Everybody wants to argue about whether AI is real. The revenue says it is. Anthropic — the company behind Claude — did $41 million of revenue in 2023, $4.6 billion in 2025, and $16.3 billion in the first half of 2026 alone. Existing customers are spending nine times what they spent a year ago. Numbers like that are not a bubble artifact. They are what a genuine technological land rush looks like from inside the income statement.
So the doomers have moved. The serious version of the doom case is no longer “nobody will pay for this.” It’s “look at what it costs to build it — and look at who’s paying.” That argument deserves a hearing, because it isn’t really about Anthropic. It’s about the price of money in America, and Anthropic’s IPO filing, made public on October 26, is the best exhibit both sides have produced yet.
The doomer exhibit
Start with the liabilities. Anthropic has committed to roughly $800–900 billion of compute, data center, and chip contracts over the next decade — and about 80 percent of that is non-cancelable. It is owed whether the revenue shows up or not. The company’s first-half net loss was $80.7 billion on paper; most of that is a non-cash accounting charge from marking its convertible notes to a soaring valuation, but the direction of travel is unmistakable: this is a company whose obligations are measured in the hundreds of billions and whose profits, so far, are measured in the occasional good quarter.
Then look at how the whole industry is funding itself. The five big hyperscalers will spend about $795 billion on capital projects in 2026 and over a trillion in 2027. They used to pay for this out of cash flow. They don’t anymore — they issued around $230 billion of bonds in the first nine months of 2026, against an average of roughly $30 billion a year in the early 2020s. Count the whole AI complex and 2026 issuance runs to about $570 billion. One Meta data-center vehicle borrowed $27.3 billion at 6.581 percent, due 2049. Read that again: a private project paying six and a half percent for 23-year money, in a country whose government borrows for thirty years at about five.
And the market has started to flinch. Corporate spreads on the hyperscalers widened. Amazon watched a large share of the orders on one bond deal walk away when the pricing got tight. AI-linked issuance fell from $113 billion in June to $23 billion in September. Alphabet — the strongest balance sheet in the business — reported a quarter with $51 billion of capital spending and negative free cash flow, and the stock dropped 12.6 percent in a day. Even the Anthropic IPO itself was pushed back past the midterm elections, explicitly because Treasury yields had surged and investors had started asking harder questions.
The doomer conclusion: seven giant companies and one federal government are all trying to borrow from the same pool of savings at the same time, and the pool is not bottomless. Somebody’s cost of capital has to rise. Everybody’s does.
The bull exhibit
The other side reads the same filing and sees something else. A company growing revenue fourteen-fold year over year, with 900 percent net revenue retention, that posted an operating profit in the second quarter, sitting on more than $110 billion of cash, does not need your sympathy. It needs a ticker symbol.
The IPO — expected this month on Nasdaq, under the symbol ANT — targets a valuation of up to $2 trillion and a raise of up to $100 billion, which would make it the largest public offering in history, bigger even than SpaceX’s listing in June. NYU’s Aswath Damodaran, nobody’s idea of a cheerleader, ran the filing and landed at about $1.475 trillion — the low end of the range, but not a refutation of it. His base case has Anthropic approaching a trillion dollars of revenue by 2036.
And on the crowding-out question, the Treasury Department’s own Borrowing Advisory Committee just studied exactly this — they gave it a formal charge, under the heading of “Super Intelligence-related issuance” — and reported that Treasury auctions show no evidence yet that tech borrowing has crowded out demand for government bonds. The government’s paper still clears. The pool, so far, has stretched.
What the fight is actually about
Both sides are describing the same machine from different ends. The doomers are wrong if they think this is Pets.com — the customers and the revenue are real. The bulls are wrong if they think revenue settles the question — because the question was never demand. The question is the discount rate.
Damodaran’s valuation uses a cost of capital of 9.74 percent. Sit with that number. The entire AI build-out — the trillion-dollar capex years, the $900 billion of non-cancelable compute contracts, the $2 trillion IPO — has to clear a hurdle rate near ten percent, in a world where the ten-year Treasury peaked at 5.4 percent in October and the Federal Reserve’s own projections hold policy rates above four percent through 2027. That is a fundamentally different world from the one that priced the last decade, when money was free and any future cash flow, however distant, was worth nearly face value today.
This is why the crowding-out question matters more than the bubble question. A bubble is an opinion about one sector. The cost of capital is a fact about every sector. When the federal government runs deficits of six percent of GDP — $1.8 trillion a year, rising past $3 trillion within a decade on current projections — and the AI complex simultaneously demands hundreds of billions a year of long-dated financing, the result isn’t necessarily a crash. It’s a term premium: investors demanding to be paid more, permanently, to lend long. We lived with a term premium near zero for a decade because central banks suppressed it. The historical norm is 150 to 200 basis points. Put that norm back on top of a four-percent policy rate and you get a ten-year Treasury near six percent without a single thing going wrong.
The verdict to watch
The Anthropic IPO pricing, when it comes, will be the cleanest read we get. If it clears at $2 trillion, the market is saying the growth outruns the cost of capital. If it prices at $1.5 trillion — the number disappointed investors were kicking around at the October briefing — the market is saying the arithmetic is tightening. If it gets pulled again, the doomers won’t need to argue anymore.
My own read, for what it’s worth: the doomers have the better of the financing argument and the worse of the technology argument, and most commentary confuses the two. AI is real. Money is no longer free. Both things can be true at once, and the collision between them — not a crash, just a long repricing — is the story of the next several years for anyone who borrows, lends, builds, or holds assets priced off the old world.
If you own a house, the ten-year Treasury is your mortgage rate’s anchor. If you own a business, it’s your loan rate and your buyer’s loan rate. If you hold savings, it’s the first good news in fifteen years — a Treasury bill pays over four percent again, tax-free at the state level here in California. The doomers and the bulls will keep shouting about a stock. Watch the bond market instead. That’s where this gets decided.
