Anthropic spent the last two days getting both halves of the squeeze.

On Tuesday, Bloomberg reported the company is in talks to raise a new round at a valuation north of $900 billion. That would put it ahead of OpenAI, which is currently being benchmarked at $852 billion. CNBC and PYMNTS confirmed the figure on Wednesday. Sources told Bloomberg the round may be Anthropic's last private raise before an IPO that bankers have been pencilling in for October.

Twenty-four hours later, the Wall Street Journal reported the Trump White House had told Anthropic it disagrees with the company's plan to expand controlled access to Mythos, the cyberattack-capable model the Pentagon classified as a supply chain risk in March. Anthropic wanted to extend the program from a handful of cleared partners to roughly 120 organisations. The White House said no.

The reason given to Anthropic, per the Journal, was that the company does not have enough compute to serve more Mythos users without degrading service for the federal civilian agencies that are already on it. Sherwood News, TipRanks, The Next Web, Crypto Briefing and Yahoo Finance all confirmed the same reporting between Wednesday night and Thursday morning AEST.

That is the squeeze. The $900 billion valuation is being marketed on Mythos and Claude Security as the two products that justify a price above OpenAI. The White House just told Anthropic it cannot ship Mythos to the customers it would need to support that price.

The Compute Wall

Running Mythos at scale is reportedly part of why Anthropic needs the cash. CNBC cited infrastructure as a primary driver of the raise. The company has been pulling capacity from cloud partners on contracts that look more like utility supply agreements than software licenses.

The federal civilian carveout we covered in cycle 179 is now the bottleneck Anthropic cannot grow past. Treasury, Commerce, DHS, Justice, State and a handful of other agencies have priority on Mythos compute. The White House is, in effect, telling Anthropic that until it can prove additional capacity, no commercial expansion of the program is acceptable.

This is the first time the federal government has functionally rate-limited a private AI company's product roadmap. It will not be the last.

The IPO Math Just Got Harder

Bloomberg's reporting on the October IPO target is now sitting in a different room than it was 48 hours ago. An S-1 filed in the next 90 days has to disclose, in plain English, that the company's flagship enterprise product is gated by Trump administration approval that has been explicitly withheld.

That is not a footnote risk factor. It is the centerpiece of the offering document. Pentagon classified contracts went to eight competitors last Friday, per our cycle 181 coverage. The Mythos civilian program is now stalled. Claude Security is shipping into a market where Microsoft Defender and CrowdStrike already own the budget.

Some bankers will read this as a reason to push the IPO into 2027. Others will argue the opposite: get the offering out the door while the $900 billion mark is still fresh, before any S-1 disclosure can compress it. Goldman, Morgan Stanley and JP Morgan are reportedly all in the bake-off. Watch which one wins. Aggressive timing signals confidence the disclosure will not break the price. Conservative timing signals it will.

Three Things to Watch

First, Dario Amodei's public response. Anthropic has stayed remarkably disciplined through the Pentagon ban, the Mythos carveout fights and the safety-clause standoff. A frustrated public statement here from Amodei or Daniela Amodei would tell us the company has decided the political cost of silence is now higher than the cost of pushback.

Second, the secondary market price. Forge Global was printing Anthropic at $1 trillion two weeks ago. The first trade after Wednesday's reporting tells us whether the safety-brand premium survives a White House product veto, or whether it just took the first real haircut of the cycle.

Third, the executive order. The same Journal report noted the White House is simultaneously exploring an executive order on AI cyber capabilities. If that text lands and codifies the Mythos restrictions into formal federal policy, the $900 billion valuation conversation effectively ends until October at the earliest.

Anthropic spent six weeks getting the federal door reopened after the March supply chain designation. They got it open just enough to onboard a handful of agencies. Now the door is being held there by the same administration that opened it.

The capital markets will figure out what that means in real time. Probably this week.