The Largest Bet in Market History
Some bets are so large they stop being bets and become facts of financial gravity. Three AI companies filing for public offerings within weeks of each other, carrying combined implied valuations approaching three trillion dollars, constitute exactly that kind of fact. SpaceX-xAI seeks to raise roughly $75 billion at a $1.75 trillion implied valuation. OpenAI filed confidentially on June 8, 2026, anchored by an $852 billion post-money valuation from its March round. Anthropic filed shortly after at a $965 billion valuation following a $65 billion raise. Together they represent the largest coordinated transfer of private AI equity into public markets ever attempted.
This is a capital markets story. The technology is the engine, but the stakes belong to anyone who holds an index fund, manages a pension allocation, or simply owns equities through a retirement account. Index funds do not choose; they absorb. When these companies enter major indices, hundreds of millions of ordinary investors will hold AI risk without selecting it. Whether that risk is priced correctly is not an academic question.
The numbers here require a caveat before they do any work. None of the three companies has yet produced an audited S-1 for public review. The figures cited derive from investor briefings, pre-IPO marketplace disclosures, and management-guided projections reported by the financial press. OpenAI's annualized revenue of $20 billion, confirmed by CFO Sarah Friar in January 2026, is a run-rate figure, not a trailing audited result. Anthropic's projection that annualized run-rate revenue will surpass $50 billion is a forward management estimate with no disclosed churn or concentration data attached. SpaceX's $135 per share offering price comes from a regulatory filing, but the $1.75 trillion implied valuation rests on structure and sentiment as much as reported financials. The deals are real. The precision of the numbers is not.
What is also real is the valuation logic these filings must sustain once quarterly reporting begins. Private companies negotiate their narratives; public companies report theirs. The moment these shares trade, the machinery of earnings calls, analyst coverage, and disclosed losses starts running. OpenAI's internal projections point to a $14 billion loss in 2026 alone. Anthropic's first profitable quarter remains prospective. The distance between the story these companies have sold to private investors and the arithmetic they will file with the SEC is the central tension of the summer ahead.
Three Deals, Three Structures, One Summer
Each deal carries its own architecture, and the differences matter more than the proximity of the filings.
SpaceX filed its S-1 on May 20, 2026, announcing 555 million shares priced at $135 apiece for a total raise of approximately $75 billion. At that price, the implied valuation exceeds $1.75 trillion, which would make it the largest IPO in history by a margin. One structural feature stands out immediately: SpaceX has reserved up to 30% of the offering for retail investors, a deliberate break from the standard book-building process in which institutional allocations dominate. Whether that reflects a populist bet on the Musk brand or a calculated strategy to broaden the shareholder base before governance critics gain institutional footing is not stated in the filing. Both interpretations fit.
Anthropic filed confidentially on June 1, one week after completing a $65 billion funding round that valued the company at $965 billion post-money. That valuation, finalized in late May, marked the first time Anthropic had exceeded OpenAI's reported valuation. The filing itself discloses nothing public yet; the $965 billion figure comes from the funding round's terms, and the revenue projections attached to Anthropic's name circulate through investor communications rather than audited statements. Management has told investors that annualized run-rate revenue will surpass $50 billion by the end of next month; Q2 revenue is projected at $10.9 billion, more than double the prior quarter. Goldman Sachs, JPMorgan Chase, and Morgan Stanley are expected to advise on the offering. An IPO as soon as fall 2026 has been reported.
OpenAI filed confidentially seven days later, on June 8. Its most recent funding round, closed March 31, 2026, raised $122 billion at $687.69 per share, producing a post-money valuation of $852 billion. Goldman Sachs and Morgan Stanley are expected as lead underwriters here too. The Wall Street Journal has reported that OpenAI is targeting a public listing in the fourth quarter of 2026, which would place all three offerings within roughly a six-month window.
Read together, the three filings represent something public markets have never absorbed at once: three companies, all projecting trillion-dollar-adjacent valuations, none yet subject to the disclosure standards that would allow independent verification of the numbers their investors have been quoting.
