Anthropic details nearly $9 billion in AI infrastructure spending in its IPO filing

Anthropic flags U.S. data center moratoria among concentration risks

Anthropic said it spent nearly $9 billion on AI infrastructure in the first half of 2026, up from $2.3 billion in the same period a year earlier, as the company released its 92-page S-1 on Monday to seek a U.S. stock-market listing. The filing reported $16.3 billion in revenue and an $80 billion loss for the period, much of the loss attributable to a $72.4 billion one-off charge.

The S-1 said Anthropic’s computing infrastructure remained concentrated in the United States, making it extremely vulnerable to regulatory changes, power-grid constraints, natural disasters and opposition to AI or data centers. The filing specifically identified data-center moratoria that had been passed or were being considered among Anthropic’s concerns.

The prospectus ranged from predictions about AI’s “incredible potential” to existential risks and the company’s expensive growth, The Guardian said. Filed with U.S. financial regulators, the document provides the public’s first look at audited Anthropic finances. Founded in 2021, the five-year-old company reported explosive growth every year of its existence. Some investors had valued it between $1.5 trillion and $2 trillion, but its S-1 did not list a target valuation.

Anthropic reported $4.585 billion in revenue and a loss of roughly $33.9 billion for 2025. The filing said securing computing power accounted for the vast majority of the company’s costs, while nearly $9 billion in first-half infrastructure spending represented a sharp increase from a year earlier.

Unlike OpenAI, Anthropic has said its customers are largely other businesses. Company executives wrote in the filing: “Frontier AI is most valuable in real-world settings, as deployment across enterprises is where economic impact is greatest.” Anthropic said its five largest customers accounted for 25% of its business. The Guardian described that concentration as risky and said the loss of a top customer could deal the company a major blow.

Anthropic said AI could eventually reach what its executives called “a country of geniuses in a datacenter,” comprising millions of virtual agents. The company said those agents could affect software, physics, math, biology, medicine, finance, economics, commerce, robotics, and military and intelligence capabilities.

The filing identified four opportunity categories: faster progress in biology and medicine; economic growth; spreading that growth globally; and streamlining government services such as social services and medical programming.

Anthropic said it was working with several pharmaceutical and biotechnology companies and had improved the clinical-trial process, “cutting the time it takes to write clinical trial reports from months to minutes.” Executives also claimed Anthropic had made biological discoveries with Claude, including a “molecular machine” with “potential to be developed into a novel biotechnology tool.” The company presented gene editing as one purported use.

The S-1 discussed risks in two contexts: dangers AI may pose to society and human beings, and ways Anthropic’s business could fail. Anthropic divided the societal risks into misuse by bad actors, loss of control, authoritarian abuse and economic disruption. It acknowledged that its most capable models can provide “meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons.”

Anthropic said it deploys technical safeguards but warned that “such safeguards are incomplete, may be outdated in light of evolving threat models, have been and may in the future again be circumvented, or may fail to operate as intended.”

The company also disclosed what it believes was the first documented case of a large-scale cyberattack carried out with little human intervention. Anthropic said a Chinese state-sponsored hacking group manipulated a Claude code tool “into attempting infiltration into roughly 30 global targets, including large technology companies, financial institutions, chemical manufacturing companies, and government agencies.”

Anthropic said AI may be used by authoritarian governments to “deepen repression” and develop “high-tech surveillance states.” It committed to working with “free societies to provide a counterweight to authoritarian abuse and to promote freedom and democracy.” The Guardian said Anthropic presented little detail about a defense or potential safeguard against those uses and appeared to suggest that its work with what it described with scant detail as “free societies” might offset the AI-enabled repression of authoritarian regimes.

The company nevertheless mentioned military capabilities while remaining in a standoff with the Trump administration over restrictions on the Department of Defense’s use of Claude. The dispute followed Anthropic’s refusal to change its restrictions to allow its models to be used to build domestic surveillance or autonomous lethal weapons.

Anthropic also detailed risks to its business. The filing said the company’s prospects depend partly on retaining top executives, competing with Google and OpenAI, navigating the regulatory landscape affecting AI and data centers, and avoiding public-relations fallout if its models are used for purposes it says they should not be, including creating bioweapons. The Guardian said investors were taking stock of a company new to the public market in a highly competitive space.

As a public benefit corporation — a for-profit structure that balances financial interests with a stated social goal — Anthropic must report progress toward its public-benefit goals alongside its finances. The company said the structure could make short- or medium-term financial goals harder to achieve and that its progress on those goals must be reported in the same way as its financial results.

Anthropic also addressed the risk of an AI-enabled catastrophic event, the kind of scenario its employees have warned about. Chief executive Dario Amodei has called for a pause in AI development to ensure firms develop the technology responsibly and avoid such a catastrophe. The Guardian described the potential public harm and fallout from a scenario in which Claude went rogue, while Anthropic said it was concerned about substantial liability, especially as insurance options for AI-related risks are still evolving. The filing warned: “We may be unable to obtain coverage for certain AI-related liabilities on commercially reasonable terms, or at all, leaving us exposed to potentially significant uninsured losses.”

Anthropic said it was diversifying its concentrated infrastructure footprint through international expansion. The company wrote: “We are expanding our data center capacity internationally as part of our efforts to diversify our infrastructure footprint and reduce concentration risks.”

The company also described its workforce as central to its value proposition. Anthropic said U.S. immigration policies could impede its ability to recruit and retain highly qualified employees. It noted that the Trump administration had targeted H-1B visas, a tool technology firms rely on to attract and hire from the global pool of technology and engineering talent.