Anthropic has acquired Decart in a $6.0 billion deal, bringing the AI infrastructure startup under its corporate umbrella. The transaction, confirmed by both parties, transfers full ownership of Decart’s technology and team to Anthropic, which will integrate the acquired operations into its existing model development pipeline.
Decart specializes in systems-level AI infrastructure that delivers a tenfold improvement in both training and inferencing for large generative models. The company also develops its own foundational generative interactive models, designed for real-time accessibility. Anthropic, known for its frontier AI research and safety-focused model development, will now control these capabilities directly rather than relying on external partners for such efficiency gains.
The acquisition is driven by a need to compress compute costs and latency across Anthropic’s model lifecycle. Decart’s infrastructure expertise is expected to shorten training cycles for Anthropic’s largest models while enabling faster, more responsive inferencing for deployed systems. By owning this layer of the stack, Anthropic reduces its dependence on third-party optimization vendors and gains tighter control over performance benchmarks.
For Decart, the deal provides access to Anthropic’s scale, including substantial compute resources and distribution channels for its interactive model research. The combined entity will likely prioritize the integration of Decart’s real-time inference capabilities into Anthropic’s consumer and enterprise products, though no specific product timelines have been disclosed. The transaction is expected to close within the current quarter, subject to standard regulatory review.
Post-acquisition, the focus will shift to harmonizing Decart’s infrastructure tools with Anthropic’s existing training frameworks. Early priorities include benchmarking the tenfold efficiency gains on Anthropic’s production workloads and determining which Decart models will be maintained independently versus folded into Anthropic’s broader model family. The outcome will determine whether the combined operation can sustain its cost advantage in an increasingly competitive AI market.