That source problem deserves direct acknowledgment. The figures above derive from regulatory filings where they exist, and from investor briefings, pre-IPO marketplace data, and financial press reporting where they do not. SpaceX's share price and share count come from its S-1. Anthropic's $965 billion valuation comes from its May funding round terms. OpenAI's $852 billion post-money figure comes from its March round. The revenue projections attached to Anthropic and OpenAI are management-guided estimates communicated to private investors, not trailing figures reviewed by auditors. Anthropic's $50 billion annualized run-rate target carries no disclosed customer concentration data, no churn figures, no cohort analysis. OpenAI's $20 billion annualized revenue figure, confirmed by CFO Sarah Friar in January 2026, is a run-rate snapshot, not a full-year audited result.
None of this makes the deals less real. It makes the numbers less precise than their apparent specificity implies. Three filings, three structures, three sets of figures that will require an S-1 to resolve. The market is pricing them regardless.
Revenue, Losses, and the Arithmetic of Belief
Start with OpenAI, because its numbers are the most public and the most stark.
CFO Sarah Friar confirmed in January 2026 that OpenAI's annualized revenue had passed $20 billion, up from $6 billion in 2024. That comparison travels well in a press release. It travels less cleanly through an audit. The $20 billion figure is a run-rate snapshot: monthly revenue at one point in time multiplied by twelve, not a full fiscal year of recognized revenue reviewed by an external auditor. The $6 billion figure from 2024 is a prior-period reference from the same management channel. The implied growth is real and probably directionally accurate. The precision is not.
At the $830 billion valuation from its March 2026 round, OpenAI's price-to-sales ratio sits at roughly 65 times 2025 revenue. That multiple is approximate in both numerator and denominator, since the denominator is the annualized run-rate figure rather than an audited annual result. Sixty-five times revenue is a valuation that requires the growth to continue and the losses to eventually stop. Neither is guaranteed.
The losses are large and structural. Internal projections put OpenAI's 2026 loss at $14 billion. The company carries roughly 18 to 24 months of operating runway on its current cash position. HSBC analysts estimate it may need more than $207 billion in additional funding by 2030, even after accounting for projected revenue growth. That figure is one analyst's point estimate, not a consensus view, and the methodology behind it is not public. Treat it as a framing device rather than a forecast. The underlying pressure it describes is real regardless: a company spending enormously on compute, talent, and infrastructure while its core product still generates less cash than the operation consumes.
Stargate makes this concrete. OpenAI stood beside the president in January 2026 to announce a $500 billion AI infrastructure project. Banks declined to underwrite it when OpenAI tried to self-finance. The company subsequently walked away from a planned expansion in Abilene, Texas. Total projected compute for the initiative was revised down from $1.4 trillion to roughly $600 billion. Instead of owning data centers, OpenAI now looks toward renting significant capacity from Amazon. The arc of that story captures the structural tension more vividly than any ratio: extraordinary revenue momentum, and a capital requirement that revenue alone cannot yet service.
Anthropic's numbers carry different problems. The company told investors its annualized run-rate revenue will surpass $50 billion by the end of the filing month, having reportedly experienced an 80-fold increase in annualized revenue over some prior period. It also reported expectations of posting $10.9 billion in revenue for the second quarter, more than doubling from the prior three-month period. These are management communications to private investors, not disclosed financials. The $50 billion target has no accompanying churn data, no customer concentration disclosure, no cohort analysis showing whether the revenue base is durable or front-loaded by a small number of enterprise relationships. Anthropic is also reported to be on pace for its first profitable quarter, which, if sustained, would distinguish it sharply from OpenAI. The qualifier matters: one quarter of operating profit is a waypoint, not a path to profitability.
The market is pricing all of this on belief, calibrated against momentum. Anthropic's $965 billion valuation briefly eclipsed OpenAI's, the first time the younger company had crossed that threshold. OpenAI's $852 billion post-money from March 2026 implies the gap is tight and unstable. Both figures derive from private round terms, where investors negotiate on projected trajectories rather than current fundamentals. The 65x multiple on OpenAI, and whatever equivalent multiple attaches to Anthropic at near-parity valuation with a smaller revenue base, describes a market buying the story of what these companies will earn, not what they have earned. Public markets will eventually demand the latter.
Who Controls What You're Buying
Public shareholders buying into any of these three offerings will own equity. What they will not own, in any meaningful sense, is control. The governance architectures underneath these companies are not incidental to the investment thesis; they are the investment thesis.
