Anthropic’s Reported Decart Talks Show Compute Efficiency Is The New Prize

Anthropic is reportedly in talks to buy Decart AI for about $6 billion, and the deal, if it happens, would say a lot about where the AI race is heading.

Reuters reported that Anthropic is in talks to acquire the Nvidia-backed Israeli startup, citing a source familiar with the matter. Bloomberg first reported the possible $6 billion price, while Reuters noted that the discussions may not result in a deal. Anthropic declined to comment, and Decart did not immediately respond outside regular business hours.

Decart is not a household name in the way Claude is. But the company’s work sits in a very important part of the stack: AI infrastructure, inference optimization and real-time generative systems. In simpler terms, Decart is the kind of company that can help make AI cheaper, faster and more usable at scale.

That makes the reported talks more interesting than a normal acquisition rumour. Anthropic already has strong models, serious enterprise traction and one of the most valuable AI brands in the market. What it needs now is the ability to serve demand efficiently without drowning in compute costs. Every frontier AI company has the same problem: users want more intelligence, longer contexts, richer media and faster agents, but the infrastructure bill keeps rising.

This is why optimisation startups are suddenly strategic. A model company can win benchmarks and still struggle if inference is too expensive. It can win enterprise pilots and still lose customers if competitors offer similar results at lower cost. The next phase of AI competition may therefore be less about who has the biggest model and more about who can deliver useful intelligence most efficiently.

Anthropic’s own business direction makes this obvious. Claude is now used across coding, enterprise workflows, research and creative work. We have covered Claude’s role in mathematical research, the company’s watermarking response to AI regulation and investor expectations around Anthropic’s public-market future. All of that growth requires infrastructure.

Decart also fits the generative video and real-time AI direction. Video, world models and interactive AI experiences are far more demanding than text chat. If Anthropic wants Claude to move deeper into multimodal agents, enterprise tools or simulation-like systems, it will need more than model talent. It will need better infrastructure economics.

There is also a defensive angle. OpenAI, Google, xAI, Meta and Chinese labs are all trying to cut inference costs and improve deployment speed. DeepSeek, Kimi and other Chinese models have sharpened the price pressure. A $6 billion acquisition may look large, but if it helps Anthropic reduce compute waste or open new product categories, investors may see it as cheaper than falling behind.

Still, this is not done. Reported talks can fall apart, and a big acquisition would bring integration risk. AI infrastructure teams are valuable because they are specialized. Buying them does not automatically mean their technology slots neatly into a large frontier lab.

The important point is the signal. Anthropic is not only competing for the best model. It is competing for the best AI economics. If the Decart talks become a real deal, it would show that the next AI acquisition wave may target the companies that make intelligence faster, cheaper and easier to serve, not just the companies that create flashier demos.