Anthropic Reportedly Pursues $6 Billion Decart Acquisition to Expand World Models and Compute Efficiency

Anthropic is in talks to acquire Israeli artificial intelligence startup Decart for about $6 billion, according to people familiar with the matter cited by Bloomberg and Reuters. The discussions have not been publicly confirmed by either company, and the reported valuation would rank among the largest acquisitions in the AI sector.

The talks place Decart at the center of an accelerating race among AI labs to secure specialized technology rather than build every capability internally. Israeli outlets Jpost and Haaretz also described the target as an Israeli startup, while The Edge Malaysia repeated Bloomberg’s account of a potential US$6 billion transaction.

Decart’s technology is associated with so-called world models, systems designed to represent how environments, objects and actions behave over time. That work has applications in video generation and simulation, while Bloomberg Intelligence analyst Matthew Bloxham said the approach could make computing hardware more effective and reduce AI training costs.

Tradingkey’s headline linked the reported acquisition to video generation and chip optimization, while Stocktwits framed the talks within the broader search for cheaper and faster compute. Those capabilities would give Anthropic a way to expand beyond language-model development and strengthen the infrastructure economics behind its products.

The reported interest reflects a basic constraint in frontier AI: training and serving advanced models require large amounts of specialized computing capacity. Technology that extracts more useful work from each chip can reduce costs, improve throughput or make larger experiments practical within the same hardware budget.

For Anthropic, compute efficiency would matter across model training, inference and new multimodal products. Bloomberg’s account connected Decart’s world models with more effective chip use, while Tradingview and Ultrathink presented the talks as part of a competitive push involving major AI labs, including a challenge to OpenAI.

A $6 billion purchase would signal that Anthropic views Decart’s capabilities as strategically important enough to justify a large premium. The deal would also show how competition among AI companies is widening from model quality to control over data, simulation, video systems and the cost of computation.

Anthropic has become one of the leading providers of general-purpose AI models, but it competes with companies that have deeper distribution, larger infrastructure budgets or broader product portfolios. Ultrathink explicitly linked the potential transaction to OpenAI, while the reported focus on compute economics points to pressure shared across the sector.

The reported transaction remains unverified, and talks can change or end without an agreement. The $6 billion figure is described as an approximate value, with no disclosed structure, financing terms or timetable for a possible announcement.

Important questions include whether Decart’s technology would be integrated into Anthropic’s core models, developed as a separate product or used primarily for internal research. The limited public details also leave the commercial contribution of video generation and chip optimization unclear. Reuters and Bloomberg supplied the central account, while the other coverage largely repeated or interpreted that report.

The next concrete development would be a formal announcement, confirmation that negotiations have ended or evidence of a revised price. Until then, the reported talks serve as a signal of where AI investment is moving: toward systems that model the physical or digital world and technologies that reduce the cost of operating frontier models.

If completed, the acquisition would give Anthropic a larger foothold in world models and video-related AI while adding a potentially important compute-efficiency capability. It would also place Decart’s technology inside one of the industry’s most closely watched AI labs, with consequences for competitors, chip suppliers and startups working on simulation and multimodal systems.