Some IPO prospectuses you read to understand a stock. Others you read to understand a market. The draft that Reuters was able to review on Monday evening, since also reported on by the Financial Times, Fortune and TechCrunch, clearly belongs in the second group. Anthropic, the developer of the Claude family of AI models, is aiming for a valuation of more than $2 trillion in its Nasdaq listing. That would be the largest initial public offering in history, bigger even than SpaceX in June, which priced at roughly $1.77 trillion and raised $86 billion.
The headline that ran across the terminals on Tuesday morning, however, was a different one: a $42 billion net loss on $4.6 billion in revenue for fiscal 2025. On paper, Anthropic lost a little over nine dollars for every dollar it brought in. Put those two numbers side by side and the obvious question writes itself: is Wall Street about to float the next bubble? The honest answer is more complicated — and for investors who have no plans to buy a single Anthropic share, possibly more important than the question of the offering price. Because the number most likely to move markets over the next few years is neither the loss nor the valuation. It is the $518 billion in future commitments for data centers, cloud capacity and infrastructure. That is not only Anthropic’s problem. It is the revenue of Amazon, Alphabet, Microsoft and their suppliers.
What leaked — and what is still missing
First, the caveat that gets lost in many headlines: this is a draft. Anthropic confidentially submitted a draft S-1 to the Securities and Exchange Commission on June 1, and the document has still not been made public. Figures in a draft can change before the public filing, and there is neither a price range nor a ticker symbol yet. In mid-September Bloomberg reported that Anthropic had chosen Nasdaq and was eyeing a listing as early as October, with Morgan Stanley, Goldman Sachs and JPMorgan named as lead underwriters. According to the latest Reuters report, however, the deal could now slip until after the U.S. midterm elections on November 3, following several delays.
The core figures in the draft, as consistently reported by several outlets:
2025 revenue: just under $4.6 billion, twelve times the prior year. 2025 operating expenses: about $12.65 billion, of which $7.33 billion went to compute and infrastructure alone — triple the 2024 figure. Operating loss: more than $8 billion. Net loss: $42 billion, up 425%. Liquidity at the end of 2025: $20.28 billion in cash, cash equivalents and short-term investments. Then come the 2026 numbers: first-quarter revenue of $4.73 billion — more in three months than in all of the prior year — and second-quarter revenue of $11.5 billion. On an adjusted basis, Anthropic was profitable at the operating level in the second quarter and expects to be again in the third.
And then there are the passages that are unusual in any securities filing. Roughly a third of the risk factors, according to the reports, deal not with competition or regulation but with the behavior of the company’s own models. Anthropic describes conduct its systems have shown or could show, including attempts to “resist shutdown,” to “conceal or manipulate information,” and behavior “resembling blackmail” — and it refers to “existential risks to humanity.”
Why $34 billion of the $42 billion loss is a sign of success
The first reflex on seeing a $42 billion net loss is wrong, and it pays to understand exactly why. According to consistent reports, about $34 billion of it is an accounting charge tied to financing instruments. In earlier rounds Anthropic raised part of its money through instruments that can later convert into shares. When the company’s valuation rises, the estimated value of those instruments rises with it — and accounting rules require that increase to be booked as a loss on the income statement.
In other words, most of the loss is the mirror image of the valuation explosion. In early 2025 Anthropic was valued in the tens of billions of dollars; by May 2026 it was at roughly $965 billion, and on the Nasdaq Private Market secondary platform it recently changed hands at around $1.32 trillion. The more successful the company looked to investors, the larger this paper loss became. No cash left the building. Comparing the $42 billion to the loss of an automaker or an airline is comparing apples to accounting oranges.
That does not make the numbers harmless. Strip out the charge and you are left with an operating loss of more than $8 billion on $4.6 billion of revenue. In 2025 the company spent about $2.75 for every dollar it took in. Critic Ed Zitron put it bluntly: Anthropic spent $12.6 billion to make $4.6 billion. That is the number worth arguing about — not the $42 billion.
The real turn happens in 2026
This is where the prospectus gets interesting, because it tells two very different stories depending on which year you open. 2025 is the build-out year: revenue up twelvefold, costs tripled, heavy losses. 2026 is the year the curve bends. Second-quarter revenue of $11.5 billion is two and a half times the prior quarter and fourteen times the same quarter a year earlier. Back in August it was reported that the annualized revenue run rate had reached about $65 billion by the end of July, up from roughly $9 billion at the end of 2025.
