The insider's guide to Anthropic's path to the public markets: the filing, the cap table, the compute bill, the governance no exchange has ever seen, and the parts the headlines skip.
On June 1, 2026, Anthropic confidentially submitted a draft Form S-1 to the SEC, four days after closing a $65 billion round that valued the company at $965 billion - Anthropic. Six months earlier the same company had flatly denied it was going public. That reversal, and everything packed inside it, is the real story.
The headlines all said the same thing: the most valuable AI startup on earth is heading for Wall Street. What the headlines mostly missed is why now, who actually controls the company, where the money really comes from, and whether the $965 billion number means anything. The answers sit in places most coverage never looks: a Delaware benefit-corporation charter, a purpose trust that quietly took control of the board in April, a gross-margin line that behaves nothing like software, and a compute bill so large that private capital can no longer carry it alone.
This guide goes deep on all of it. It starts with exactly what happened on June 1, then works through the valuation ladder from a $124 million seed in 2021 to a near-trillion-dollar mark, the revenue and burn behind the run-rate, the compute deals that explain the timing, the governance structure that has no precedent on any exchange, the race with OpenAI that set the clock, the bubble debate in honest terms, the under-reported angles (Gulf money, regulation, the jobs reversal), and finally what carries the valuation and what an Anthropic IPO would actually mean. It assumes you are smart but not an insider. By the end, you will understand this listing better than most people who will buy the stock.
Contents
- The filing that flipped the script: what actually happened on June 1
- The valuation ladder: from a $124M seed to $965 billion
- Reading the books before the books are public: revenue, margins, and burn
- The compute bill behind the IPO: why a private company needs public markets
- The governance puzzle Wall Street has never seen: PBC, the Trust, and Class T
- The race that set the clock: Anthropic versus OpenAI
- Bubble or build-out: the honest case for and against $965 billion
- The parts the headlines skip: Gulf money, regulation, and the jobs reversal
- What carries the valuation: Claude, Claude Code, and the moat
- What an Anthropic IPO would actually mean
1. The filing that flipped the script: what actually happened on June 1
The single most important sentence in this entire story is the one Anthropic published itself. "Today, Anthropic, PBC confidentially submitted a draft registration statement on Form S-1 to the U.S. Securities and Exchange Commission for a proposed initial public offering of our common stock" - Anthropic. The company added that "the number of shares to be offered and the price have not yet been set" and that going public remains an option contingent on SEC review and market conditions. That language matters, because it is deliberately non-committal. A confidential draft S-1 is not a commitment to list. It is a company buying itself the right to list quickly if and when it decides the window is open.
To understand why this is a flip, rewind to December 5, 2025. Anthropic's communications chief Sasha de Marigny told reporters the company was "keeping its options open" with "no immediate plans" to go public - Axios. Roughly six months later, the draft S-1 was filed. Companies do not casually reverse a public denial of this kind in half a year. Something changed, and that something was not a sudden surge of investor demand (there was never any shortage of that). It was the competitive and financing calendar, which we will trace through the rest of this guide.
The funding context is what gives the filing its weight. The S-1 landed only four days after Anthropic closed a $65 billion Series H at a $965 billion post-money valuation, a mark that vaulted it past OpenAI's last private valuation of roughly $852 billion - Fortune. For a brief moment, the company best known for being the safety-first AI lab became, on paper, the most valuable startup in the world. Multiple outlets reported a target listing window around October 2026, with Goldman Sachs, JPMorgan, and Morgan Stanley in early underwriting talks, though none of this is confirmed in the confidential draft itself - CNBC.
It helps to be precise about what "confidential" buys a company, because this is where a lot of reporting gets sloppy. Under the JOBS Act, an emerging growth company can submit a draft registration statement that the SEC reviews privately, without revealing revenue, margins, risk factors, or the cap table to the public or to competitors. The company then has the option to "flip" the filing public later, and a public S-1 must be on file roughly 15 days before any IPO roadshow begins - Proactive Investors. In plain terms: Anthropic has started the clock and kept the details hidden. It can move fast when it wants to, and it can quietly walk away if the market turns. The reported choice of Wilson Sonsini, the firm that took Google public in 2004, as counsel signals a conventional IPO rather than a direct listing or a reverse merger.
There is a reason the most valuable startups now default to this confidential path rather than a splashy public filing. A confidential draft lets a company complete the entire SEC comment process in private, ironing out accounting questions and risk-factor language without a competitor (or a journalist) parsing every line in real time. Airbnb, Coinbase, and most large technology debuts of the past decade used the same mechanism, then flipped the filing public only when the deal was nearly ready to price. For Anthropic the confidentiality does double duty: it shields the gross-margin and compute-commitment disclosures that would draw the most skeptical scrutiny, and it keeps OpenAI guessing about timing. The cost is that the public, including future retail buyers, learns the hard numbers last and fastest, compressed into a roughly two-week window before the roadshow begins.
The psychology of the reversal is worth dwelling on, because it tells you how the company itself reads the moment. A management team does not publicly say it has "no immediate plans" to go public and then file six months later unless the cost of waiting suddenly rose. Nothing about investor appetite changed in that window; Anthropic could clearly have raised even more privately, and did. What changed was the strategic value of being able to move. Filing now converts an open-ended question (should we ever go public?) into a closed tactical one (do we pull the trigger this quarter or next?), and that optionality is itself worth a great deal when your chief rival is sprinting toward the same finish line.
At the center of every one of these decisions sits one person, the CEO whose public denial in December became a filing in June.
The man in that photo is the one whose statements, leaked memos, and shifting public forecasts run through the rest of this story, which is why understanding his incentives matters as much as understanding the balance sheet.
The first-principles read here is that the filing is an option, not a verdict. Anthropic is not telling you it will be public by Halloween. It is telling you it wants the door open and the paperwork done so that the decision becomes purely tactical: list when the comps are high and the rival is distracted, hold when they are not. Everything that follows in this guide is really about the conditions that determine when, and whether, that door swings open. For broader context on how the largest labs are pulling capital and influence toward themselves, our analysis of AI market power consolidation maps the same gravitational pull from a different angle.
2. The valuation ladder: from a $124M seed to $965 billion
The $965 billion number did not appear from nowhere, but the path to it is stranger and more revealing than the clean upward line most charts show. Anthropic was founded in 2021 by Dario Amodei and Daniela Amodei with a group of former OpenAI researchers, and its first announced round was a $124 million Series A in May 2021 led by Skype co-founder Jaan Tallinn, with Eric Schmidt and Dustin Moskovitz among the backers, at roughly a $623 million post-money valuation - Anthropic. For a company now flirting with a trillion dollars, the starting point was modest, and the second round carries an asterisk almost no clean recap mentions.
