Anthropic hits first profitable quarter at $10.9B revenue; Karpathy joins to lead pretraining

Anthropic's first profitable quarter at $10.9B in revenue — roughly double the prior period — lands as a watershed for the frontier-lab business model. Until now, the consensus was that frontier labs would burn cash for years; Claude's enterprise ramp via KPMG, AWS Bedrock, and direct API has flipped that script faster than analysts modeled.

The more consequential signal is Andrej Karpathy's move from OpenAI alumni status to leading Claude pretraining. Karpathy's public position throughout 2025 was that post-training and tool-use carried most of the modern model's gains. Taking a pretraining seat at Anthropic is a reversal — or at least a hedge — and the community is reading it as a bet that scaling laws still have meaningful runway.

Competitively, this puts Anthropic on a collision course with OpenAI on raw capability, not just enterprise polish. It also lands the same week as the Opus 4.8 Vertex sighting (with Sonnet 4.8 reportedly close behind) and the Project Glasswing disclosure — a portfolio of moves that suggests Anthropic believes 2026 H2 is the window to close or open a capability gap.

Skeptics note that profitability at $10.9B is meaningful but fragile: API margins compress every time DeepSeek cuts prices, and Microsoft's Claude Code cancellation hints that even Anthropic's whales are starting to feel token costs. Watch next: how Karpathy reorganizes pretraining, and whether the next Claude generation cites a new scaling regime.