Add to that the claim of a gross margin above 80% — with an important qualifier: before revenue sharing with distribution partners and before the cost of training new models. The distribution partners are essentially the big cloud platforms through which enterprise customers buy access to Claude. Training costs are the line item that, across the AI industry, separates a software margin from an industrial one. And adjusted operating profit is, well, adjusted. Investors should look very closely at what gets excluded once the public filing lands.
Still, the contrast with the usual cash furnaces that come to market in a boom is substantial. A company that books more than double its entire prior-year revenue in a single quarter while turning an adjusted operating profit is not a concept deck. The question is not whether there is a business. The question is what that business is worth — and who is paying for its growth.
The valuation: $2 trillion against a $65 billion run rate
Take the targeted $2 trillion and put it in context. Against 2025 revenue it would be more than 430 times sales — a meaningless multiple nobody will seriously use. Against the annualized run rate of about $65 billion it is roughly 31 times. Against second-quarter revenue annualized ($46 billion), about 43 times.
For reference: Nvidia, the biggest winner of the AI boom, trades at a far lower revenue multiple with a net margin above 50%. Palantir, the most expensive large-cap software name, has at times traded at 70 to 100 times sales. So $2 trillion is not a number from another galaxy, but it assumes growth continues for several more years at a pace no company of this size has ever sustained. And it arrives in an environment more hostile to long-duration valuations than at any time in almost two decades: the 10-year Treasury yield stood at 5.24% on Tuesday morning, its highest level since 2007. The higher the risk-free rate, the less profits that arrive five or ten years from now are worth today.
Then there is the concentration risk Anthropic discloses itself: nearly a quarter of 2025 revenue came from just two customers, who are not named. Large customers, the draft says, lack long-term contracts and could reduce their spending. For a company seeking thirty times sales, that is an unusually candid warning.
$518 billion: the number that matters for Amazon, Google and Microsoft
Now to the number whose significance reaches well beyond Anthropic. According to the draft, the company has taken on $518 billion in commitments for cloud services, compute and infrastructure over the coming years. For comparison, cash stood at $20.28 billion at the end of 2025. The commitments are roughly 25 times the cash pile and more than a hundred times 2025 revenue. Even against the current $65 billion run rate they equal about eight years of sales.
Who sits on the other side of those contracts? Publicly known are the agreement with Amazon, which can cover up to 5 gigawatts of capacity and under which Anthropic uses Amazon’s Trainium chips, and the one with Google, which will supply another 5 gigawatts from 2027 through Google Cloud and its TPU chips. In November 2025 a deal with Microsoft and Nvidia followed, under which Anthropic committed to $30 billion of compute capacity on Microsoft’s Azure. Contracts with SpaceX and the British data center operator Nscale are also cited for 2026.
This is the point investors need to grasp: for the hyperscalers, the $518 billion is not a risk hanging off a startup — it is backlog. A large share of the reported remaining performance obligations at the cloud units of Amazon, Alphabet and Microsoft now comes from a handful of AI labs. And those labs can only honor their contracts if they keep raising capital. Equity analyst Ross Hendricks framed the doubt sharply: even if the IPO pulls in $100 billion of equity, is another $400 billion supposed to come from the debt markets? One report puts roughly $71 billion already running through off-balance-sheet special purpose vehicles, mostly to finance Google’s TPU chips.
The IPO, then, is not just a liquidity event for existing shareholders. It is the funding source that underwrites the cloud providers’ capacity build-out for years to come. If it fails, or only succeeds at a sharply lower price, the damage will not stop at Anthropic. It will run through the revenue forecasts of the entire AI infrastructure chain.
What it means for the stocks you already own
For most retail investors, getting shares at the offering price will be hard. Allocations in a U.S. deal of this size go overwhelmingly to institutions and to clients of the big banks; retail IPO-access programs at brokers such as Robinhood or SoFi usually receive only a sliver. Most individuals will be buying on the open market after the first trade — historically with violent swings in the opening days. Any gain sold within a year is taxed as a short-term capital gain at ordinary income rates, which argues for patience or for holding it in a tax-advantaged account.
