Anthropic in talks to buy Nvidia-backed AI startup Decart for $6 billion ahead of mega IPO

Anthropic is in talks to acquire Nvidia-backed AI startup Decart in a deal that could be worth roughly $6 billion, a rare acquisition push that could give the Claude maker new technology to squeeze more performance from its costly computing infrastructure as it prepares for a potential public listing, according to an exclusive report from Bloomberg.

“Anthropic PBC is in talks to buy the artificial intelligence startup Decart AI for about $6 billion,” Bloomberg reported, citing people familiar with the matter.

A source familiar with the discussions separately confirmed the talks to Reuters. The negotiations remain at an early stage, meaning a transaction is far from guaranteed.

The potential acquisition stands out for its size and timing. Anthropic has largely built its technology internally rather than pursuing multibillion-dollar takeovers. Buying Decart would signal a different approach at a moment when demand for Claude is putting increasing pressure on the company’s computing capacity.

Decart sits at an interesting intersection of those problems. The startup develops AI infrastructure and optimization software aimed at making chips operate more efficiently during AI workloads. It has built its own AI models, including technology focused on simulated environments and real-time video.

That combination could make Decart more valuable to Anthropic than a conventional model startup.

Decart raised $300 million in May in a round led by Radical Ventures, with Nvidia joining as a new investor. A roughly $6 billion acquisition would represent a significant jump from the valuation implied by a typical venture round and show just how much strategic value Anthropic may place on infrastructure efficiency.

Neither company is saying much publicly. An Anthropic spokesperson declined to comment, and Decart did not immediately respond to Reuters outside regular business hours.

Anthropic’s AI bottleneck is increasingly about compute

The bigger story behind the potential deal is Anthropic’s race to get more Claude capacity from an enormous and expensive computing footprint.

Training frontier AI models consumes vast amounts of computing resources, but serving those models to millions of users creates another challenge. Every Claude query requires inference capacity, turning efficiency gains at the chip and software level into potentially meaningful improvements in cost, latency and the number of customers Anthropic can serve.

Decart’s optimization technology could fit directly into that problem. A person familiar with the matter told Reuters that the startup’s technology could help Anthropic’s existing infrastructure absorb greater demand. If a transaction closes, Decart’s team is expected to join Anthropic’s inference and performance organization.

That would line up with another move Anthropic disclosed last week. The company is recruiting engineers with experience across hardware and software to help co-design custom chips and AI models that can make Claude run faster and more efficiently.

Decart brings another piece to the equation: world models.

Its Oasis model generates simulated environments that can be used to train and test robotics and autonomous-driving systems. Decart has also developed Lucy, an AI model capable of editing live video in real time, according to the company’s website.

World models have become an increasingly important research direction as AI companies push beyond text generation into systems capable of reasoning about physical environments. For Anthropic, Decart’s expertise could open research opportunities beyond Claude’s current strengths, though the immediate attraction appears closely tied to infrastructure and inference performance.

The reported talks arrive as Anthropic moves toward what could become one of the largest technology IPOs in years. Preparing for public markets puts a sharper spotlight on revenue growth, operating costs and the enormous capital requirements behind frontier AI.

That makes the proposed Decart deal particularly revealing. The AI race is no longer simply about who can train the smartest model. For companies operating at Anthropic’s scale, the ability to serve those models efficiently may become just as important.

A $6 billion acquisition would be an expensive way to attack that problem. It could also show how valuable every additional unit of AI compute has become.