🚨 Emergency Ep: Anthropic’s Massive IPO. Will It Sink? 🚨

Adam and Adir pull apart Anthropic's S-1: $16.3B of half-year revenue, a 900% NRR, $475B of compute commitments, an 85% margin with an asterisk, and whether a $2T Nasdaq float gets away.

Listen on Spotify here or Apple Podcasts here.

🔵 REMINDER! Join us at our next Contrarians Live on the 8th December hosted with BlueRock. Buy your tickets here. 🔵

1,300% growth, 900% NRR and the end of token maxing

Anthropic filed its S-1 on 26 October to list on Nasdaq as ANT. Revenue for the six months to 30 June was $16.3B, up from $1.2B a year earlier, growth of 1,300%. Revenue in 2023 was $41M. The filing reports net revenue retention of about 900% as of 30 September: customers who were paying a year ago are now paying roughly nine times as much.

Adam: “Have you ever seen a business grow this fast?” Adir: “Never. The question is, have you seen a business with over a billion dollars of revenue grow this fast? I doubt there’s been one in history.”

Adam: “My theory is they grew significantly through getting more cash out of their existing customers, from dabbling in Claude to using it massively and really token maxing. And we’ve seen that token maxing is over for most smart companies. People are asking: how do I reduce my spend? How do I divert to cheaper LLMs? How do I use open weight models?”

Adir: “In the first half of ‘25 it was a bit over a billion dollars of revenue, and in the first half of ‘26 it was $16 billion. That sounds amazing until you realise that the net revenue retention of 900% is nine of that 16.”

Adam: “Claude’s an incredible product. But is it five times as good as OpenAI? Is it a hundred times as good as the OpenAI models? No, we think it’s slightly better. And all the other guys are catching up as well.”

$475B of compute, customers who can walk at will

The S-1 shows at least $475B of future cloud compute commitments, most of them non-cancelable, including about $114.5B with Google and $110B with Amazon. Elon Musk’s SpaceXAI (formerly xAI) is in line for up to $120.5B through 2029, though once the initial periods end in early 2027 either side can cancel on 90 days’ notice. Around 80% of Anthropic’s revenue is consumption-based, and its largest customers buy on usage with no long-term contracts. Anthropic held $112.4B in cash at 30 June, out of $141.1B raised.

Adam, borrowing a line from Evan Armstrong: “Revenue, meanwhile, is consumption priced and cancellable at will. So this sounds a lot like a bank to me. They’re borrowing long and lending short, which is a recipe for a bank run and a recipe for a collapse. Am I being too dramatic?”

Adir: “You say they’re non-cancelable, but they are cancelable by the liquidator. These businesses have to not only stay alive, but continue to access funding in order for all your promises and dreams to come true as a supplier.”

Adir: “A hundred billion dollars feels like a lot of money to raise, but it turns out it’s not that much between friends when your business is AI, because you blow more than that in one year’s worth of compute. This cannot possibly be the end of their aggressive capital raising.”

Adam on what Claude said when he ran the S-1 through it: “The angle isn’t that it’s secretly losing a fortune today. The angle should be: today’s numbers are a snapshot taken at the most flattering moment. The bill hasn’t arrived yet.” Adir: “Good on you, Claude, for throwing your masters under the honesty bus.”

Hollywood, not SaaS: an 85% margin and models that expire in six months

Anthropic doesn’t report a gross margin, but cost of revenue of $2.38B against $16.3B of revenue implies about 85% for the first half. Compute for model development sits in R&D, and fees paid to cloud channel partners, up $2.7B year on year, sit in sales and marketing. Models released in the six months to 30 September accounted for more than 80% of third-quarter revenue.

Adir: “Basically in the gross profit, the only thing taken out is the air that people breathe inside the Anthropic office. It’s basically a useless, pointless number except for the purposes of catching out the lazy investor.”

Adam: “I think of foundation models as being akin to Hollywood, not Silicon Valley SaaS. In Tinseltown, you’re only as good as your next movie. In the LLM game, you’re only as good as your next model.”

Adam: “The more they scale, the more it costs them in compute to build the best training model. So it feels like the scale benefit here is at best weak and at worst non-existent.”

Adir: “All of this work and all of this effort to create this amazing software has turned out to just be building a commoditised piece of software that is interchangeable with the five other commoditised pieces of software, and the 500 that are probably not as good but good enough.”

One IPO record in seven years, three in one year

Saudi Aramco’s $25.6B float in 2019 held the record for the largest IPO for almost seven years. In 2026, SpaceX raised $85.7B on Nasdaq in June, SK Hynix raised $26.5B through a Nasdaq ADR listing in July, and Anthropic could raise $100B or more. Closer to home, AI data centre operator Firmus withdrew its ASX float on 9 October after weak demand.

Adir: “When you say the previous record holder held for seven years and then this year there’s going to be three record IPOs, that possibly tells you it’s bubble time.”

Adam: “Australia had its test two weeks ago with Firmus, and Australia passed with flying colours.”

Adir: “Ian Patrick, the chief investment officer of the Australian Retirement Trust, which is the second largest superannuation fund in Australia, told the AFR conference that AI-related investments accounted for more than 25% of the fund’s portfolio. I think this is the singularly highest risk moment in the AI boom slash bubble, this IPO.”

$2 trillion needs $10B of profit. Where’s the road?

Reports put the IPO at $100B or more at a valuation of $1.5T to $2T, with listing targeted for the week of 16 November. Anthropic was valued at $965B in its May private round, and at $183B in September 2025. Nvidia has reportedly considered investing as much as $10B in the offering.

Adir: “I’m going to be generous and say I’ll pay 200 times earnings for it. You’d have to make $10 billion of net profit after tax on a PE of 200 to justify $2 trillion. And what’s the road to $10 billion of NPAT on this business? There is no obvious road to get there.”

Adam: “Anthropic isn’t a luxury brand. People aren’t going to pay 30x for what you can get from DeepSeek.”

Adir: “I think this will get away. I think it will be a kicking-the-can-down-the-road moment, because they’re going to need more money.”

Adir: “Will you lose 90% of your money or more with Alphabet or Microsoft or Nvidia? Very unlikely. But much more likely you’ll lose 90% of your money here. If you want to bet on AI, this is not the risk-return curve to bet on.”