That Series B in April 2022 was about $580 million, and roughly $500 million of it came from FTX and Sam Bankman-Fried - Wikipedia. When FTX collapsed, that stake became part of the bankruptcy estate and was later liquidated, transferring a large block of Anthropic shares to other buyers at fire-sale-era prices. It is one of the great ironies of the AI boom that a meaningful slice of the safety-first lab traces back to crypto's most infamous fraud, and that the buyers of those liquidated shares may turn out to be among the biggest winners of any eventual IPO. The Series C of $450 million in May 2023, led by Spark Capital with participation from Google and Salesforce, put the company near a $4 billion valuation and marked the start of the hyperscaler era.
The hyperscaler era introduced a quirk that public investors should understand before they read a cap table. Both Google and Amazon deliberately took capped, non-controlling stakes with no board seat and no voting rights, a structure designed to keep regulators at bay (the FTC and the UK's competition authority both scrutinized these tie-ups) and to preserve Anthropic's nominal independence. The practical effect is that the company's two largest financial backers wield almost no formal governance power, which is unusual for investments of this size and which becomes pivotal later when we examine who can actually overrule the company's mission trust. The strategic money bought compute relationships and balance-sheet upside, not control, and that distinction is one of the most consequential design choices in the entire structure.
The FTX chapter also left a lasting fingerprint on price discovery that is rarely mentioned. When the bankruptcy estate liquidated its Anthropic shares across 2024, it did so through negotiated secondary sales to sovereign and institutional buyers, and those transactions became some of the first real external marks on the company's worth at a time when it was still officially valued in the single-digit billions. Secondary markets in private shares have shadowed Anthropic ever since, and they consistently price below the primary-round headline, for the simple reason that a secondary buyer gets none of the preferences, information rights, or strategic perks that a lead investor negotiates. Keep that wedge in mind: the number on the press release is the most optimistic point on a distribution of prices, not the consensus.
After that, the public sequence does something unusual: it jumps from Series C straight to Series E, with no widely announced Series D. The reason is that the intervening capital arrived not as a named priced venture round but as strategic infrastructure money from Google and Amazon. Google put in roughly $300 million in early 2023 for about 10 percent, then far more. Court filings surfaced in March 2025 revealed Google held about 14 percent of Anthropic with over $3 billion invested, capped near 15 percent, with no voting rights and no board seat - TechCrunch. Amazon, meanwhile, committed up to $4 billion starting in September 2023 and then doubled it to a total of $8 billion by November 2024, taking a stake capped below 33 percent while becoming the primary cloud and training partner. Our earlier coverage of Amazon's multibillion-dollar Anthropic investment captured the moment that relationship became the spine of the company's compute strategy.
From there the climb turns vertical. The Series E in March 2025 raised $3.5 billion at a $61.5 billion post-money valuation, led by Lightspeed - Anthropic. The Series F in September 2025 brought in $13 billion at a $183 billion valuation, co-led by ICONIQ, Fidelity, and Lightspeed, and crucially pulled in sovereign and institutional heavyweights including BlackRock, Blackstone, GIC, Goldman Sachs Alternatives, and the Qatar Investment Authority - Anthropic. Then the Series G in February 2026 added $30 billion at a $380 billion valuation, led by GIC and Coatue, with MGX (the Abu Dhabi vehicle) among the co-leads, which at the time was the second-largest venture deal in history. Three and a half months later, the Series H roughly tripled that to $965 billion. The company announced that final private round itself, with its own graphic, days before filing to go public.
The speed of that re-rating, from $380 billion in February to $965 billion in May, is itself a data point. Valuations that move this fast are driven less by fundamentals (which cannot triple in fourteen weeks) than by a scarcity premium: a small number of investors competing for access to one of the only frontier labs that will sell them equity, ahead of a public listing that would dilute that scarcity.
The most under-reported fact about this ladder is what it hides about realizable value. In April 2026, Anthropic completed an employee tender offer that priced shares at a roughly $350 billion pre-money valuation, and many employees declined to sell, betting on a bigger payday later - Bloomberg. Sit with that. The price at which actual humans could convert paper into cash was $350 billion, while the primary round just weeks later carried a $380 billion mark and the next one hit $965 billion. Primary-round post-money valuations are set by the price of the last marginal share sold to a strategic investor with liquidation preferences and information rights. They are not the price at which the whole company could be sold, and they are certainly not the price an employee can get today. A headline valuation is a marketing number as much as a financial one, and the gap between the tender price and the round price is the closest thing we have to an honest signal.
The second hidden truth is that the headline cumulative fundraising figure is double-counted. The Series H included roughly $15 billion of previously committed hyperscaler money rather than entirely new dollars, so adding every round plus every Amazon and Google commitment overstates the real cash that has entered the company. And there is a reflexive loop worth naming: Amazon has said its original $8 billion is now worth over $70 billion on paper, and a meaningful share of both Amazon's and Google's reported AI-related gains comes from revaluing their Anthropic stakes rather than from operating profit. The investors who fund Anthropic book the gains from Anthropic's rising valuation, which makes the next round easier to raise. For a wider frame on how this capital is sloshing across the sector, our AI investment analysis tracks the same dynamic across regions and stages.
3. Reading the books before the books are public: revenue, margins, and burn
The confidential S-1 means the public cannot yet see audited financials, so this section relies on what Anthropic has disclosed and what credible reporters have surfaced. The headline growth is genuinely staggering. Run-rate revenue went from about $1 billion at the start of 2025 to about $9 billion by year-end, then to $14 billion in February 2026, $30 billion in April, and $47 billion crossed in early May - Anthropic. That is roughly 80x growth in three years, a curve almost no enterprise software company has ever produced. Anthropic's own April disclosure to CNBC framed it as "crazy 80x growth" - VentureBeat.
The first insider distinction to internalize is the difference between run-rate and actual revenue. A $47 billion run-rate means the single most recent month, annualized by multiplying by twelve. It is not booked annual revenue. Full-year 2025 revenue landed closer to $4.5 billion (up from $381 million in 2024), and the projected Q2 2026 quarter is around $10.9 billion. When a curve is bending this steeply, the run-rate sits far above trailing revenue, and conflating the two (as much coverage does) inflates the apparent scale by a wide margin. The number is real and the growth is real, but the $47 billion is a forward-looking annualization, not money that has hit the bank.