Start with SpaceX-xAI. Elon Musk holds 12.3% of Class A shares and 93.6% of Class B shares, with Class B carrying ten votes per share, producing a combined voting power of 85.1%. SpaceX filed its S-1 with controlled-company status, which exempts it from Nasdaq's requirement that a majority of the board be independent. Musk serves simultaneously as CEO, CTO, and chairman. A public shareholder voting against any board resolution has roughly the same effect as not voting at all. The xAI merger compounds this. That transaction transferred xAI's losses, $6.4 billion in 2025 and widening into 2026, onto SpaceX's balance sheet. Morningstar flagged the merger as a material threat of value destruction. The controlled-company exemption means shareholders have no structural mechanism to challenge that judgment. AkademikerPension, the Danish pension fund, reached its own conclusion before the IPO even priced, placing SpaceX on its investment blacklist and describing the governance structure as "catastrophic". Governance-focused institutional capital will not be the buyer here. The retail allocation of 30% of the offering suggests the deal knows its audience.
Anthropic's structure is more intricate and, in certain respects, without precedent in the history of public markets. The company has three share classes: Class A for public investors carrying one vote per share; Class B for founders Dario and Daniela Amodei carrying super-voting rights at a ratio still undisclosed pending the S-1; and Class T, held by the Long-Term Benefit Trust. That last class demands attention. The LTBT consists of five AI safety experts who hold no equity in Anthropic, receive no economic benefit from its success, and yet retain sole authority to elect and remove a designated portion of the board. Over time, that portion escalates until the LTBT elects a majority of directors. No technology company has gone public with a structure in which a non-shareholder body holds escalating authority over board composition. A public investor buying Class A shares is purchasing a claim on future cash flows governed by a trust whose members have no financial alignment with those cash flows whatsoever.
Amazon's position illustrates the paradox cleanly. In Q1 2026 alone, Amazon booked $16.8 billion in pre-tax gains from its Anthropic stake, including $12.3 billion from an upward revaluation following the latest funding round. Amazon has committed $8 billion to Anthropic as its primary cloud provider. It has zero governance rights, no board seats, and no board observer rights. Google owns 14% of the company, capped contractually at 15%, and holds no voting rights and no board representation. The largest financial stakeholders cannot direct the company. That is the design. The $80 billion in compute commitments Anthropic has made through 2029, flowing primarily to Amazon and Google, will require SEC scrutiny as related-party transactions in the prospectus. There is also the FTX legacy: FTX held approximately 8% of Anthropic before its collapse, with the stake liquidated for $884 million through bankruptcy proceedings to buyers including Abu Dhabi's Mubadala. Anthropic's cap table was partially shaped by a criminal bankruptcy process. That sentence will appear, in some form, in the risk factors.
OpenAI's governance story is best understood through what happened in November 2023. The nonprofit board that fired Sam Altman had no financial stake in the company. That design feature, intended to insulate mission from commercial pressure, became the vulnerability that nearly destroyed the company when investors and employees forced Altman's reinstatement within days. The restructuring that followed converted OpenAI from its capped-profit LP structure to a Public Benefit Corporation in 2025, but it left a specific question unanswered: Altman's equity stake remains listed as TBD in IPO filing analysis. The CEO of a company targeting a valuation above $800 billion has an undisclosed ownership position. Microsoft owns approximately 27% of OpenAI following its $13 billion investment and receives 20% of OpenAI's revenue under their current agreement. The nonprofit Foundation retains approximately 26%. Public shareholders in a Q4 2026 listing would own a minority economic interest in a company whose largest single revenue obligation runs to a strategic partner that is simultaneously a competitor in cloud infrastructure and enterprise software.
Across all three offerings, the question is whether investors understand that buying shares in any of these companies differs fundamentally from buying a proportional stake in the enterprise's future direction. The voting math, the trust mechanisms, and the revenue agreements have already decided who steers. The public market is being invited to fund the journey.
Regulatory Fire and the Pentagon Confrontation
Of the three companies preparing to file, Anthropic carries the most combustible near-term legal exposure. The dispute with the Pentagon did not begin as litigation; it began as a deadline.
