Anthropic Files for IPO at a Valuation of Around $965 Billion
Anthropic, the developer of the Claude language models, has confidentially filed documents with the SEC to go public. Its valuation has reached nearly one trillion dollars after closing its Series H round at $650 billion.
IPO of the Year No One Saw Coming: Why Anthropic’s $965 Billion Valuation Is the Biggest Bubble in History
You’ve seen the headlines. Anthropic confidentially filed for an IPO at a valuation of around $965 billion. This came just four days after the company closed its Series H round at $65 billion. It has now officially overtaken OpenAI ($852 billion) and become the world’s most valuable private AI company.
I’ve been tracking the AI market since GPT-2 and have seen plenty of wild valuations. But what’s happening now exceeds even my most optimistic expectations for irrationality. In five years Anthropic has grown from zero to a valuation higher than Tesla’s and Meta’s combined market caps. It has 134 million users versus OpenAI’s 900 million, yet it earns more. Its annual run-rate revenue stands at $47 billion—three times what it was in February 2026.
Let’s be honest. What we’re witnessing is not the triumph of a brilliant business. It is the largest bet in history that the B2B AI market will remain supply-constrained forever. And that bet carries three fatal flaws no one in the press is talking about. I’ll break them down.
[The Core]: What’s Really Happening
Anthropic isn’t just going public. It is committing hara-kiri in front of the world, and no one notices. The key point: a confidential S-1 is filed when a company does not want to reveal its real numbers. It is a legal way to hide gross margin, cost structure, and, most importantly, cash-burn rate from competitors.
Why does this matter? Because PitchBook already dropped the key line you won’t see in headlines: “the number that will decide everything is the gross margin that no one outside Anthropic has ever seen.” If that margin turns out below 60–65 % (and I suspect it is closer to 40–50 %), the entire “trillion-dollar valuation” story collapses on the first day of trading.
The inside detail that changes everything: look at their revenue structure of $16.2 per user per month—seven times higher than OpenAI’s ($2.2). That figure is only possible if almost all users are large corporations with multi-million-dollar annual contracts. This is not a diversified customer base. It is a bet on 500–1,000 mega-clients, each of which could defect to a competitor or build its own solution at any moment.
The scariest part for investors: to generate $16.2 per user, Anthropic must spend enormous sums on compute for those same corporate clients. Its newly released Claude Opus 4.8 model requires 5–10 times more compute per query than GPT-4o. The higher the revenue per user, the higher the cost per user. No one knows whether revenue covers those costs. That is exactly what is hidden inside the undisclosed gross margin.
Timeline and Context
Let’s look at the timeline to understand how insane the last month has been for Anthropic.
February 2026: Anthropic closes Series G at a $183 billion valuation. At the time it felt like science fiction. Annual revenue run-rate: $14 billion.
March 2026: OpenAI closes a round at $122 billion with a $852 billion valuation. The market thinks OpenAI has won the race.
May 28, 2026: Anthropic announces a $65 billion Series H at a $965 billion valuation. Investors: Altimeter, Dragoneer, Greenoaks, Sequoia. Amazon participates (another $50 billion on top of prior investments), along with… Micron, Samsung, and SK Hynix. Yes, memory manufacturers suddenly became AI investors. The signal: they see Anthropic burning chips like crazy and want a piece of the fire.
June 1, 2026: Anthropic files a confidential S-1 with the SEC. This happened five days after the Series H round. No tech company of this scale has ever moved from private funding to IPO filing this fast. Normally six to twelve months pass between the last round and an S-1.
Why the rush? There are two explanations, both alarming.
Optimistic version (for Anthropic): they want to go public before the market sees OpenAI’s real numbers. Their S-1 will remain confidential for another two to three months. They are riding peak AI FOMO and want to raise at the highest possible valuation.
Realistic version: they are running out of money. Run-rate revenue jumped from $14 billion to $47 billion in four months. Delivering that growth required tens of billions in GPUs. The $65 billion Series H is not “we’re rich.” It is “we just burned $40 billion and need more, and we don’t want to borrow at 12 %.”
Winners and Losers
This IPO race is zero-sum: the winner takes all, the loser loses everything. Here’s the scorecard.
Biggest winner: Sequoia Capital and early investors. Sequoia entered at valuations in the tens of billions. Its stake is now worth 50–100 times more. If the IPO happens at $965 billion, they can exit with tens of billions in profit. Caveat: the lock-up period is usually 180 days. A lot can change in six months after an IPO.
Next winner: D.A. Davidson and other underwriters. The fee for an IPO of this size is 2–3 % of the amount raised. If Anthropic sells $50–100 billion in shares, the banks pocket $1–3 billion. One of the largest paydays in investment-banking history.
Biggest loser: retail investors who buy at the top. The same people now buying synthetic Anthropic tokens on Solana (yes, they exist and are up 640 % since late 2025). When the real S-1 opens and everyone sees 40 % gross margin instead of the hoped-for 70 %, the stock will drop 30–40 % on day one.
OpenAI finds itself in a strategically weak position, but not the way it looks. PitchBook is right: Anthropic just “volunteered to take all the disclosure risk first.” OpenAI can now watch how the market reacts to Anthropic’s real numbers and adjust its own story before its own IPO. Classic second-mover advantage in public markets.
