Anthropic Hits the Public Markets While Wall Street Still Can't Count AI Returns
Anthropic filed its S-1 this morning. Dario Amodei stood on a stage in San Francisco and told a room full of suits that AI return doubts are, quote, "a problem of imagination, not economics." The stock will list next week. The valuation target sits at $68 billion. That's not a typo.
The Thing
Here's what the paperwork actually says. Anthropic pulled $4.2 billion in revenue for fiscal 2025, up from $1.6 billion in 2024. Gross margins landed at 34%. That's brutal for a software company. Most SaaS runs 70-80%. But Anthropic isn't SaaS. It's an infrastructure play running on compute that costs a fortune before you serve a single request.
The net loss? $3.8 billion last year. Yes, they almost matched revenue with losses.
Amodei's pitch is simple. The spending is the moat. Every dollar burned on training and inference hardware stacks capabilities competitors can't replicate without spending the same dollar. He called it "the forge" three separate times during the presentation. The audience ate it up.
On X, the reaction split hard. @swyx pointed out that 34% margins with $3.8B in losses means Anthropic needs to roughly triple revenue just to break even on current spend rates. Others noted that Google and Microsoft can subsidize their AI divisions with search and cloud revenue forever, but Anthropic has no such sugar parent post-IPO. The company has Amazon as a backer, sure. But Amazon's obligation ends at the contract line.
The S-1 also reveals something interesting. Enterprise API revenue now makes up 61% of total revenue, up from 38% eighteen months ago. Consumer Claude subscriptions grew too, but the real traction is developers and companies building on the API. That's the play.
So What
This IPO is a referendum on whether pure-play AI companies can exist as independent businesses.
OpenAI has Microsoft's credit card. Google DeepMind lives inside a $300B ad machine. Meta stuffs AI into a social graph that prints cash. Anthropic is walking into public markets alone. It'll be the first real test.
The margin problem is the story most people will focus on. I think that's the wrong frame. The question isn't whether 34% margins improve. The question is whether they need to improve, or whether the market decides AI infrastructure belongs in a different category altogether.
AWS operated at negative margins for years. Amazon's stock didn't care because the narrative was about capture, not profit. Amodei is making the same bet. He's telling investors that the forge runs hot, and if you're cold about it then stay out.
For builders and startup founders, this matters because it sets the pricing floor. If Anthropic can sustain $68B in market cap while losing nearly $4B a year, the signal is clear. Growth and capability capture trump profitability for the next several years. You have runway. Use it.
But there's a darker read. If the IPO underperforms or the stock tanks six months out, every AI startup with sub-40% margins gets repriced overnight. Venture firms will start demanding path to profitability tomorrow, not in year five. The downstream effects on funding, hiring, and compute budgets would be severe.
What's Next
Watch the first week of trading. Not the pop. The volume after the pop. Institutional money will decide if the forge thesis holds.
Then watch the 90-day lockup expiration. That's when insiders sell. That's when you see what Dario and the early team actually think about the company's prospects. Talk is an S-1. Money on the table is truth.
For anyone building in AI right now, the actionable takeaway is straightforward. If you're selling picks and shovels, you're fine. If you're burning compute to build a moat, you have maybe 18 months of grace before the math comes due regardless of how this IPO goes. Build something that compounds. Build something that gets cheaper to run as it gets better. Everything else is just standing next to the forge and hoping.
Amodei made his bet. Now the market gets to make theirs.
🪶