Anthropic is reportedly in talks to buy Decart, an Israeli AI startup, for roughly $6 billion. The companies have not confirmed a transaction, and The Next Web reports that negotiations could still collapse. Even at the discussion stage, the proposed price puts a spotlight on a less visible prize in the AI race: software that helps each expensive chip do more work.
Decart is not principally a competing chatbot maker. Its infrastructure work aims to improve the performance of AI applications across hardware, while its public products also include generative video and world-model systems. That makes the target relevant to an AI lab whose rivals are debating how frontier intelligence should be deployed as compute demand becomes a central business constraint.
Why Decart Matters | More Work From Existing Chips
Training and serving large models require vast quantities of accelerated computing. Improvements to the software layer can reduce the hardware, power, and time needed for a given workload. Livemint, citing Reuters reporting, said a Decart deal could give Anthropic additional infrastructure expertise as it scales resources for Claude.
That focus has become more valuable as lower-cost models put pressure on the economics of leading labs. The market shift is visible in the rise of open-weight Chinese models competing on price. For a model provider, lower inference costs can support better margins, more competitive pricing, or both.
Nvidia's Investment | Strategic Interest in Infrastructure
Decart announced a $300 million funding round led by Radical Ventures in May, with Nvidia participating as a new investor. The company had been valued at about $4 billion after that financing, according to The Next Web. A $6 billion acquisition would therefore represent a material premium for a startup founded in 2023 by Dean Leitersdorf, Orian Leitersdorf, and Moshe Shalev.
Industry reporting also described preliminary interest from Nvidia before higher offers emerged from other companies. Those reports do not establish a bid or a formal sale process, but they illustrate why infrastructure talent has become strategically contested. Cloud operators need more capacity, chip companies want software that strengthens their platforms, and model developers want to curb the cost of every generated response.
Anthropic's IPO Logic | Margin Control Before Public Markets
The reported transaction comes as Anthropic expands access to computing capacity and prepares for a potential initial public offering. The company recently signed a multi-year GPU arrangement with CoreWeave to operate Claude at production scale, according to The Next Web's report on the capacity agreement. Securing capacity and making that capacity more efficient are complementary strategies.
For prospective public-market investors, the key question is whether rapid revenue growth can outpace the cost of running frontier models. A proprietary optimization layer would not end Anthropic's dependence on outside hardware, but it could give the company more leverage over inference margins. That financial discipline matters in a market shaped by debates over low-cost AI competition, open-model strategy, and the capital requirements of ever-larger systems.