Amazon, which has invested a total of $80 billion in Anthropic, becomes a hostage to the situation. It is the largest shareholder. If Anthropic collapses after the IPO, Amazon suffers not only direct losses but a reputational hit: “How could the world’s smartest company put $80 billion into a bubble?”
What the Media Isn’t Saying
Journalists are writing about a “historic IPO” and “the triumph of safe AI.” Here are three things they are omitting because they don’t grasp their significance.
Insight #1: Tokenized shares on Solana are a ticking time bomb.
While everyone discusses the S-1, decentralized exchanges like Jupiter DEX are trading synthetic tokens that give “exposure to Anthropic.” Their price has risen 640 % since late 2025 and implies valuations between $850 billion and $1.7 trillion.
Anthropic has officially stated: “Any sale or transfer of Anthropic shares without explicit board approval is invalid and will not be recognized on the company’s cap table.” It named eight platforms selling unauthorized exposure.
What this means: thousands (if not tens of thousands) of holders of these synthetic tokens will not receive real shares at the IPO. They hold an instrument Anthropic legally refuses to recognize. When that becomes clear, panic will follow. That panic will hit the price of real shares because retail traders do not distinguish between “an Anthropic token” and “a real Anthropic share.”
Insight #2: Micron, Samsung, and SK Hynix’s participation is not a vote of confidence but a hedge against disaster.
Everyone writes: “Look, memory makers are investing in Anthropic, so they believe in the future.” I see something else. These three companies know the HBM memory market is oversupplied. They understand that demand for GPUs to train large models could collapse at any moment—one breakthrough in training efficiency or one recession is all it takes.
Their investment in Anthropic is insurance. If the AI market implodes, they at least own a piece of one of the survivors. It is not optimism. It is risk hedging. Samsung’s investment committee is not looking at Claude and saying “wow, what a great model.” It is looking at HBM consumption forecasts for 2027–2028 and seeing downside risk.
Insight #3: The 2028 breakeven forecast is a trap.
Wall Street analysts write that Anthropic plans to reach profitability in 2028, two years ahead of OpenAI (2030). It is presented as a sign of business strength.
Let’s do the math. Current run-rate revenue is $47 billion. To break even in two years they must either:
- keep revenue flat and cut costs by $47 billion (impossible, because costs are mostly Nvidia GPUs and Nvidia is not lowering prices), or
- grow revenue to $100 billion+ while maintaining current margins (also unlikely, because the enterprise AI market is saturating).
“Profitability in 2028” is a pretty slide for investor presentations. In reality they will remain unprofitable at least until 2030–2032. The only question is how large those losses will be. If gross margin is 40 %, then at $100 billion in revenue they will lose $20–30 billion a year. No public market will tolerate losses of that magnitude.
Forecast: The Next 30 Days and 90 Days
Forget $965 billion. Forget the “new AI era.” Here is what will actually happen in the next three months.
Next 30 days: Wave of lawsuits from synthetic-token holders.
Within a month at least three class-action suits will be filed against the platforms that issued Anthropic tokens (Forge Global, Hiive, Unicorns Exchange). Plaintiffs will demand either recognition of their rights to real shares or compensation for losses. Anthropic will defend itself by saying it warned everyone. The token market will crash 80–90 %. This will create negative sentiment around the IPO itself.
At the same time the SEC will launch a non-public investigation into these platforms. Regulators have long wanted to clean up trading of private securities on crypto rails. The Anthropic case will become the perfect test case.
Next 90 days: SpaceX will outshine Anthropic and change everything.
SpaceX plans an IPO at roughly $1.8 trillion around June 12, 2026. Polymarket puts the probability of closing above $2 trillion on day one at 78 %.
What this means for Anthropic: two to three weeks after SpaceX goes public (assuming success), investors will have a fresh example of what a “good tech IPO” looks like. If SpaceX rises 20–30 % on day one, expectations for Anthropic will skyrocket. If SpaceX falls (unlikely but possible), the IPO window for unprofitable tech companies will close for years.
Anthropic is trapped by its own choice: it filed the S-1 and there is no going back. If it withdraws after a SpaceX failure, it will be seen as admitting weakness. If it goes public in a weak market, its valuation will drop 40–50 % on day one.
What about 12 months from now? Either Anthropic becomes a public company with a market cap of $400–500 billion (because the market will adjust its valuation to realistic multiples) or it withdraws the S-1 and stays private, burning $30–40 billion a year until Series H investors demand an exit.
The second scenario is more likely. The $65 billion Series H gives it a 1.5–2-year runway. It can wait out market turbulence and try again in 2027 or 2028. But that requires revenue to keep growing at least 50 % a year. That becomes harder and harder when you are already at a $47 billion run-rate.
I don’t know whether anyone will ever buy that $4.85 million Mill Valley mansion in exchange for Anthropic stock. But I do know one thing: the person who sells that house for Anthropic stock today will either be a genius in five years or lose everything. And the stakes in this game are higher than any real estate in California. Because we still don’t know the one number that matters—Anthropic’s gross margin. And the fact that the company is working so hard to hide it tells me far more than all of its $47 billion in revenue.
— Editorial Team
No comments yet.