On February 24, 2026, Defense Secretary Pete Hegseth gave Anthropic CEO Dario Amodei until 5:01 p.m. on February 27 to allow unrestricted use of the company's AI models for all legal purposes. The underlying conflict was specific: Anthropic had maintained usage restrictions limiting military applications it considered dangerous, including autonomous weapons and mass surveillance. Hegseth's demand was that those restrictions go. When Amodei did not relent, President Trump directed federal agencies to cease using Anthropic's products, and Hegseth formally designated the firm a supply-chain risk. That designation carries particular legal weight; it is the category typically reserved for foreign adversaries, not American AI startups.
Anthropic sued the federal government on March 9, filing in both the Northern District of California and the D.C. Circuit Court of Appeals, arguing the administration's actions had caused irreparable harm. On March 26, Judge Rita Lin granted a preliminary injunction, blocking enforcement of the ban and the supply-chain designation on grounds that the government had likely taken retaliatory actions that violated the law.
A preliminary injunction is a finding of likely merit, not a verdict. The litigation remains active. Any S-1 Anthropic files must disclose ongoing material litigation, which means every prospective investor reads this dispute in the risk factors before reaching the financial projections. That placement in a prospectus is not cosmetic; it signals that a federal court has found the government's conduct likely unlawful but has not resolved whether Anthropic recovers its former agency contracts, faces renewed restrictions, or wins cleanly. The government can appeal. The underlying policy disagreement about what restrictions an AI company may impose on military customers has no obvious resolution in law, because no settled law governs it.
The White House's March 2026 national AI policy framework cuts against this risk in one specific direction. The framework recommended a single federal standard preempting state AI laws and explicitly called for relying on existing sector-specific regulators rather than creating any new regulatory body. That posture signals an administration more interested in accelerating deployment than in regulating it. For Anthropic's commercial operations outside the Pentagon dispute, a light federal touch is favorable. The framework does nothing to resolve the existing lawsuit, however, and the administration that issued it is the same one that designated Anthropic a supply-chain risk three weeks earlier. Both facts belong in any honest reading of the regulatory environment.
The SEC disclosure challenges extend beyond the Pentagon matter and differ by company. OpenAI's conversion from a nonprofit to a Public Benefit Corporation in 2025 has no clean precedent for how a registrant discloses the residual claims and obligations of its predecessor nonprofit structure; the Foundation's retained 26% stake and the governance rights that flow from it will require disclosure treatment the SEC has never reviewed. Anthropic's S-1 will need to address both the LTBT's escalating board authority and the $80 billion in compute commitments running primarily to its two largest equity investors; that related-party structure demands disclosure depth well beyond standard related-party tables. SpaceX's controlled-company exemption must be disclosed, and its shelf life depends on Musk retaining voting control; any scenario in which he does not creates governance discontinuity that prospective investors cannot price from current filings.
None of these disclosure challenges prevents the offerings from proceeding. They determine how much friction each company encounters in SEC review and how much risk language lands in front of retail buyers before the roadshow begins.
The Competitive Clock Public Markets Start
Once these companies file, the quarterly clock starts. Private companies govern their own narrative; public ones answer to consensus estimates, and the market does not pause between capability announcements.
The competitive pressure is already visible, though the data deserve caution. According to Similarweb figures cited in pre-IPO investor materials, ChatGPT's web traffic share among AI assistants fell from 86.7% to 64.5% over twelve months, while Google Gemini's share rose from 5.7% to 21.5% over the same period. Similarweb measures web traffic, not revenue, not retention, not enterprise penetration; the numbers are directionally plausible but methodologically unverified as a proxy for competitive position. What they illustrate is that 800 million weekly active users does not insulate a product from share erosion when a better-capitalized rival ships faster. Public investors will see that chart. They will ask about it every quarter.
The agentic workflow transition is the structural reason these three companies chose 2026 rather than waiting. Through 2025, the dominant question in financial media was whether AI infrastructure spending had decoupled from any realistic revenue base; Sequoia's analysis noted that the combined revenue of all AI companies could not justify the compute being built. What shifted sentiment was the actual deployment of agents into production workflows, consuming large volumes of inference tokens and delivering enterprise value concrete enough to measure. Private equity is now funding OpenAI and Anthropic joint ventures specifically structured around agentic workflow implementation; consultancies have repositioned toward shipping agents directly into enterprise accounts rather than advising on strategy. The value sits in the completed workflow, not the underlying model. That shift from research artifact to operational infrastructure gave these companies a revenue story credible enough to bring to public markets.
