Anthropic targets a $2 trillion IPO with $518B in non-cancellable compute bonds

Anthropic targets a $2 trillion IPO with $518B in non-cancellable compute obligations

🔎 San Francisco, October 14: the biggest stock market filing in AI history enters orbit

On October 14, Anthropic gathered its prospective investors at its San Francisco headquarters. The message left no room for ambiguity: the roadshow is underway. According to Yahoo Finance, the IPO could be wrapped up before Thanksgiving, on November 26, with an expected valuation between $1.8 and $2 trillion.

The figure turns heads. It would place Dario Amodei's company above the record set by SpaceX ($1.77 trillion in June), with a raise of around $100 billion — an order of magnitude never seen for a debut listing.

But the leaked prospectus accompanying the filing tells a very different story. $4.6 billion in revenue in 2025. $8.06 billion in operating losses. And above all, $518 billion in future cloud/compute/infrastructure obligations, "almost all non-cancellable," as reported by Yahoo Finance. This gap between the narrative and the books makes this filing far more than financial news: it's the first real-world test of the thesis that "AI capex is not a bubble."

Key takeaways

• Timeline: roadshow launched on October 14 in San Francisco, IPO targeted before Thanksgiving (November 26), expected valuation between $1.8T and $2T.
• Record raise: ~$100B expected, with Morgan Stanley, Goldman Sachs, and JPMorgan Chase as lead underwriters.
• 2025 financials: $4.6B in revenue (12x year-over-year), $8.06B operating loss, ~$42B net loss including ~$34B in non-cash accounting charges.
• The ticking time bomb: $518B in upcoming cloud/compute/infrastructure obligations, nearly all non-cancellable (~$15B/year for SpaceX's Colossus 1, $44.6B with Nscale, up to $42B Broadcom facility).
• Concentration risk: two anonymous customers account for ~25% of revenue (~$1.1B in 2025), and many large customers are not bound by long-term contracts.
• The context: run-rate above $65B by end of July 2026; $2T represents ~435x 2025 revenue, ~31x that run-rate. OpenAI, for its part, has pushed back its IPO.

Recommended tools

To follow this story without relying on rumors, here are the tools we use in our newsroom:

$2T on the stock market: the timeline, the banks, and the super-voting shares

Anthropic's IPO could be wrapped up before November 26, 2026, at a valuation between $1.8T and $2T, with a raise of around $100B. That's the scenario described by Yahoo Finance, which mentions a possible roadshow as early as the week of November 9.

The offering is sized like a mega-deal. Morgan Stanley, Goldman Sachs, and JPMorgan Chase are steering the operation. The expected raise, ~$100B, would surpass anything the market has ever absorbed for a listing.

Another detail that matters: the founders are negotiating super-voting shares. Dario Amodei and his cofounders would retain disproportionate control over decisions after the IPO. Classic in American tech, but worth watching closely in a deal where $518B in commitments bind the company for a decade.

The valuation trajectory, meanwhile, is dizzying. The $30B funding round was still targeting $900B (our analysis). The Series H in May pushed the bar to $965B, a few weeks before the filing of a confidential S-1 with the SEC (our coverage). Two months later, the leaked prospectus points to twice that.

And according to six investors cited by Yahoo Finance, annualized revenue would reach $100B to $120B before the end of the year. A target that, if met, will change the game between the prospectus leak and pricing.

The leaked figures: 12x growth, operating losses up 2.7x

Yes, the growth is real and historic: revenue up 12x in 2025. No, the losses are not a rounding detail: $8.06B in operating losses. Both are true at the same time, and that's precisely what makes this story so difficult.

Here are the key figures from the S-1 as reconstructed by Fortune and Reuters:

*Our own calculation, based on the tripling announced by Yahoo Finance.

Three takeaways emerge from this table.

The growth is violent. Revenue up 12x in 2025, reaching nearly $4.6B. In the first quarter of 2026, Anthropic booked $4.73B — more than in all of 2025. In the second, $11.5B, more than double the previous quarter. By late July, the run-rate exceeded $65B according to Morningstar.