This matters for valuation because the multiple you pay looks wildly different depending on which number you anchor to. At roughly $4.5 billion of trailing 2025 revenue, a $965 billion valuation is more than 200 times sales, a figure that has no precedent among durable public companies. At the $47 billion run-rate it is about 20 times, expensive but no longer absurd by AI-era standards. At the projected $70 billion of 2028 revenue it is under 14 times, which would be a bargain if the growth and margins materialize. The entire bull-bear argument, stripped to its core, is a fight over which of these denominators is the honest one, and that fight cannot be settled until the audited S-1 shows trailing numbers rather than annualized snapshots.
The revenue mix is the part that makes Anthropic structurally different from OpenAI, and it is the part a public listing would put front and center. Roughly 70 to 75 percent of revenue is token-based API and first-party usage, only about 10 to 15 percent is consumer and team subscriptions, and the remainder is reserved-capacity enterprise commitments - Sacra. Put differently, about 80 percent of revenue comes from more than 300,000 business customers, the near-inverse of OpenAI's consumer-heavy mix. Anthropic is, financially, a developer and enterprise infrastructure company that happens to have a chatbot, not a consumer phenomenon that happens to have an API. That distinction shapes everything from churn dynamics to how a public-market analyst would model it.
Inside that mix, one young product carries an outsized share of the story. Claude Code, the company's agentic coding tool, scaled from about $400 million annualized in July 2025 to about $1 billion in November to roughly $2.5 billion by February 2026, with enterprise accounting for more than half - Sacra. A single product, barely a year into general availability, is generating a Claude-Code-sized business on its own. We break down its economics in our Claude Code pricing guide, but the strategic point is concentration: a huge share of the growth narrative rests on developers choosing Claude to write software, and that is both the moat and the risk.
Now the number that actually matters for a public valuation, and the one the growth headlines bury: gross margin. In late 2025 Anthropic cut its projected 2025 gross margin to about 40 percent, down from an earlier 50 percent, after inference costs ran about 23 percent higher than expected - Investing.com. In 2024 the gross margin was a startling negative 94 percent. The reason is structural and it is the single most important thing to understand about every frontier lab: inference is a true variable cost of goods, not the near-zero marginal cost of traditional software. Every Claude response burns GPU or TPU time rented from Amazon and Google. A SaaS company sells the same code a million times at almost no incremental cost; an AI lab pays real money for every token it generates. That is why Anthropic's margins look like a compute-intensive infrastructure business (think a cloud provider) rather than software's 80-percent-plus. Independent analysis from Epoch AI confirms compute is the dominant expense line across AI companies.
To make the margin concrete, picture a single heavy coding session. A developer using Claude Opus 4.8 might feed in hundreds of thousands of input tokens (the codebase, the surrounding context, the back-and-forth) and pull out tens of thousands of output tokens. Anthropic collects real revenue on that session, but it also pays Amazon or Google for the GPU and TPU seconds the model burned to produce it, and that cost rises with longer context windows and heavier reasoning. A traditional software company books almost pure profit on its millionth user; an AI lab pays its cloud bill on the millionth query just as it did on the first. This is why scale alone does not fix the margin the way it does in software, and why the bull case hinges on driving cost-per-token down through custom silicon and efficiency rather than on volume.
The contrast with a normal software business is stark enough to reframe how you value the company. A healthy SaaS firm runs 75 to 85 percent gross margins because its marginal cost of serving one more customer rounds to zero, which is why the market happily pays it 10 to 20 times revenue. Anthropic, at a reported 40 percent gross margin in 2025, behaves financially more like a cloud-infrastructure provider or even a manufacturer with real cost of goods sold. If that 40 percent is the steady state, then a software-style revenue multiple is unjustified and the stock should trade closer to infrastructure comps. If the margin genuinely marches to 70-plus percent as the deck claims, then the software multiple is earned. The valuation, in other words, is a leveraged bet on one line item: gross margin. Everything else is secondary.
The forward picture Anthropic has shown investors is aggressive. Internal projections reported by The Information point to up to $70 billion in revenue and $17 billion in cash flow by 2028, with gross margin climbing toward 77 percent and the company turning cash-flow positive around 2027 - TechCrunch. More provocatively, Anthropic reportedly told investors that Q2 2026 would be its first profitable quarter, at roughly $10.9 billion in revenue and a $559 million operating profit, a thin 5 percent margin - Yahoo Finance. If true, that would make Anthropic the first frontier lab to reach operating profit, a powerful pre-IPO talking point.
The hype filter has to come on here, because this is exactly the kind of claim that gets repeated uncritically. The "first profitable quarter" figure is unaudited, pre-IPO, and contested. Critics note that a temporarily discounted compute deal (reportedly with SpaceX) landed in precisely the May to June window used to show profit, that enterprise prepayments and discounted token commitments can pull revenue forward while deferring delivery, and that Anthropic's costs have historically scaled roughly linearly with revenue, which argues against durable margin expansion - Where's Your Ed At. The honest position is that the projections are reported, not verified, and that the 50-to-40-percent margin cut already shows these forecasts get revised down. The 77-percent margin endgame assumes aggressive, unproven inference-cost deflation through custom silicon, batching, and distillation. It might happen. It is not a fact. The deeper structural reason this is hard to predict is the subject of our piece on how LLM inference is eating software, which argues the cost of the "big pipe" is the central economic variable of this entire era.
4. The compute bill behind the IPO: why a private company needs public markets
The cleanest way to understand why a company that just raised $65 billion privately would also file to go public is to look at the compute bill, because no venture round, however large, can carry it indefinitely. Anthropic's reported 2025 cash burn was around $2.8 billion, and 2026 estimates for combined training and inference run as high as $14 billion - Cybernews. On top of that sit enormous multi-year commitments: a reported $21 billion purchase of Alphabet TPUs and over $80 billion in cloud commitments stretching to 2029. These obligations dwarf even the $47 billion run-rate, and most of them are off the balance sheet as future contractual cash, which means a reported quarterly "operating profit" can coexist with staggering future liabilities.
