Anthropic's $2T IPO: 12x revenue, $518B bet

Anthropic just pulled back the curtain on its IPO filing, and the numbers are staggering. The company behind Claude is targeting a $2 trillion valuation — which would make it one of the largest public debuts in history — while revealing that it lost $42 billion in 2025 alone. The prospectus, obtained by Reuters and reported across multiple outlets on September 28-29, 2026 (UTC), lays bare the extreme economics of running a frontier AI lab.

The numbers behind the hype

For fiscal year 2025 (ending December 31, 2025), Anthropic posted revenue of $4.59 billion, up roughly 12-fold from $386 million in 2024. That kind of growth is virtually unheard of for a company at this scale. But the cost side is even more eye-popping: operating losses hit $8.06 billion, and the net loss swelled to approximately $42 billion once non-cash charges related to prior financing rounds are included.

The real jaw-dropper is what comes next. Anthropic has committed to $518 billion in future compute and infrastructure spending — more than 100 times its 2025 revenue. That figure covers cloud contracts with Amazon and Google, plus deals with SpaceX and smaller providers to lock in GPU capacity for years to come.

Source: Anthropic S-1 filing as reported by Reuters, 36Kr, and Securities Times on September 28-29, 2026 (UTC).

The growth has only accelerated in 2026. Second-quarter revenue reached $11.5 billion, a 14-fold year-over-year increase, and the annualized revenue run-rate (ARR) climbed past $65 billion by the end of July. Anthropic is projecting revenue of $190-200 billion by 2028.

The Amazon dependency problem

Buried in the risk factors — which consume roughly one-third of the filing — is a structural paradox. Approximately 45% of Anthropic's revenue comes from Amazon, which is simultaneously its largest cloud compute provider and its biggest customer. Nearly one-quarter of total revenue comes from just two clients: Amazon and Google.

This creates a circular revenue pattern. Anthropic pays Amazon billions for GPU time on AWS Bedrock. Amazon then resells Claude access to its own enterprise customers, and those API calls count as Anthropic revenue. PitchBook analysts estimate that on a net, ex-related-party basis, Anthropic's "real" revenue could be closer to $25-35 billion rather than the headline $65 billion ARR, which would put the valuation multiple at 28-39x revenue instead of the optically lower figure.

Compounding the risk: many of Anthropic's largest customers have not signed long-term contracts, meaning they could reduce or halt spending at any time.

Why it matters

A $2 trillion Anthropic IPO would be more than a financial event — it would set the ceiling for how the market values AI infrastructure companies. SpaceX's $1.78 trillion IPO in June 2026 currently holds the record, and Anthropic is explicitly targeting above that. If the listing succeeds, it validates the thesis that frontier AI labs can become trillion-dollar enterprises on the back of enterprise adoption. If it stumbles, every private AI valuation gets marked down.

The $518 billion compute commitment is the number that should make investors pause. That is more than the entire annual revenue of Microsoft. It means Anthropic has effectively pre-committed to spending over a hundred times its 2025 top line just to keep training and serving models. If revenue growth decelerates — if enterprise AI spending hits a wall, if open-source models catch up, if a recession hits — those fixed costs become an anchor. The filing itself acknowledges that the company may never achieve sustained profitability.

The Amazon dependency deserves scrutiny too. A company where nearly half your revenue comes from the same entity that controls your compute supply is not fully independent. If Amazon decides to prioritize its own models (Titan, Olympus) on Bedrock, or renegotiates the compute contract on less favorable terms, Anthropic's economics shift overnight. The S-1 flags this explicitly, but investors tend to underprice concentration risk until it materializes.

There is also the question of whether the $2 trillion target is realistic. At that valuation, Anthropic would be worth more than NVIDIA was in early 2025. The bull case rests on Claude Code's dominance in AI coding, enterprise adoption of Claude Cowork, and the $30 trillion total addressable market Anthropic cites. The bear case: 2025 operating margins were deeply negative, customer concentration is extreme, and the company is asking public markets to underwrite a half-trillion-dollar infrastructure bet at a moment when data center construction is facing headwinds worldwide.

What to watch next

The IPO is expected to price in November 2026, with the S-1 likely to be publicly filed in the first two weeks of October. The key indicators to track: whether the final valuation comes in below $2 trillion (a signal of weak demand), how much of the $100 billion+ raise is primary capital versus secondary shares, and whether Amazon's revenue share declines in the quarters before listing.

If Anthropic prices at $2 trillion or above, expect OpenAI to accelerate its own IPO timeline. If it prices at $1.5 trillion or lower, the entire private AI funding complex faces a reckoning. Either way, this filing is the most detailed financial window into frontier AI economics the public has ever seen — and the picture is equal parts exhilarating and terrifying.