The losses are widening at the same time. Operating loss of $8.06B in 2025, versus $2.98B in 2024. Compute and infrastructure expenses: $7.33B in 2025, three times more than a year earlier — more than the company's annual revenue that year. Total operating expenses: $12.65B.

And the $42B net loss? That's the headline-grabbing figure, and it's misleading. Roughly $34B corresponds to an accounting charge — the revaluation of financing instruments convertible into equity — and not to actual cash outflows, Reuters notes. That still leaves ~$8B in real operating losses for the year, and an adjusted operating result that only turned positive again in the second quarter of 2026.

$518B in compute commitments: "almost all non-cancellable"

The most important figure in this story is neither $2,000B nor $42B. It's $518B.

According to the prospectus reported by Reuters, Anthropic has committed to $518B in future cloud/compute/infrastructure obligations. And these commitments are, for the most part, "almost all non-cancellable" — the phrase used by Yahoo Finance. In other words: whatever happens to demand, the bill will come due.

The commitments detailed in the financial press give a sense of the mechanics:

Put these figures in perspective and the vertigo returns. $518B is roughly 70 times Anthropic's 2025 compute spend ($7.33B) and more than 110 times its annual revenue. Even measured against the July 2026 run-rate ($65B), the commitment represents nearly eight years of theoretical revenue — our calculations based on prospectus data.

Why bind itself to such sums? Because that's the price of access to the compute capacity needed to serve demand projected to explode. Anthropic is pre-buying the compute that is meant to serve its $65B run-rate of tomorrow. If demand shows up, the bet is brilliant. If it falls off, these obligations become an anchor. There's no middle scenario — and that's exactly what's called a structural bet.

It's this liability that weighs most heavily on the valuation equation, as Markets Insider and TipRanks point out: $2,000B gross is not $2,000B net.

Two clients, a quarter of revenue: the hidden fragility of the case

Nearly a quarter of Anthropic's revenue rests on two anonymous clients — roughly $1.1B out of the $4.6B for 2025. And that's not the only red flag: many large clients are not bound by long-term contracts, Reuters points out in its breakdown of the prospectus.

The double exposure is classic — and fearsome. Concentration first: if either of the two major clients were to scale back, the growth curve would fall flat on its face. Revenue quality next: clients not bound by multi-year commitments can walk away at every renewal, without any economic warning.

This is precisely what Morningstar invites you to look at before getting carried away: growth in spending from existing clients, the share of contracts with minimum spend commitments, and the utilization rate of capacity already paid for. Three metrics that will tell whether the $518B in commitments are backed by real demand or by a projection.

Our take: at a $2,000B valuation, the market will want to see this concentration go down, not up. A quarter of revenue from two unnamed clients is a venture capital profile, not that of an income stock.

The full-scale test of the thesis that "AI capex is not a bubble"

Anthropic's IPO is the first opportunity for the market to publicly put a price on what a massive compute commitment is worth — and therefore to settle, in real time, the bubble debate.

This debate has been going on for months. Anthropic's CIO, Krishna Rao, entered the arena to assert that "our capex is not a bubble" (our analysis). On the other side, Microsoft, Meta and Google have climbed to $725B in combined AI capex (our analysis), fueling fears of a systemic overheating.

Until now, the debate remained theoretical: AI labs were private, their accounts opaque, their commitments invisible. Anthropic's S-1 changes that. For the first time, we can see the internal mechanics of a frontier lab: the revenue, the losses, and above all the take-or-pay contracts that lock in its future.

In this respect, Morningstar's reading is the most useful: the leak strengthens the case for rapid growth, but weakens the case for a $2T valuation being justified today. In other words, the fundamentals are excellent; the price, however, requires an act of faith.

November's pricing will therefore be a referendum. If the market absorbs $100B worth of shares at a $2T valuation, the "not a bubble" thesis receives its strongest endorsement to date. If the order book struggles and the price gets trimmed, the correction will spread beyond Anthropic — starting with the hyperscalers' $725B of capex.

Anthropic vs OpenAI: one window, two strategies

Anthropic is moving forward, OpenAI is waiting. And for the first time, the one moving forward is overtaking the one waiting on the revenue line.