The supplier relationships are where the timing of the IPO really gets explained. Amazon's Project Rainier, an Indiana data center costing roughly $11 billion, went live in October 2025 with about 500,000 Trainium2 chips, and a April 2026 expansion pushed the partnership toward up to 5 gigawatts and more than $100 billion over ten years - Anthropic. Google committed tens of billions for up to 1 million TPUs and more than a gigawatt of capacity in 2026, later expanded with Broadcom toward 5 gigawatts from 2027 - Anthropic. In November 2025, Anthropic added a $30 billion Azure commitment plus up to a gigawatt of Nvidia Grace Blackwell and Vera Rubin systems, alongside investments of up to $5 billion from Microsoft and up to $10 billion from Nvidia - Microsoft. A separate multi-year deal with CoreWeave followed in April 2026.
This is the structure critics call circular financing, and naming it plainly is more useful than dismissing it. A strategic backer invests equity in Anthropic; Anthropic uses that capital (and more) to commit to multi-year compute purchases from that same backer; the backer books those purchases as revenue and marks up its equity stake as Anthropic's valuation rises. Nvidia, notably, backs both Anthropic and OpenAI, and CoreWeave is itself heavily Nvidia-financed. Across the sector, analysts have identified more than $800 billion in such arrangements - J.P. Morgan Asset Management. The loop is not inherently fraudulent (the compute is real and it gets used), but it does make headline demand harder to interpret, because supply, investment, and revenue are entangled among the same handful of balance sheets.
There is a historical rhyme worth holding in mind. During the late-1990s telecom build-out, equipment makers like Lucent and Nortel lent their own customers the money to buy their gear, which inflated reported demand right up until those customers could not pay and the vendors' revenue evaporated. The AI version is more sophisticated and the underlying demand is far more real, but the accounting question is identical: how much of today's booked revenue and marked-up equity is genuinely independent, and how much is the same dollars circulating among a few giants? The related worry is depreciation. If Nvidia's chips remain useful for the four-to-six years companies assume, today's reported margins are roughly honest; if they obsolesce in two-to-three years as skeptics argue, then the true cost of compute is understated across the entire industry, Anthropic included, and reported profits are flattered by a bookkeeping choice rather than earned in cash.
The physical dimension of all this is easy to forget behind the dollar figures. Project Rainier is not a metaphor; it is a sprawling campus in Indiana drawing power measured in gigawatts, the output of a small fleet of power plants, and the Google and Microsoft commitments imply similar build-outs elsewhere. Frontier AI has quietly become an energy and real-estate business as much as a software one, and the constraint that increasingly governs how fast a lab can grow is not talent or even chips but electricity and the time it takes to build substations and data halls. This is part of why the compute commitments stretch to 2029 and beyond: you cannot conjure a gigawatt of capacity in a quarter. A public investor buying Anthropic is, in a real sense, buying exposure to an industrial supply chain that runs through the power grid.
Now zoom out to the macro, because the funding-wall logic is the first-principles answer to "why go public at all." Western hyperscalers are committing on the order of $600 to $725 billion to capex in 2026, up sharply year over year, with roughly 75 percent AI-tied - Futurum Group. That build-out is increasingly debt-funded: AI-related firms tapped debt markets for at least $200 billion in 2025, and Morgan Stanley expects $250 to $300 billion of hyperscaler debt issuance in 2026, with banks already offloading data-center loans to private credit as their own capacity strains - Insurance Journal. When the debt leg of financing a build-out is straining bank balance sheets, the equity leg has to do more work. An IPO is the equity leg. Anthropic is not going public because it ran out of private investors. It is going public because the scale of compute it must buy to stay at the frontier exceeds what private venture and debt can comfortably carry, and public equity is the deepest pool of capital on earth.
The practical takeaway for anyone trying to value this company is that the compute commitments are the real liabilities, and they are larger and longer-dated than the revenue is high. The path to the 77-percent margins and 2027 breakeven runs directly through Anthropic's ability to drive per-token compute costs down faster than usage scales up, which is why the company is buying custom Trainium and TPU silicon rather than only renting Nvidia GPUs. If that cost-deflation thesis holds, the off-balance-sheet commitments become a moat. If it does not, they become the thing that keeps the company burning cash well past the dates in the investor deck. A public S-1 would force these commitments into a risk-factors section, which is one reason the confidential route is so appealing: it lets Anthropic control exactly when that disclosure becomes public.
5. The governance puzzle Wall Street has never seen: PBC, the Trust, and Class T
If the compute bill explains the timing, the governance structure explains why this listing is genuinely unprecedented, and why most coverage gets it backwards. Anthropic is a Delaware Public Benefit Corporation, which means its directors are legally required to balance three things: the financial interests of stockholders, the interests of those affected by the company's conduct, and the chartered public benefit, defined as the "responsible development and maintenance of advanced AI for the long-term benefit of humanity" - Harvard Law School Forum on Corporate Governance. Read that again: a PBC board has a legal license to not maximize shareholder value. For most public investors, who are trained to expect shareholder primacy, that is unfamiliar territory.
The PBC form itself, though, is not actually the obstacle, and this is where a lot of commentary is simply wrong. Multiple public benefit corporations have already gone public, including Lemonade, Vital Farms, Allbirds, Warby Parker, and Coursera, and a 2020 Delaware amendment lowered the vote to convert to or from PBC status from two-thirds to a simple majority, which actually spurred PBC IPOs - National Law Review. PBC directors are arguably better insulated from shareholder suits than ordinary directors, and bringing a derivative claim for breach of PBC duties requires holding at least 2 percent of shares - Cooley GO. So the benefit-corporation wrapper is well-trodden. What is novel is the layer on top of it.
The reason the governance design matters at all is written in recent history. In November 2023, OpenAI's nonprofit board abruptly fired Sam Altman, only to reinstate him days later after employees and investors revolted, an episode that exposed how a mission-control structure can collide violently with commercial reality and nearly vaporize tens of billions in value over a weekend. Anthropic's design tries to solve the same tension more gracefully by vesting control in a standing trust rather than an ad hoc nonprofit board, and by phasing that control in gradually rather than handing it overnight to a few people who can act on impulse. Whether that is genuinely more stable, or simply a slower-motion version of the same conflict, is exactly the question a public listing puts to the test. Public shareholders would be buying into a structure explicitly engineered to be able to tell them no.
That layer is the Long-Term Benefit Trust (LTBT): an independent body of five financially-disinterested trustees who hold a special class of stock, Class T, whose sole power is to elect and remove a phased-in portion of Anthropic's board - Anthropic. The Class T shares are non-economic: they carry no dividend and no claim on assets. Their only function is control. The board-election power phased in over time and fundraising milestones, from one of five directors, to two-fifths in July 2024, to three-fifths by November 2024, designed to reach a board majority within four years - AI Lab Watch. On April 14, 2026, that majority became real: when Novartis CEO Vas Narasimhan joined the seven-member board, Trust-appointed directors crossed into actual majority control - Anthropic.
