Anthropic's IPO Tests the Cost of Government Conflict

Anthropic says government deals account for less than 1% of its annual revenue, yet its confidential IPO prospectus warns that US government actions could damage relationships with commercial customers and partners. An investor can use that percentage to gauge direct exposure to government purchasing. It says much less about what happens when a business loses access to a model it uses or decides a supplier’s dispute with Washington makes that supplier harder to rely on.

The prospectus has not been made public. Reuters reviewed it; the warning and revenue figure come through The Next Web’s account of that review. Anthropic is warning of possible commercial harm, rather than disclosing a tally of customers who have left. Its recent disputes with Washington show two different routes by which government decisions can reach the rest of its business.

Anthropic may begin marketing an initial public offering during the week of November 9, with trading possible before Thanksgiving, Bloomberg reported, as recounted by International Business Times. The company appears to be targeting a valuation as high as $2 trillion. The dates could move, and the target is not a price investors have agreed to pay. Prospective buyers must evaluate a company whose commercial demand could support enormous infrastructure commitments, provided customers keep using the service.

Two Ways Washington Reaches a Customer

In February, President Donald Trump told federal agencies to stop using Anthropic’s models, and the Pentagon designated the company a supply-chain risk, according to the prospectus reviewed by Reuters. Anthropic challenged the designation, and a judge temporarily blocked it in March. A federal appeals court later declined to block the designation, International Business Times reported. The dispute was still live as Anthropic prepared to pitch investors.

The Pentagon decision followed Anthropic’s refusal to remove restrictions on using its technology for autonomous weapons and mass surveillance, International Business Times reported. Anthropic has argued that the designation damaged its business and reputation. The immediate purchasing question concerns agencies directed away from its products. The commercial question is different: whether businesses considering Anthropic as a supplier regard the government conflict as a reason to hesitate. A customer could agree with Anthropic’s restrictions and still want assurance that its chosen AI service will remain available.

The second episode involved availability rather than a customer’s judgment. US Commerce Department export controls affected two Anthropic models, according to the reported prospectus. Anthropic disabled the affected models for all customers on June 12 and restored access on July 1 after the controls were lifted. Those are the shutdown and restoration dates, not the date the controls were imposed. A business using either model could have its service interrupted without selling anything to the government or changing its opinion of Anthropic.

February’s instruction targeted federal agencies. June’s shutdown reached customers outside them. One route from Washington to Anthropic’s commercial business runs through buyers’ confidence in the company; the other ran through access to the product itself. The reported less-than-1% share describes neither route. The shutdown demonstrates an interruption, while the warning about damaged commercial relationships describes a potential consequence whose financial size remains unclear.

The Cost of Building Ahead

Anthropic’s growth offers a reason investors might accept some of that uncertainty. Reuters’ review of the prospectus, recounted by International Business Times, put 2025 revenue at nearly $4.6 billion, twelve times the previous year’s level. It also put the operating loss at about $8 billion and the net loss at roughly $42 billion. Much of the larger net figure came from accounting charges involving financial instruments that could convert into shares. The operating loss and the future cost of supplying AI services are more useful here than treating every dollar of the net loss as a bill for computing.

Anthropic also disclosed plans for $518 billion in cloud-computing and infrastructure obligations over the coming years. That is a multiyear figure, not a payment due at once. Computing capacity has to be arranged before every future customer purchase is known. If usage and renewals grow, that capacity can support more sales. If a government action interrupts a model or customers become reluctant to depend on it, the revenue case for capacity arranged in advance gets harder to make. How flexible those obligations are depends on terms that the reported headline figure does not specify.

Joseph Montezuma, writing a financial analysis for Seeking Alpha, points to the gap between shorter-duration customer contracts and longer-term infrastructure obligations, alongside customer concentration. That is an assessment of the investment case, not a prediction that contracts will be canceled. Reuters’ prospectus review also described Anthropic’s reliance on a relatively small group of customers and major technology partners, including Amazon and Google, for revenue, computing and other services. A large relationship can matter to both the sales side and the capacity side of this business; the reported government-revenue percentage covers only a narrow slice of the first.

There is a strong counterargument to reading the June interruption as a forecast of lasting damage. Access was restored on July 1 after the controls were lifted, and growing demand could give customers ample reason to stay. A customer might also distinguish a temporary regulatory interruption from Anthropic’s ability to deliver its service over time. The prospectus warning deserves attention because access was actually interrupted, while any lasting effect on commercial sales remains uncertain.

SpaceX offers a narrow IPO comparison. Its filing warned that weaker ties with US agencies could cost it contracts, The Next Web noted. Anthropic’s reported warning also involves Washington, but extends to business customers and partners beyond government work. The similarity is exposure to government decisions; the relevant difference is where each reported warning locates the possible loss. SpaceX’s contract warning supplies no basis for assuming that the companies have comparable government dependence, customers or infrastructure costs.

For Anthropic, the next useful disclosures would connect these pieces: how much commercial demand depends on the affected models, how customer agreements renew, and what room the company has to adjust planned computing capacity. Government deals account for less than 1% of reported annual revenue; in June, a government action interrupted models for all customers, while Anthropic was planning $518 billion in obligations to supply computing over future years. That combination, rather than the government sales figure alone, is the risk an IPO buyer would have to price.