Yet OpenAI had prepared its own confidential filing, valued at around $730B — while Anthropic overtook it in revenue (our analysis). Then it pushed back the deadline: Sam Altman deems the current IPO ill-advised, reports Yahoo Finance.

The market balance of power has flipped in just a few months. Anthropic is seizing the public window, with an expected raise of ~$100B that would give it a war chest capable of funding part of its $518B in commitments without depending solely on private capital. OpenAI, which remains private, will have to keep raising in a closed circle, in a context where its own filing awaits a better window.

Two strategies, one underlying bet: that AI demand holds long enough to justify the billions committed. The difference is that Anthropic is now putting that bet in writing, figure by figure, before the regulator and the market.

❌ Common Mistakes

Mistake 1: Confusing net loss with cash burn

The $42B net loss includes ~$34B in accounting charges (revaluation of instruments convertible into equity), not actual cash outflows. The real operational thermometer: $8.06B operating loss in 2025, and adjusted operating income back in positive territory in Q2 2026. Solution: read the S-1 systematically distinguishing accounting charges from actual cash flows.

Mistake 2: Extrapolating the run-rate in a straight line

A run-rate above $65B in late July doesn't promise $65B in guaranteed annual revenue. One quarter's growth can't be mechanically projected, especially with customers not bound by long-term contracts. Solution: track Morningstar's three metrics — spending by existing customers, share of contracts with minimum spend commitments, utilization of paid capacity.

Mistake 3: Reading the valuation without the $518B

$2T gross says nothing about net value. Non-cancellable commitments represent a future liability that weighs on the equation, as Markets Insider and TipRanks point out. Solution: reason in terms of valuation net of commitments, and in the number of years of revenue needed to cover them.

Mistake 4: Believing the prospectus leak will derail the IPO

The confidential S-1 had been on file since June; qualified investors have known these figures for a long time. A media leak doesn't change the mechanics of a roadshow. Solution: separate the media noise from the actual process — October 14 invitation, November roadshow, pricing targeted before Thanksgiving.

❓ Frequently Asked Questions

When will Anthropic go public?

The roadshow could start as early as the week of November 9, targeting a listing before Thanksgiving on November 26, 2026. The expected valuation is between $1.8T and $2T, with a raise of around $100B led by Morgan Stanley, Goldman Sachs and JPMorgan Chase.

What exactly do the $518B in obligations represent?

These are future cloud/compute/infrastructure commitments disclosed in the prospectus, nearly all of them non-cancellable: ~$15B per year for SpaceX's Colossus 1 data center, a $44.6B cloud deal with Nscale, and a credit facility of up to $42B with Broadcom.

Does the $42B net loss mean Anthropic is burning $42B in cash?

No. About $34B corresponds to a non-cash accounting charge — the revaluation of financing instruments convertible into equity. The actual operating loss stands at $8.06B in 2025, and adjusted operating income turned positive again in Q2 2026.

Why is the $2T valuation being challenged?

Because it represents ~435x 2025 revenue ($4.6B) and ~31x the July 2026 run-rate ($65B). Add to that two customers accounting for ~25% of revenue and $518B in non-cancellable commitments. Morningstar believes the leak strengthens the growth thesis, not the valuation thesis.

What risks are cited in the prospectus?

Revenue concentration — two customers accounting for nearly a quarter —, the absence of long-term contracts for many large customers, and the weight of the $518B in non-cancellable obligations. All factors that expose the company to a downturn in demand.

How can you follow the story live?

The confidential S-1 was filed with the SEC in June. The public version will appear on SEC EDGAR before pricing. In the meantime, Reuters, Yahoo Finance and Fortune are publishing verified excerpts from the prospectus.

✅ Conclusion

Anthropic is going public with the fastest growth in the sector and the heaviest compute bill ever put on paper in a prospectus — and the market will decide, before Thanksgiving, whether the "AI capex isn't a bubble" thesis survives its first real-world test. To keep reading, check out our breakdown of CIO Krishna Rao's response on AI capex.