Anthropic builds compute team with 12+ senior hires for recursive AI self-improvement

Anthropic has quietly executed a year-long hiring spree, pulling over a dozen senior executives from top AI labs and cloud providers. The pattern is unmistakable: the densest cluster landed on the Compute team. This is not a prestige play—it is a strategic acquisition of the three critical inputs for recursive self-improvement: frontier research talent, raw compute capacity, and the enterprise/government distribution channels that generate the revenue to sustain both. The hires include researchers, infrastructure operators, and go-to-market leaders, all aligned around one mission: making AI that helps build better AI.

Notable moves include Andrej Karpathy (ex-OpenAI, ex-Tesla) joining to work on Claude improving its own architecture; John Blomfield (ex-Microsoft) to turn compute into a founder-level operating problem; and Maria Carlson (ex-AWS) to secure institutional permission and demand. Others like Marcus Nordeen (ex-Google DeepMind) focus on physical capacity activation, while Gabriel Fontoura (ex-xAI) works on extracting more intelligence from every cluster. The combined team forms an 'intelligence-production flywheel' where research, compute, and revenue feed each other. For competitors, this signals that Anthropic is betting on self-improving models, not just scaling existing ones.