Here is the framing almost every outlet gets wrong: this is not a founder-entrenchment device. The dual-class playbook that public investors know (Google, Meta, Snap) hands founders super-voting shares so they can never be overruled. Anthropic's structure does close to the opposite: the founders ceded long-term board control to an independent trust of safety-oriented outsiders. The trustees include figures like Richard Fontaine of the Center for a New American Security and Mariano-Florentino Cuellar of the Carnegie Endowment, not Amodei loyalists. The genuinely interesting question for a public listing is not "how do founders keep control" but "how does a trust of non-shareholders keep control of a public company's board, and will the market tolerate it."
The honest answer involves a failsafe that critics say guts the whole thing, and the precise math is almost never reported. A stockholder supermajority can amend the Class T provisions and strip the Trust's power without trustee consent, but only at a high and rising bar: 75 percent of total voting power before November 2025, escalating to 85 percent after, plus 75 percent board approval - AI Lab Watch. Skeptics on the EA Forum argue this makes the Trust "quite subordinate to stockholders" and note that Anthropic has never published the full Trust Agreement. The counterpoint, rarely placed beside the criticism, is that Amazon's and Google's shares are non-voting and cannot be counted toward any such supermajority, which sharply weakens the "big investors will just vote it out" thesis. An IPO scatters new voting common stock into public hands, and exactly how that reshapes the 85-percent math is the open question no filing has yet answered.
There is one more piece that makes the governance more than ceremonial, and its timing is conspicuous. The Responsible Scaling Policy (RSP) v3.0, effective February 24, 2026, wires the Trust directly into operational safety: when a deployment relies on marginal-risk reasoning, it requires explicit approval from both the Board and the LTBT, not just the CEO and the Responsible Scaling Officer - Anthropic. That makes the Trust a deployment gatekeeper, a far more consequential power than "elects some directors" suggests. But v3.0 also quietly dropped the prior unconditional pause commitment, replacing it with Risk Reports and an explicitly non-binding Frontier Safety Roadmap - GovAI. Loosening the hardest safety commitment roughly three months before filing to go public is the kind of juxtaposition that barely gets connected in mainstream coverage. For public markets it cuts both ways: it removes a clause investors would have hated, while inviting the question of whether any safety commitment can bind under quarterly-earnings pressure. No precedent exists for a trust of safety experts controlling a majority of a public company's board, and that, not the PBC wrapper, is the real frontier here.
The composition of the Trust reinforces how unusual this is. The trustees are policy and security figures, people like the head of a national-security think tank and the president of a major peace-and-diplomacy endowment, chosen precisely because they are financially disinterested and bring an outside-the-company perspective on long-run risk. That is the opposite of a typical board, which is stacked with investors and operators whose incentives align tightly with the share price. The deliberate insertion of disinterested outsiders into the control structure is what gives the design its credibility with safety advocates and what makes it so alien to the proxy-advisory machinery (ISS, Glass Lewis) that public-market investors rely on to evaluate governance. There is simply no rating framework for "a purpose trust controls the board," which means early price discovery on the stock will have to invent one in real time.
The disclosure question sits underneath all of this and rarely surfaces in coverage. Anthropic has published a summary of how the Trust works but not the full Trust Agreement, so the precise removal mechanics, the exact triggers, and the fine print of the override provisions remain unseen by outsiders. Critics treat that opacity as the crux: a structure is only as strong as the clauses you cannot read. A public S-1 would, in principle, force far more of this into the open, because risk-factor disclosure obligations are stringent and a structure this consequential cannot be hand-waved past institutional investors and their lawyers. One of the most telling things to watch when the public filing arrives is simply how much of the Trust Agreement Anthropic is finally willing to show, because that will reveal whether the mission control is a load-bearing wall or decorative trim.
6. The race that set the clock: Anthropic versus OpenAI
To see why Anthropic filed when it did, put the two rivals' calendars side by side, because the choreography is unmistakable once you line up the dates. OpenAI confidentially filed its own S-1 on May 22, 2026, targeting a fourth-quarter listing at a valuation between roughly $852 billion and $1 trillion, with Goldman Sachs and Morgan Stanley leading - CNBC. Anthropic then closed its $65 billion round on May 28 and filed its own confidential S-1 on June 1, four days later. That is not coincidence. It is a footrace, where being public first, or at a higher valuation, shapes the recruiting pitch, the partnership leverage, and the capital-raising narrative for years.
OpenAI's ability to file at all is itself an under-appreciated enabler, and it explains part of Anthropic's structural head start. OpenAI spent late 2025 completing a wrenching for-profit restructuring into OpenAI Group PBC controlled by the OpenAI Foundation nonprofit, a deal in which Microsoft took a roughly 27 percent stake valued near $135 billion, OpenAI committed to $250 billion in incremental Azure spend, and Microsoft's IP rights were extended to 2032 - Fortune. Without converting and resolving the nonprofit-control overhang, an OpenAI IPO was legally impossible. We tracked the contours of that fight as it unfolded, including the battle over OpenAI's for-profit plans and the AGI-clause negotiations with Microsoft. Anthropic, already a PBC since birth, never carried that blocker.
The two were close enough that even share liquidity moved in parallel. OpenAI ran its own employee liquidity events, including a tender offer letting staff sell roughly $1.5 billion of shares, which we covered when OpenAI opened its tender to SoftBank. Both labs use these events to keep employees from leaving for a payday, and both now dangle the bigger prize of an IPO. The April 2026 revamp of the Microsoft and OpenAI deal, which freed OpenAI to serve products on any cloud and capped its revenue share, further cleared the runway. The result is two of the most valuable private companies in history racing toward the same public window in the same quarter.
The divergence the headlines flatten is unit economics, and it is the real differentiator a public S-1 would expose. OpenAI reportedly hit about $25 billion in annualized revenue but lost roughly $1.22 for every $1 of revenue in early 2026, with internal projections pushing breakeven toward 2030 - CNBC. Anthropic, with its enterprise-heavy mix and reported approach to operating profit, is structurally the cleaner company to take public, even though both face the same compute gravity. The competitive intensity also shows up directly in the products: OpenAI's current flagship is GPT-5.5, detailed in our GPT-5.5 guide, and the leapfrogging between the two labs is now monthly. The strategic point is that the IPO clock is set by the rivalry and the compute bill, not by a market begging for the paper, which is precisely why filing first matters.