Anthropic's specific product bet within this transition is Claude Code, an agentic coding assistant competing in a segment where measurable productivity gains are already documented at the enterprise level. Coding agents are the most commercially legible form of agentic work: the input is a task, the output is working software, and the value is traceable without a consulting overlay. Claude Code represents a deliberate choice to own a wedge of the agentic market where buyers have both the budget and the technical capacity to evaluate what they are purchasing.
None of this competitive momentum resolves the structural tension public markets will apply. Quarterly reporting disciplines companies that have operated on investor briefings and managed run-rates. Once OpenAI is public, the equivalent of its $20 billion annualized figure becomes a reported quarter with disclosed churn, segment breakdowns, and a comparison period. The narrative of exponential growth either survives contact with GAAP accounting or it does not. Public markets do not extend the benefit of the doubt that pre-IPO rounds do; they price the current quarter and punish anything that reads as stagnation. Companies that have spent years operating under private narrative control are about to find out what that discipline actually costs.
What the Capital Rotation Means for Everyone Else
Three simultaneous mega-IPOs do not merely test investor appetite for AI. They restructure the entire capital stack beneath it.
The most immediate pressure falls on smaller AI companies still in private markets. When liquid, index-eligible positions in OpenAI, Anthropic, and SpaceX become available, allocators who currently hold illiquid stakes in Series B and C companies face a straightforward choice: stay locked up in something unproven, or rotate into a named position they can exit. The compression this creates for sub-unicorn AI startups is real. Capital that once funded the next tier of the market will gravitalize toward the liquid mega-caps, and the median path to IPO for venture-backed companies is already eleven years. For founders currently raising, the window narrows further.
Index fund mechanics carry this rotation into portfolios that never made an active AI bet. Once any of these three companies clears the size and liquidity thresholds for S&P 500 inclusion, every passive investor holding a broad equity index owns a slice, without deliberation and without the ability to opt out. Pension funds, sovereign wealth vehicles, and retail retirement accounts will carry AI concentration risk they did not choose.
Some institutional capital will refuse on governance grounds. AkademikerPension placed SpaceX on its investment blacklist before the IPO, describing the governance structure as catastrophic. Funds with explicit governance mandates cannot absorb a controlled company where 85.1% of voting power rests with a CEO who is simultaneously CTO and chairman. For those funds, SpaceX is an automatic exclusion, independent of any risk-reward calculation.
Anthropic's PBC structure and LTBT architecture are partly designed to attract the capital that SpaceX repels. An ESG-mandated fund that cannot hold SpaceX can hold a Public Benefit Corporation with a stated mission of responsible AI development. The LTBT's non-shareholder governance reads as a liability to some allocators and as a moat to others. Whether the SEC views it as a disclosure challenge or simply an unusual but legal structure will determine how much of that ESG-adjacent capital actually arrives at the offering.
The dot-com parallel surfaces here, not as a prediction but as a structural question worth sitting with. In the late 1990s, the scarcity of public internet exposure drove valuations to levels the underlying businesses could not support. When share supply increased, the narrative of scarcity dissolved faster than the businesses themselves. Sun Microsystems fell from $70 to $5. The companies were real; the price was a function of who could get in. Today the mechanism operates the same way: the price is set by willingness to buy, and demand for these shares reflects eagerness to participate in the AI moment more than it reflects audited cashflow. Supply was scarce because these companies were private. Supply is now arriving in quantity.
Several questions will determine whether this cohort writes a different ending. Sam Altman's equity stake remains listed as TBD in IPO analysis, an extraordinary omission for a CEO leading a company toward a trillion-dollar valuation. The LTBT's escalating board-control mechanism has no precedent in public markets and will receive SEC scrutiny that no private filing required. Microsoft's 20% revenue share, manageable in a private context, becomes a disclosed line item in a public income statement that analysts will model and challenge every quarter. And Anthropic's path to profitability runs on a clock set partly by OpenAI's IPO timeline: if OpenAI lists first and absorbs the market's appetite for pure-play AI exposure, Anthropic's pricing leverage shrinks. The race is not only technological. It is calendrical.