each layer rather than absorb it into a single multiple.
- How does the SpaceX framework apply here?
- Identically in structure, differently in weighting. Layer 1 (fundamentals) is much larger for the AI issuers than for SpaceX — real recurring enterprise revenue, contracted hyperscaler deployments. Layer 2 (discounted optionality) is the AGI / frontier-capability bet, which must be valued as a real option, not as base-case enterprise value. Layer 3 (narrative premium) is the residual the institution decides whether to underwrite.
- What does the public-benefit-corporation structure mean for minority shareholders?
- The board is legally permitted, in some cases required, to weigh mission against shareholder return. For OpenAI, the operating company sits beneath a non-profit parent with reserved authorities; for Anthropic, the Long-Term Benefit Trust holds reserve voting power. A disciplined minority shareholder applies a control discount that does not depend on mission attainment.
- Why does Microsoft's ~27% interest in OpenAI matter?
- It is the single largest external economic interest in the issuer, the counterparty on the most material commercial agreements (compute, distribution), and a competitive AI principal in its own right. The S-1 will need to disclose related-party economics, exclusivity terms, and the consequences of any change-of-control on the partnership architecture.
- What is the infrastructure read-through?
- OpenAI and Anthropic together drive a material share of Nvidia GPU allocation, Oracle and CoreWeave capacity contracts, and the hyperscaler capex cycle. A re-rating of either issuer transmits into the infrastructure stack — and the stack's market cap is, in turn, partly underwritten by the assumption that AI capex continues to grow at current pace. The concentration is reciprocal.
- Could two simultaneous mega-IPOs force a valuation cut?
- It is the live institutional question. Allocator budgets are not infinite; underwriter syndicates overlap; index inclusion mechanics will compete. The historical pattern is that the second deal in a tight window prices at a discount to the first, or that one is delayed. Cabier's diligence cadence assumes both scenarios.
- How does this connect to the SpaceX listing?
- SpaceX is the precedent. The fixed-price mechanic, thin float, retail-allocation profile and founder-control structure of SPCX will be referenced — either as a template to follow or a template to repudiate — in the AI listings' offering documents and roadshow materials. Risk committees should test both deals against the same standing questions.
- What should an institutional allocator be testing first?
- Three things. First, the sum-of-the-parts walk, with optionality named separately from fundamentals. Second, the governance-discount calculation, given PBC structures and reserved control. Third, the suitability defence for retail allocation, given the magnitude and the absence of a public operating history at this scale.
- Does Cabier express a price view?
- Cabier expresses a view on the methodology and the governance structure. We do not publish price targets and we do not issue recommendations to buy, sell or hold.
- Will Cabier engage on the listings directly?
- Yes — through the IPO Governance Watch desk, under signed terms, with the institutional risk committee, the chief investment risk officer, and the suitability function. Conversations begin under non-disclosure.
- Is there a public Cabier price list?
- No. Every engagement is custom-quoted under signed terms.
- What is the regulatory perimeter forming around this class?
- Securities disclosure on one side, the EU AI Act's general-purpose model obligations on the other, and prudential interest wherever regulated institutions are exposed to the issuers.
- Is this cohort systemically relevant?
- Increasingly. When banks, insurers, and public administrations depend on the same small set of model providers, provider failure becomes an operational-resilience question for their supervisors.
- What concentration should investors measure?
- Revenue by counterparty, compute by supplier, model dependency by weight provider, and capital by investor. Where the same names recur across all four, the exposure is one risk, not four.
- What is the exit risk for enterprise customers?
- Substitution cost. Where prompts, fine-tunes, and workflows are provider-specific, the customer carries concentration risk it has not priced. Contractual portability is the control.
- Do open-weight alternatives reduce the concentration?
- They change its shape. Hosting, evaluation, and accountability move in-house — a governance gain and an operational burden. The dependency does not vanish; it becomes yours to assure.
- What would change our reading of the class?
- Disclosed diversification of compute supply, durable enterprise contracts with portability terms, and an AI governance record that survives independent testing.
Glossary
- ARR
- Annual recurring revenue — the metric most often cited in pure-play software and AI valuations.
- Burn rate
- The rate at which a loss-making business consumes cash, net of revenue.
- Path to profit
- The set of operating assumptions under which losses converge to break-even within a credible time horizon.
- Public-benefit corporation (PBC)
- A US corporate form that legally permits the board to weigh mission against shareholder return.
- Long-Term Benefit Trust
- Anthropic's governance instrument holding reserved voting authority on certain material decisions.
- Reserved control
- Voting or veto rights conferred on a class of holders independent of equity ownership.
- Sum-of-the-parts (SOTP)
- Valuation by separating an enterprise into its constituent businesses and valuing each on its own basis.
- Real-options overlay
- Valuing strategic optionality as discounted real options rather than embedded base-case revenue.
- Narrative premium
- The portion of valuation not supported by disclosed fundamentals or a defensibly discounted real option.
- Frontier capability
- The leading edge of model performance, generally undisclosed in detail at filing.
- Related-party economics
- Commercial terms between the issuer and a holder of a material economic interest.
- Hyperscaler capex
- Capital expenditure by the largest cloud platforms on compute and data-centre infrastructure.
- Suitability
- The regulatory obligation on intermediaries to ensure a security is appropriate for the investor offered it.
- Index-inclusion mechanics
- Rules governing entry into major equity indices and the forced buying they generate.
- Control discount
- The valuation reduction a disciplined investor applies to securities whose voting structure permanently subordinates minority holders.