Anthropic's IPO Filing: $16.3B Revenue, $80B Loss and 92 Pages of Risk

Anthropic made $16.3 billion in the first half of 2026 and lost $80 billion in the same period. That single pair of numbers, disclosed in the company's S-1 stock market filing, captures the paradox at the heart of what could become one of the largest public offerings in tech history — a five-year-old firm that some investors value between $1.5 and $2 trillion, burning cash at a pace no company has attempted before.

The filing, submitted without a stated target valuation, pulls back the curtain on the maker of Claude. And it arrives at a moment when public investors are being asked to underwrite a new species of company: capital-intensive, research-driven, safety-obsessed and politically entangled. Here is what the document actually says.

How much money is Anthropic making — and losing?

A lot of both. Anthropic brought in $16.3 billion in revenue in the first half of 2026, up from $4.585 billion for all of 2025, per the filing. The company lost roughly $33.9 billion in 2025 and $80 billion in the first half of 2026 — though much of that half-year figure came from a $72.4 billion one-off charge.

Founded in 2021, the company has posted explosive growth every year of its existence. Computing power accounts for the vast majority of its costs. Anthropic spent nearly $9 billion on AI infrastructure in the first half of this year, up from $2.3 billion in the same period last year — a near-quadrupling that shows how fast the arms race for compute is escalating.

Unlike OpenAI, Anthropic has made clear that its customers are largely other businesses. "Frontier AI is most valuable in real-world settings, as deployment across enterprises is where economic impact is greatest," the company's executives wrote in the filing.

There's a catch. The company said its top 5 customers account for 25% of its business. That concentration makes the revenue engine fragile — lose a few anchor accounts, and the growth story wobbles.

What does Anthropic think AI will become?

In the filing's most striking passage, Anthropic's executives describe a future they call "a country of geniuses in a datacenter" — in their words, "millions of virtual agents with the ability to revolutionize every area of human endeavor: software, physics, math, biology, medicine, finance, economics, commerce, robotics, military and intelligence capabilities."

That last word — military — is notable. Anthropic is currently in a standoff with the Trump administration over its refusal to loosen restrictions on how the Department of Defense could use Claude. Listing military capability as an opportunity, while refusing to expand the Pentagon's access, shows the tightrope the company is walking between commercial ambition and its own stated principles.

The filing highlights four categories of opportunity:

• Acceleration of progress in biology and medicine
• Economic growth
• Spreading that growth globally
• Streamlining government services, such as social services and medical programming

Some of the claims are concrete. Anthropic said it is already working with several pharmaceutical and biotechnology companies and has improved the clinical trials process, "cutting the time it takes to write clinical trial reports from months to minutes." The executives also claim Claude has helped make biological discoveries, including a "molecular machine" with the "potential to be developed into a novel biotechnology tool." One purported use case: gene editing.

What does the filing say about safety failures?

The most consequential section may be the company's own admission that its guardrails might not hold. Anthropic's most capable models, its executives wrote, are capable of providing "meaningful assistance to malicious actors seeking to develop chemical, biological, or other dangerous weapons."

The company deploys technical safeguards, but with a caveat stated plainly in the filing: those 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."

That's a rare public acknowledgment from a frontier lab that its own protections can be beaten.

The filing also discloses what Anthropic believes is the first documented case of a large-scale cyber-attack with little human intervention. The company says 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 breaks the risks AI poses to society into four categories: misuse by bad actors, loss of control, authoritarian abuse and economic disruption. On authoritarian abuse, the company said AI may be used by authoritarian governments to "deepen repression" and develop "high-tech surveillance states." It offered less of a technical defense against those uses than against others, saying instead it is committed to working with "free societies to provide a counterweight to authoritarian abuse and to promote freedom and democracy."

That position has a political cost. The company had a falling out with the Trump administration after saying it would not change its restrictions to allow its models to be used to build domestic surveillance or autonomous lethal weapons.

Why are there 92 pages of risk factors?

The S-1 devotes 92 pages to risks facing the business, a document that reads as a checklist of everything that could sink a frontier AI company.

The company acknowledges that much of its success is tied to whether its top executives stay. It must outcompete Google and OpenAI, navigate a tumultuous regulatory landscape around AI and datacenters, and survive any public relations fallout if Claude is used to do something it wasn't supposed to — like help create bioweapons.

The enterprise focus cuts both ways: if any of its top buyers leave, the company could take a huge hit. And Anthropic's structure as a public benefit corporation — a for-profit company that balances financial interests with a stated social good — may make it harder to achieve short- or medium-term financial goals, the company said. The firm must report on its progress toward public benefit goals the same way it reports financials.

Then there is the tail risk. The filing flags liability tied to any AI-enabled catastrophic event — the doomsday scenario Anthropic's own employees have warned about. CEO Dario Amodei has called for a pause on AI development to ensure firms develop it responsibly and avoid such a catastrophe. And insurance may not cover the fallout. "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," the filing reads.

The physical footprint is a risk too. Anthropic's AI infrastructure is concentrated in the US, leaving it exposed to regulatory changes, power grid constraints, natural disasters and local opposition to datacenters. The filing specifically calls out data center moratoriums that have been passed or are being considered. "We are expanding our data center capacity internationally as part of our efforts to diversify our infrastructure footprint and reduce concentration risks," the filing states.

People are the final vulnerability. The company, which says talent is a big part of its value proposition, worries that US immigration policy — specifically the Trump administration's targeting of H-1B visas — could affect its ability to recruit and retain highly qualified employees from the global tech and engineering pool.

What happens next?

The filing now goes to investors, who must price a company with triple-digit growth, eight-figure customer concentration, self-described incomplete safeguards and a business model dependent on compute no one has fully built yet. How the market values that trade-off — between $1.5 and $2 trillion, if current private estimates hold — will set the benchmark for every AI lab lining up behind it.