The indirect exposure is already in many portfolios. Amazon (AMZN) and Alphabet (GOOGL) are the largest strategic shareholders. Analysts estimate Amazon’s stake at 15% to 20% and Alphabet’s at 10% to 15% — at a $2 trillion valuation, that would be worth $300 billion to $400 billion for Amazon and $200 billion to $300 billion for Alphabet. Amazon reportedly booked $16.8 billion in pre-tax gains from revaluing its Anthropic holding in the first quarter of 2026 alone. Gains like that inflate reported earnings per share while saying little about the underlying business — and in a weaker market they can turn into paper losses just as fast.
The $518 billion also flows further down the chain. Microsoft (MSFT) carries the $30 billion Azure commitment in its backlog. Nvidia (NVDA) is both an investor and a supplier. Broadcom (AVGO), which co-designs Google’s TPUs, is exposed through the Google agreement. And the data center power and cooling names — Vertiv (VRT), Eaton (ETN), GE Vernova (GEV) — have built much of their recent growth story on the same hyperscaler spending plans that deals like these are used to justify. None of them needs Anthropic to succeed tomorrow. All of them need the capital markets to keep funding the AI labs.
For index investors: an S&P 500 fund will not own Anthropic on day one; inclusion depends on seasoning periods and eligibility rules, including profitability tests. But the concentration in a small group of megacaps whose businesses are increasingly entangled with the AI labs is already in the index.
The objections: control, circularity and competition
Three counterarguments deserve to be taken seriously before getting carried away by the revenue curve.
First: buying shares does not buy control. Anthropic is a public benefit corporation. An independent body, the Long-Term Benefit Trust, appoints a majority of the board through its own class of stock. In September the company also proposed giving its seven co-founders, led by CEO Dario Amodei, a combined 50.1% of the vote on most matters through a special share class, modeled on Palantir. That is by design: the structure is meant to shield the company from short-term return pressure. For public shareholders, though, it means carrying the economic risk without a say on strategy.
Second: part of the growth is a closed loop. Amazon and Google are at once shareholders, compute suppliers and distribution partners whose clouds enterprise customers use to buy Claude. Money that flows in as investment flows back out as cloud spending, and the rising value of the stake in turn lifts the investors’ earnings. That does not make the revenue fake — end customers ultimately pay for the usage. But it makes it hard to say how much of the growth would exist without the entanglement.
Third: prices are falling. The cost per unit of text processed is dropping fast across the industry, and Chinese developers are pressuring prices with freely available open-weight models. Anthropic has so far shown it can charge a premium for its frontier models, and independent benchmark rankings currently put its latest models on top. But leadership in AI has rarely gone unchallenged for more than a few months. Tuesday itself showed how quickly the picture can shift: OpenAI has reportedly shelved a planned frontier model after it failed internal safety tests, and on the same day AMD announced it will acquire Fei-Fei Li’s AI startup World Labs for $8.2 billion in stock. The race for the next generation of models is being run by far more than two labs.
Bottom line: the story is not the loss, it is the bill for everyone else
The leaked prospectus answers a question that has hung over the AI trade for two years: is there a business behind the enormous data center spending that will eventually pay for it? Anthropic’s 2026 numbers say yes, there is one, and it is growing faster than almost anyone expected. The $42 billion net loss is largely an accounting echo of the company’s own valuation surge.
The more uncomfortable insight sits in the other number. $518 billion of commitments against $20 billion of cash means that a significant part of the big cloud providers’ capacity build-out rests on the capital markets continuing to fund the AI labs — through IPOs, bonds and special purpose vehicles. As long as that works, the commitments are backlog. If it stops working, they become a question of who ultimately pays for data centers that are already under construction.
For investors, that means two things. If you want to buy Anthropic after the IPO, wait for the public prospectus and scrutinize the adjusted margin, the customer concentration and the funding of the commitments. And if you do not want to buy it, know that through Amazon, Alphabet, Microsoft, Nvidia and a plain S&P 500 fund you already own a piece of the bet. The offering price, whenever it is set, will therefore decide not just the value of one company but the valuation of an entire supply chain — at a moment when the 10-year Treasury pays more than it has since 2007.
Try TradingView Free for 30 Days
Plus get a $15 discount on your first subscription through this link.
Read more in our topic hub: Topic Hub: AI Stocks Investing 2026
