Anthropic Prospectus Reveals Massive Losses, Explosive Growth, and AI Extinction Risk
TL;DR
- Anthropic's newly released prospectus shows staggering scale: around $9 billion in 2025 revenue, up more than 10x in two years, paired with net losses in the tens of billions as it burns cash on chips and data centers.
- The filing sets the stage for a blockbuster potential IPO, with Anthropic targeting a valuation well above $300 billion backed by Amazon, Google, and Microsoft, despite never having turned a profit.
- In a striking risk disclosure, Anthropic warns investors that its own frontier AI models could pose an existential risk to humanity if misaligned, misused, or autonomous beyond control.
Burning Billions To Chase Trillions
Anthropic has finally opened its books, and the picture is both breathtaking and brutal.
According to its new prospectus released ahead of a potential public listing, the maker of Claude generated roughly $9 billion in revenue in 2025, up from about $1 billion in 2024 and just $100 million in 2023. For the first half of 2026 alone, the company says it already topped $7 billion in revenue, putting it on an annualized run-rate approaching $15 billion to $17 billion.
That is hypergrowth by any Silicon Valley standard, faster on a percentage basis than Google, Facebook, or even OpenAI at a similar stage. The driver is enterprise: Claude Code and Claude for business now account for the vast majority of sales, with API usage and $200-per-month power-user subscriptions exploding as coding agents went mainstream.
But growth is coming at an eye-watering cost. The prospectus shows a net loss of around $14 billion to $16 billion for 2025, after losing about $5.7 billion in 2024. In other words, Anthropic is currently spending roughly $2.50 to $3.00 for every $1.00 it brings in.
Where is all the money going? In one word: compute.
Anatomy Of A $15 Billion Loss
Like OpenAI, Anthropic is locked in a full-scale arms race for Nvidia GPUs, custom silicon, talent, and gigawatts of data center capacity. The filing breaks out more than $20 billion in 2025 operating expenses, dominated by cloud infrastructure costs paid largely to Amazon Web Services and Google Cloud — two of its biggest investors and partners.
Research and development alone topped $10 billion, reflecting a hiring spree for elite AI researchers commanding $5 million to $10 million packages. Sales and marketing also surged as Anthropic built a global enterprise sales force from scratch to compete with Microsoft-backed OpenAI and Google Gemini.
Crucially, Anthropic says it has secured tens of billions in committed compute and capital, including more than $20 billion raised from Amazon, Google, and venture firms like Lightspeed, Menlo Ventures, and Sequoia. It ended 2025 with over $20 billion in cash and marketable securities on its balance sheet, but warns it will need to raise substantially more debt and equity to fund its next generation of models, dubbed Claude 5 and beyond, and its Project Prometheus data center buildout.
Chief Financial Officer Krishna Rao frames the losses as intentional: burn now to capture what Anthropic believes will be a multi-trillion-dollar market for AI agents and knowledge work.
The Road To A Monster IPO
The prospectus does not set a date, price, or ticker, confirming it is a confidential-turned-public S-1 intended to test Wall Street appetite. But bankers say all signs point to a 2027 debut that could be the largest tech IPO since Alibaba.
Sources cited in the filing coverage suggest Anthropic is eyeing a valuation of $350 billion to $400 billion or more, nearly double its $183 billion private valuation from its $13 billion Series G in September 2025. That would place it ahead of Oracle and just behind Meta in market cap on day one.
The timing is no accident. With annualized revenue now in the eight figures per employee and net dollar retention above 140%, Anthropic can pitch itself as the efficient, enterprise-safe alternative to OpenAI, which is also losing tens of billions per year but spending even more aggressively on consumer hardware and Stargate infrastructure.
Still, the filing is blunt: Anthropic has never been profitable, expects losses to widen in the near term, and may never achieve profitability. It lists fierce competition from OpenAI, Google DeepMind, Meta, xAI, and Chinese labs like DeepSeek as a core risk, alongside U.S. export controls, copyright lawsuits from publishers and music labels, and dependence on just two cloud providers.
Our Own Models Could Kill Us
Buried deep in the 200-plus pages of risk factors is the disclosure that has stunned even jaded Silicon Valley lawyers.
Under a heading on safety and alignment, Anthropic warns that increasingly capable AI systems could pose an existential risk to humanity, including threats from loss of control, cyberattacks on critical infrastructure, and development of chemical, biological, radiological, and nuclear weapons.
The company states plainly that its stated mission — to build safe superintelligence — may conflict with investor returns, and that there is no guarantee its safety techniques, including Constitutional AI and responsible scaling policies, will work. It even cautions that a catastrophic misuse or autonomous action by Claude or a future model could destroy its business, invite draconian regulation, and cause widespread harm or death.
It is extraordinarily rare for a company to tell potential shareholders: our core product could end civilization. But for Anthropic, co-founded by ex-OpenAI safety researchers Dario and Daniela Amodei, the warning is both a legal shield and a philosophical statement. The Amodeis retain super-voting shares through a long-term benefit trust designed to let them overrule investors on safety decisions, even after going public.
What Happens Next
For now, Wall Street seems willing to overlook the red ink. Private shares traded on secondary markets jumped after the prospectus leaked, and analysts argue that 200%+ annual growth justifies a money furnace — if Anthropic can convert enterprise pilots into durable $100 million-plus contracts and bring inference costs down.
The key metrics to watch in the next amended filing: gross margins on API calls, which are improving as custom chips come online, cash burn per quarter, and whether Claude Code can hold off OpenAI Codex and Google Jules.
One thing is clear from the prospectus: Anthropic no longer wants to be seen as just an AI lab. It wants to be seen as the next AWS — the infrastructure layer for the agent economy. Whether investors will pay cloud-like multiples while absorbing existential risk is about to become the biggest gamble in tech history.
Get All The Latest Updates Delivered Straight To Your Inbox For Free!