Anthropic is in talks to acquire Decart, an AI infrastructure startup, for $6 billion, according to reporting by Bloomberg. The deal would come less than four months after Decart raised $4 billion in May 2026 at a $4 billion valuation.
The acquisition targets compute efficiency rather than raw model capability. Decart has built systems to optimize how AI models use computing resources during training and inference, a constraint that major AI labs face as model scale increases. Rudina Seseri, founder of venture firm Glasswing Ventures, told Bloomberg that leading AI companies like OpenAI and Anthropic have reached a point where their existing architectures create inefficiency. "Their success is also their limitation, which is they're not efficient," Seseri said in an interview.
Anthropic has raised $7.3 billion in total funding and completed a $2 billion investment from Google in January 2025. The company competes directly with OpenAI and other frontier labs on large language model development, with the efficiency gap widening as models approach and exceed 100 billion parameters. Infrastructure constraints have become the binding constraint in AI scaling rather than algorithmic innovation, according to industry observers.
Decart was founded by researchers focused on compiler optimization and GPU acceleration. The startup's technology allows models to run with lower memory bandwidth and power consumption without degrading output quality. A $6 billion price tag would represent a 50 percent premium over Decart's most recent valuation.
The deal has not closed. Antrophopic and Decart declined to comment on the talks when contacted by Bloomberg on August 13, 2026.
If completed, this would be one of the largest acquisitions in AI infrastructure to date. The previous largest deal of this type was Databricks' acquisition of MosaicML for $1.3 billion in June 2023. OpenAI has not announced comparable infrastructure acquisitions.
The number to watch is whether Anthropic closes this deal by December 2026, which would determine whether the company rebuilds its training stack before launching its next-generation model.
