Anthropic, the company behind the AI assistant Claude, filed its IPO prospectus on October 26. In Q2 2026, it earned more than $970 million in operating profit, its first in the six quarters listed, yet booked a net loss of more than $66 billion. Is it actually making money?
Anthropic, the company behind the AI assistant Claude, filed its IPO prospectus on October 26. In Q2 2026, it earned more than $970 million in operating profit, its first in the six quarters listed, yet booked a net loss of more than $66 billion. Is it actually making money?
This video covers:
Produced by MILO Investments. The sole source is Anthropic's preliminary prospectus (Form S-1), filed on October 26, 2026 (U.S. Eastern Time). Financial data is as of June 30, 2026; half-year and quarterly figures are unaudited. The offering price, Q3 revenue and total compute commitments are still blank in the filing, so this video does not discuss valuation. The research reflects the filing as of October 26, 2026.
Research commentary only, not investment advice; the author may hold the stocks mentioned.
0:00In Q2 2026, the quarter ending in June, Anthropic's business earned an operating profit of about $0.97 billion, its first operating profit in the six quarters listed in its IPO prospectus. Yet in the same quarter, its books show a net loss of more than $66 billion.
0:20Anthropic is the company behind the AI assistant Claude. On October 26, it filed its IPO prospectus to list on Nasdaq under the ticker ANT. The offering price is still blank, so we won't ask whether it's cheap or expensive, only this: is it actually making money?
0:40Start with revenue. In 2023 it brought in just forty-one million dollars for the whole year; in the first half of 2026 alone, it brought in more than $16 billion.
0:53Q2 revenue grew more than 1,300% year over year, while its biggest cost, compute, grew only 300% over the same period. Revenue outran costs, and that is how operations turned profitable.
1:08So where did that huge loss come from? The answer is counterintuitive: the company became too valuable. Amazon still holds about $6.2 billion of convertible notes, money lent to the company that can be swapped for shares at an agreed price, the lowest tier being just over $2 a share.
1:29The more the company is worth, the more valuable these cheap shares become, and the bigger this 'debt' gets; under accounting rules they are marked to market, and the increase is booked as a loss. In Q2 this item came to nearly $60 billion, and another $8.4 billion came mainly from the initial recognition and remeasurement of other financing agreements.
1:54This loss isn't cash going out; these notes are expected to be settled in shares. In the first half, operations actually brought in nearly $4 billion of cash, partly thanks to the timing of payments and collections. The company also expects heavy net cash outflows from operating and investing activities ahead as it builds out compute. The cost is shares: at the agreed prices, the notes convert into more than 1.8 billion shares in total; if the IPO price is below the agreed price, they convert at the IPO price, into even more shares; that dilutes Class A shareholders, meaning each share gets a smaller slice.
2:36So the huge loss is mostly an accounting entry; the real test is the compute bill. Just four contracts, with Microsoft, Google, Amazon and Broadcom, carry minimum payments of more than $430 billion, paid over five to ten years; three must be paid whether or not the capacity is used, and one lease can't be cancelled during its term; other contracts are mostly caps or estimates and aren't counted here; the total commitment is still blank in the prospectus.
3:10The company also writes that when large amounts of new compute come online, the costs land first, margins may be squeezed, and this will swing in cycles; near term, it will also lean more on pricier short-term compute.
3:25So the evidence that it has 'started making money' is still thin: among the quarters disclosed, only Q2 was profitable, and Q1 still had an operating loss of more than $1.6 billion; the company also expects revenue growth to slow, possibly sharply.
3:43The company also estimates that 5% to 7% of its revenue in the first nine months came from customers violating its usage policy or commercial terms; it expects that share to fall, but how much that will hurt revenue is uncertain. Customers are concentrated too: the top five made up about a quarter of revenue in the first nine months, though that share is falling; nearly half of first-half revenue was transacted on cloud providers' marketplaces, and some of those cloud providers are also its investors and rivals.
4:23Then again, it has plenty of cash: more than $130 billion in cash and short-term investments at the end of June, with over $110 billion raised through financing in the first half alone; the company says that, together with operating cash and its credit facility, this covers at least the next twelve months.
4:46Finally, the Class A shares in this offering carry one vote each; but a single special share held jointly by the seven co-founders carries 50.1% of the voting power, and a majority of the board is chosen by the Long-Term Benefit Trust; the company itself says Class A shareholders may be unable to influence important decisions. And it is registered as a 'public benefit corporation': its directors must weigh the charter's public-benefit purpose alongside shareholders' interests, and the two may conflict.
5:19The company has also placed about 5% of its shares, counting options and convertible notes, into its own public benefit fund. After the IPO, each time the share price rises another 50% from the IPO price and holds there, the board may, at its discretion, issue up to another 2%.
5:43So my view: it has managed one profitable quarter of operations, but hasn't yet proven it can keep making money. The huge loss isn't the biggest risk; the compute bill is. And holding Class A shares means a proportional share of the company's economic value, not a hand on the steering wheel.
6:05The next thing to watch is the amended prospectus: the offering price, Q3 revenue and the total compute commitment are all still blank. Don't judge by whether a single quarter is profitable; watch whether revenue keeps outrunning compute and other costs.
6:24The same set of books, read one line differently, can lead to a completely different conclusion. The author may hold the stocks mentioned in this video; this is research commentary, not investment advice. See you next time.
Ahead of Tesla's Q3 earnings, a closer look at Q2: revenue rose by a quarter year over year, while operating profit fell by more than half. More vehicle sales did not translate into higher earnings.
Before Alphabet reports Q3, a closer look at Q2: of the more than $9 in earnings per share, over $6 came from gains on equity investments. The money is real, but customers did not pay it.
