The company building Claude is finally showing its books to the people who might take it public — and the numbers suggest AI inference just became a real business, not a science project.
The Summary
- Anthropic is hosting an institutional investor day, signaling potential IPO preparation as market confidence in AI business models solidifies
- SemiAnalysis estimates Anthropic's inference business could hit 88% gross margins, putting it in rarified software territory alongside the best SaaS companies
- The company reportedly signed a multiyear cloud deal with IREN, suggesting infrastructure expansion ahead of scaling demand
- If the margin numbers hold, Anthropic just proved AI companies can make money selling intelligence, not just burning capital to create it
The Signal
Anthropic is doing something most AI labs avoid: showing institutional investors the actual economics of running an AI company. The timing of this investor day matters because the AI industry is splitting into two camps. One burns billions training ever-larger models. The other figured out how to turn inference — the actual act of answering questions — into a high-margin business.
The reported 88% gross margins from SemiAnalysis put Anthropic in the same league as peak-era software giants. For context, Salesforce runs around 75% gross margins. AWS does 60%. If Anthropic is really hitting 88% on inference, they've solved the unit economics problem that's supposed to make AI companies unprofitable forever.
"Anthropic's potential high margins could accelerate its path to profitability, influencing competitive dynamics and investment in AI sectors."
The IREN cloud deal isn't about diversifying away from AWS or Google Cloud. It's about locking in compute capacity before demand spikes. Companies don't sign multiyear infrastructure deals unless they're confident revenue will scale to match. This is the move you make when you know what's coming.
The broader pattern: Anthropic is de-risking the IPO story. Strong margins prove the business model works. Infrastructure deals prove they can scale. An investor day proves they're ready to be held accountable by public market standards. Most AI companies are still in the "trust us, AGI is coming" phase. Anthropic is showing spreadsheets.
What makes this different from OpenAI's trajectory is the focus on sustainable economics over raw scale. OpenAI raised at a $157 billion valuation on the promise of what's next. Anthropic seems to be building the case on what's profitable now. Both strategies work, but only one survives public market scrutiny without a constant need for new capital injections.
Key factors driving institutional interest:
- Proven inference margins that rival best-in-class software
- Diversified infrastructure partnerships beyond hyperscaler dependency
- Repeatable revenue model selling API access vs. consumer products
The Implication
If Anthropic goes public with these margin profiles intact, it resets the valuation framework for every AI company. Suddenly "AI company" doesn't mean "cash incinerator with moonshot upside." It means "software company with compute overhead." That's a category institutional investors actually know how to price.
Watch what happens to private AI valuations in the next six months. If Anthropic proves you can run inference at SaaS-like margins, every VC-backed AI lab will need to explain why they can't do the same. The capital advantage shifts from whoever raises the most to whoever spends the least per query. That's a fundamentally different game, and most AI companies aren't built for it.