The prize in winning the race is not only valuation; it is talent and trust, and Anthropic has quietly led on both. The company reportedly retains close to 80 percent of its technical staff, losing only a couple of researchers even to Meta's reported nine-figure compensation offers, a remarkable number in a market where elite AI talent is the scarcest input of all. Being first to list, or listing at the higher mark, compounds that advantage: liquid public stock is a powerful recruiting and retention tool, and the prestige of the larger debut shapes which researchers, which enterprise customers, and which governments choose to partner with you. The footrace is really a contest over who gets to be the default frontier lab for the next decade, and the IPO is one move in that longer game.
It is worth being precise about Anthropic's structural head start, because it is easy to miss. OpenAI had to spend the better part of a year unwinding the nonprofit cap that made an IPO legally awkward, negotiating Microsoft's stake down to a clean percentage, and converting into a public benefit corporation controlled by its foundation. Anthropic was born a PBC and never carried that overhang, which means it could turn the question of going public into a simple matter of timing while its rival was still resolving existential structural questions. In a race where months matter, starting with clean legal plumbing is an underrated advantage, and it is one reason the company that was long seen as the more cautious of the two could end up reaching the public markets on equal or better footing.
7. Bubble or build-out: the honest case for and against $965 billion
Any guide that pretends to insider knowledge owes you a genuinely two-sided answer on the question everyone is actually asking: is $965 billion real, or is this a bubble? The intellectually honest position is that both the bull and bear cases are strong, and the truth depends on variables nobody can yet observe. Start with the cleanest public comp. CoreWeave, the AI cloud provider, IPO'd in March 2025 at $40 a share, ran to nearly $187, and a year later traded around $120, up roughly 175 percent from its IPO but about 35 percent off its high, on a contracted backlog near $66 billion and heavy debt - The Motley Fool. That volatility is the template: enormous demand, real revenue, real debt, and a stock that swings violently on sentiment about whether the demand is durable.
Set against its private peers, Anthropic's mark is in a class of its own, and the spread shows how concentrated the market's frontier bets have become. xAI raised $20 billion at a $230 billion valuation in early 2026, Safe Superintelligence reached $32 billion with essentially no revenue, and Mistral sits near $14 billion, while Anthropic and OpenAI tower above the field.
The gap between the top two and everyone else is the real signal: investors are not spreading bets across a dozen contenders, they are concentrating capital into the two labs with the deepest revenue and the most compute, which is also what makes a public listing for either one such a consequential bellwether for the whole category.
The bear case rests on three pillars, and none of them is frivolous. First, valuations are stretched: Palantir has traded near 87 times sales and Nvidia's price-to-sales ratio has sat above 30, the kind of multiples that historically precede sharp corrections - The Motley Fool. Second, enterprise ROI is thin: an MIT study found that roughly 95 percent of enterprise generative-AI pilots delivered no measurable financial return, blamed on integration and learning gaps rather than model quality - Fortune. Third, the circular financing described earlier means demand may be partly an accounting loop. Investor Michael Burry has gone further, calling GPU depreciation accounting "Fugazi" and arguing the 4-to-6-year useful-life assumptions overstate margins because the chips actually obsolesce in 2-to-3 years.
The analogy everyone reaches for is the dot-com fiber glut, and it is worth testing rather than repeating. In the late 1990s, telecoms laid vast amounts of fiber on the expectation of internet demand that took another decade to fully materialize, and the overbuild bankrupted many of them even though the internet ultimately did need all that capacity and more. The AI parallel is the data-center and GPU build-out on the expectation of AI demand. The crucial difference is timing: the fiber sat dark for years, earning nothing, whereas AI compute is being consumed the moment it is switched on, at prices customers pay today. That argues the AI build-out is less speculative than the fiber glut. The crucial similarity is financing: both were funded heavily with debt and vendor credit, which means a demand air-pocket could still trigger a painful deleveraging even if the long-run demand is entirely real. A bubble can pop on timing and leverage alone, without the underlying thesis being wrong.
The bull case is equally grounded, and it starts from a structural observation rather than a vibe. Jensen Huang frames the moment as "the largest infrastructure build-out in human history" - Fortune, and the demand data, however concentrated, is not imaginary. The first-principles version of the bull case is this: intelligence is becoming a metered utility, and unlike the dot-com fiber glut, the capacity being built is being consumed in real time at prices customers are voluntarily paying. The bifurcation in the MIT finding is the key nuance: even as broad enterprise pilots fail, revenue is concentrating in a few labs and a few killer applications (coding chief among them). That is not what a uniform bubble looks like; it is what an early-but-real platform shift looks like when value accrues unevenly.
So where does that leave a $965 billion private mark heading into public markets? The pressure-tested answer is that the valuation is defensible only on the forward case, not the trailing one. At roughly $4.5 billion of actual 2025 revenue, $965 billion is an absurd multiple; at the projected $70 billion of 2028 revenue with healthy margins, it is merely very expensive. The bet embedded in the price is that Anthropic compounds at something close to its recent pace for several more years and drives compute costs down enough to reach software-like margins. If a public AI-stock correction hits while the IPO window is open, that window can slam shut for both labs and force another private mega-round instead. The deeper structural read, which we develop in our agent economy analysis, is that the value will ultimately accrue to whoever turns cheap intelligence into delivered outcomes, and the open question is how much of that margin the labs capture versus the companies built on top of them. A reasonable investor can look at the same facts and reach opposite conclusions, which is exactly why this listing is so contested.
It helps to translate the debate into rough scenarios rather than a single verdict. In a bull case, Anthropic compounds toward the projected $70 billion of 2028 revenue, margins climb toward software-like levels as custom silicon bites, and the stock grows into and beyond a near-trillion-dollar valuation the way Nvidia grew into its once-stretched multiple. In a base case, growth stays strong but margins improve more slowly, profitability arrives later than the deck promises, and the shares trade like a volatile, expensive infrastructure name, closer to the CoreWeave pattern of big gains punctuated by brutal drawdowns. In a bear case, an enterprise-demand air-pocket or a single model-quality stumble collides with the heavy compute commitments, the IPO window slams shut, and the company is forced back to private markets at a flat or lower mark. None of these is fanciful, and the spread between them is enormous, which is the honest reason to treat any precise price target with suspicion.
