SemiAnalysis just released a deep-dive report on Anthropic, and it's quite something to digest.

In one sentence: Among AI labs, someone has finally managed to run both "growth" and "profitability" at the same time.

• ARR: $9 billion → $30 billion → $60 billion+, like a rocket
• Net dollar retention (NDR) 500% — existing customers keep growing on their own
• Gross margin flipped from -94% to 60%+, API business at 80%+
• 3Q26 operating profit exceeds $1 billion

The most striking contrast is with OpenAI: one charges by usage and has turned gross margin positive; the other still relies on subscriptions and has a -100% margin.

The valuation anchor in the report: $6 trillion, base case, not optimistic.

The logic is actually simple:
→ High-margin inference revenue
→ Invest in the next generation model
→ Widening intelligence gap
→ Pricing power becomes more stable
→ More high-margin revenue

The flywheel spins, and no one else can catch up.
→ It should IPO first, forcing OpenAI to raise money from a worse position.

The first to go public locks in capital — and also the narrative.

But there are a few thorns:
• Enterprises are starting to set budget caps on AI
• OpenAI is reportedly planning to cut token prices
• Compute gap is a hard constraint: 100GW+ needed by 2030
• Regulatory model lockdowns — low probability, but tail risk