The frame that cuts through the noise is the one we keep returning to in our coverage of the sector: value accrues to whoever turns cheap intelligence into delivered outcomes. If that is the labs, because frontier capability stays scarce and defensible, then a near-trillion-dollar Anthropic is reasonable. If intelligence commoditizes faster than expected, with open models and rivals compressing prices, then much of the value migrates up the stack to the application and workflow companies, and the labs become high-revenue, low-margin utilities rather than monopolist toll-collectors. The history of technology offers examples of both outcomes (Intel captured the PC era's value; the telecoms did not capture the internet's). Which pattern AI follows is unknowable today, and it is the deepest uncertainty buried inside the $965 billion price tag.
8. The parts the headlines skip: Gulf money, regulation, and the jobs reversal
This is the section the big outlets mostly will not write, not because the facts are secret but because they are inconvenient to the clean narrative. Begin with the money. In July 2025, WIRED reported a leaked internal Slack memo in which Dario Amodei told staff Anthropic would pursue investment from Gulf states, money he estimated at "easily $100 billion or more," explicitly calling the principle that "no bad person should ever benefit from our success" impractical and conceding the move could "enrich dictators" - Futurism, citing WIRED. It was a striking reversal of an earlier moral stance, framed internally as a competitive necessity. The under-reported part is that the prediction came true: by 2026, Abu Dhabi's MGX (the vehicle linked to G42 and Mubadala) co-led the Series G, and Qatar's QIA had already joined the Series F. The sovereign-wealth money Amodei once ruled out now stands to profit directly from the IPO, and coverage rarely connects the MGX deal back to the "enrich dictators" admission.
The strategic logic behind the Gulf money is as important as the ethics. The Abu Dhabi vehicle MGX is tied to G42 and Mubadala and sits within an ecosystem chaired by Sheikh Tahnoon bin Zayed, the UAE's national security adviser, which is precisely why the earlier safety-minded objections existed: this is sovereign capital with geopolitical aims, not passive index money. For a frontier lab, though, Gulf sovereign wealth offers something American venture cannot match at this scale: tens of billions of patient, price-insensitive dollars willing to fund a decade-long build-out, plus access to cheap energy and land. The same gravitational pull that drew Anthropic to Abu Dhabi is drawing every frontier lab toward sovereign capital and toward the power these states can supply, which is quietly reshaping where AI capability physically sits and who has leverage over it.
The regulatory posture is the second skipped story, and it is more strategic than it first appears. Anthropic endorsed California's SB 53, the first enforceable US frontier-AI transparency law, in September 2025 - Anthropic, and it publicly opposed a proposed ten-year federal moratorium on state AI laws, which the Senate ultimately stripped 99 to 1. On the surface this reads as a safety-first company welcoming regulation. The first-principles read is subtler: a frontier lab that already follows rigorous internal safety processes benefits when those processes become legally mandated, because compliance is a fixed cost that falls hardest on smaller, faster-moving competitors. Endorsing transparency rules you already meet, while preferring a single federal standard over a patchwork, is not only principled; it is a moat. The geopolitics of who funds and who governs frontier AI is the subject of our piece on how war is reshaping AI access and independence, which situates the Gulf-money question in a wider contest.
The third skipped story is the jobs reversal, and it is a near-perfect case study in pre-IPO narrative management. In May 2025, Amodei warned that AI could eliminate up to half of entry-level white-collar jobs within one to five years, potentially driving unemployment to 10-to-20 percent, an argument rooted in his essay "Machines of Loving Grace." Ahead of the IPO, he and other lab leaders notably softened that message, reframing it through the lens of the Jevons Paradox (cheaper intelligence creates more demand for work, not less). Fortune explicitly tied the reversal to the trillion-dollar listings on the horizon - Fortune. It is hard to sell shares in a future you are simultaneously describing as a mass-unemployment event. Independent research, including from the Yale Budget Lab, has so far found no significant AI-attributable labor shift, and Anthropic's own Economic Index shows augmentation outpacing automation on Claude.ai even as API usage trends toward heavy automation. The doom-to-optimism pivot, arriving exactly as the company files to go public, is the kind of detail that tells you more about IPO mechanics than any prospectus will.
A fair reading does not treat any of this as scandal. Taking sovereign money is standard for capital-intensive frontier work, shaping regulation to your advantage is what every large company does, and revising public messaging is rational. The point of surfacing these threads is that they are the real texture of how a mission-driven lab behaves when nearly a trillion dollars is on the line, and they complicate the tidy story that the safety-first lab is simply being rewarded for its virtue. The truth is more human and more interesting: Anthropic is making the same trade-offs every ambitious company makes, just at a scale and with a moral framing that invites more scrutiny.
9. What carries the valuation: Claude, Claude Code, and the moat
Strip away the financial engineering and the governance novelty, and a public investor is ultimately buying one thing: the belief that Claude stays at or near the frontier and that developers and enterprises keep choosing it. The current lineup is led by Claude Opus 4.8, the flagship launched in late May 2026, priced at $5 per million input tokens and $25 per million output tokens, alongside the balanced Claude Sonnet 4.6 and the fast Claude Haiku 4.5, with a more powerful model, Claude Mythos, in preview - Claude API docs. Subscriptions run from Claude Pro at $20 a month to Max at $100 and $200. We benchmark the flagship in detail in our Claude Opus 4.8 guide, and the broader competitive picture in our AI model benchmarks and pricing roundup.
The strongest pillar of the moat is coding, and the numbers are striking. Claude Code passed a $2.5 billion run-rate and reportedly accounts for more than half of enterprise spending on Anthropic, while one analysis from Menlo gives Anthropic roughly 42 percent of the code-generation market versus OpenAI's 21 percent - Yahoo Finance. Anthropic has become the default model for serious software work, and that position compounds: developers build workflows around Claude, enterprises standardize on it, and the switching cost rises. The second pillar is the Model Context Protocol (MCP), the open standard Anthropic introduced for connecting models to tools and data, which we covered when Anthropic launched MCP. MCP is ecosystem leverage: by setting the connective standard, Anthropic shapes how the whole industry wires models into real systems.
The concentration that is the moat is also the risk, and a candid S-1 would have to say so. If more than half of enterprise revenue rides on a single young product in a category OpenAI, Google, and a swarm of startups are all attacking, then a single model generation where a competitor leaps ahead could dent the growth story fast. There is an even sharper wrinkle: one of Anthropic's largest API customers is reportedly Cursor, an AI coding startup valued in the tens of billions and a direct competitor in the coding tools layer. Part of Anthropic's vaunted coding lead is therefore revenue from a rival that could, in principle, shift its model mix. The full ecosystem, from models to agents to the developer tools built on them, is mapped in our Anthropic ecosystem guide, and the rise of autonomous, multi-step agents built on these models is the subject of our Claude managed agents guide.
The cadence of the model lineup is itself part of the moat, and it is accelerating. Beyond the generally available models, Anthropic has been scaling Claude Mythos, part of a defensive-cybersecurity effort, to a growing roster of organizations across many countries, signaling that the next capability tier is close behind the current flagship. A lab that ships a meaningfully better model every few months forces competitors and customers alike to keep re-benchmarking and re-integrating, and that velocity is hard to sustain without exactly the kind of capital an IPO provides. Speed of iteration, not any single benchmark win, is the compounding advantage, because it keeps the company at the front of mind every time a buyer evaluates the market.
The enterprise footprint underneath the headline products is the quieter half of the moat. More than 300,000 business customers now run on Anthropic, and the company's revenue is concentrated in paying organizations rather than free consumers, which tends to produce stickier, higher-quality revenue than a viral consumer app. Enterprises that have standardized on Claude for coding, document analysis, and customer-facing agents incur real switching costs to move, and once MCP wires Claude into a company's internal tools and data, that integration becomes part of the furniture. The risk, as the Cursor example shows, is that some of the biggest customers are themselves fast-moving startups that could shift their model mix overnight, so the durability of enterprise revenue (the net retention, the contract lengths, the concentration) is among the most important things a public S-1 would finally reveal.
This is also where the model layer meets the layer above it, and where the value of an IPO ultimately gets decided. Frontier models like Opus 4.8 are inputs; the businesses that win are the ones that turn those inputs into delivered outcomes. A growing class of operators now wraps these models into autonomous companies, where AI agents run a business's tool stack while humans set the goals. Platforms such as O-mega sit at that layer, letting a single conversation stand up a website, an app, billing, and content operated by AI rather than assembled by hand, one example of how cheap frontier intelligence gets converted into actual work. Yuma Heymans (@yumahey), who founded O-mega and co-founded the autonomous recruiting platform HeroHunt.ai, has argued that the durable value lands with the operators who apply models to real workflows rather than with any single model vendor, a useful lens for reading what an Anthropic listing does and does not capture. The labs win the input layer; whether they also capture the outcome layer is the trillion-dollar question hiding inside the valuation.
10. What an Anthropic IPO would actually mean
Step back from the mechanics and the meaning becomes clearer than any single data point. An Anthropic IPO would be the moment the frontier AI lab as a public company stops being theoretical. For employees holding paper that the April tender priced at a $350 billion valuation while declining to sell, it would be the conversion event they bet on. For the early backers, including the buyers of FTX's liquidated stake and the sovereign funds that joined late, it would be a generational return. For the public investor, it would be the first chance to own a frontier lab directly, rather than through the proxy of an Nvidia, a Microsoft, or an Amazon whose AI exposure is one line in a vast conglomerate.
The harder meaning is governance precedent. If Anthropic lists while keeping the Class T structure and the Trust-controlled board intact, it will be the first time a public company is governed by a body explicitly empowered to put a chartered public benefit on equal footing with shareholder returns, with a safety policy that can gate product deployment. That is either a landmark in stakeholder capitalism or an experiment that quarterly-earnings pressure slowly erodes, and we will find out in real time. The alternative, which some analysts expect Anthropic may be forced toward, is converting to a conventional Google-style dual-class structure to reassure the market, which would quietly abandon the most distinctive thing about the company. Watching which path the public S-1 takes will tell you how much the mission survives contact with the capital markets.
How should a non-insider actually think about all this? A simple framework holds up. First, separate the option from the event: the June 1 filing buys flexibility, and the real signal will be the public S-1 and the price range, not the confidential draft. Second, watch the margin line, not the run-rate: a credible path from 40-percent toward software-like gross margins is the entire bull case, and any slippage there (as already happened once) matters more than another run-rate milestone. Third, read the governance disclosures closely, because how the S-1 characterizes the Trust, the Class T override math, and the RSP's deployment gate will reveal whether the structure is durable or decorative. Fourth, treat the compute commitments as the real liabilities they are, since the off-balance-sheet obligations to Amazon, Google, Microsoft, and Nvidia are larger and longer-dated than the headline profit.
It is also worth holding the scale of the moment in view. If Anthropic, OpenAI, and SpaceX all reach the public markets within months of each other, they could collectively bring close to $3 trillion of new market capitalization to public investors in a single window, a concentration of frontier exposure with no real precedent. That has a reflexive quality: a successful Anthropic debut would validate OpenAI's, and a stumble by either could sour sentiment for both and for the AI-infrastructure names already trading. The IPO is therefore not only a corporate event but a market-structure event, one that will pull index funds, pension plans, and retail portfolios into direct frontier-AI exposure whether or not they sought it. The decision the filing really forces is not only Anthropic's; it is every public investor's, about how much of this they want to own and at what price.
There is a longer arc here too, and it is the reason this listing matters beyond the trading desk. For three years, the frontier labs have operated as black boxes, their economics visible only through leaks and selective disclosures. A public Anthropic, bound by quarterly reporting and audited statements, would be the first time the world gets a continuous, verified view into the actual unit economics of frontier AI: the real gross margin, the real compute bill, the real customer concentration, the real cost of staying ahead. That transparency could puncture hype or confirm it, but either way it ends the era of guessing. In that sense the most consequential thing an Anthropic IPO produces may not be the capital it raises, but the numbers it is finally forced to show, which will reset how the entire industry is understood and valued.
The first-principles conclusion is that Anthropic going public is less a victory lap than a financing necessity dressed as a milestone, set in motion by a race with OpenAI and a compute bill that private capital can no longer carry alone. That does not make it a bad investment; some of history's best companies went public out of exactly this kind of need. It does mean the tidy story (safety-first lab earns its trillion-dollar reward) is the least interesting and least accurate way to understand it. The accurate version is a company making aggressive, contested, deeply consequential trade-offs across money, governance, safety, and narrative, all at once, in full view of a market that has never seen a company quite like it. Whether the $965 billion holds will be decided not by the filing, but by whether Claude stays at the frontier, whether the margins bend the way the deck promises, and whether a trust of safety experts can really steer a public company. Those are the things to watch. Everything else is noise.
This guide reflects the state of Anthropic's path to the public markets as of June 2, 2026. Valuations, filings, financial figures, and AI model versions change quickly, and many figures here are reported or projected rather than audited. Verify current details before making any decision based on them.