The courtship had the shape of a decisive move: an AI company with a war chest, a startup with technology that could cut the cost of running models, and a price tag near $6 billion. Then, according to Bloomberg, Anthropic completed its due diligence and walked away. The acquisition of Decart, an Israeli startup that builds software to make AI chips work harder, will not happen.
People familiar with the matter said the decision came after months of examination, and the reasons were not disclosed. What is clear is the scale of what was abandoned. Decart raised $300 million in May at a valuation near $4 billion, in a round led by Radical Ventures, and Anthropic’s offer represented a premium of roughly 50 percent. Had it closed, the deal would have ranked as the largest acquisition in the Claude developer’s history and its fifth of 2026.
The company Decart had built sits at one of the industry’s most crowded intersections. Its software targets the cost of training and running AI models, the expense that has come to define the economics of artificial intelligence. As models grow and usage spreads, the price of serving them has become as important as the cost of building them, and a startup that can squeeze more useful work from each processor has obvious appeal to any lab with a large computing bill.
That appeal explains the interest from Anthropic, which has made no secret of its hunger for computing efficiency. The company has signed enormous contracts for data-center capacity and has said that the cost of serving its models to a growing user base is one of the central constraints on its business. Acquiring Decart would have brought the optimization technology in-house, along with a team that has spent years studying where the waste is in modern AI systems.
Nvidia’s presence on Decart’s shareholder list added another layer to the story. The chip maker has invested broadly across the AI stack, and people familiar with the situation said Nvidia had offered Decart’s founders a more attractive price of its own, an approach the founders declined. Nvidia’s interest makes sense from its own perspective: a startup that makes its chips run more efficiently makes the hardware more valuable, and the company has been careful to keep such technology close.
The decision to abandon the deal is being read in the context of Anthropic’s path to an initial public offering. The company is widely expected to list its shares in the coming months, a process that subjects every major decision to new scrutiny. A $6 billion acquisition completed just before an IPO would have asked investors to value a business with a large new integration to manage, at a time when the market for AI stocks has shown it will punish complexity.
There is also the question of what the deal would have bought. Decart’s technology is real, but integrating a foreign startup into Anthropic’s model-serving infrastructure would have taken quarters, and the benefits would have shown up gradually rather than in time for the offering narrative. Walking away lets Anthropic keep its story clean: a company focused on its own models, its own infrastructure and its own growth, rather than a consolidator absorbing technology it could license instead.
The same logic has not stopped other AI companies from buying. The industry has seen a wave of acquisitions as the largest labs trade cash for talent and technology, and deals that would have seemed startling a year ago are now routine. What distinguishes Anthropic’s decision is the stage: the company is at the point where its balance sheet is about to become public, and every commitment is being weighed against how it will read in a prospectus.
For Decart, the outcome is a return to the market. The company retains its technology, its investors and its valuation, and the interest from two of the industry’s most important companies is itself a form of validation. Efficiency software for AI has not become less valuable because one buyer changed its mind; if anything, the attention has raised Decart’s profile among the other labs and cloud providers that face the same cost pressures.
The episode also reflects a broader shift in how AI companies think about efficiency. For most of the industry’s brief history, the winning strategy was to spend on scale: more chips, more data, more electricity, on the assumption that size defeated sophistication. That arithmetic has begun to change as serving costs climb and investors ask when the spending produces profit. Technologies like Decart’s, which extract more value from existing hardware, have moved from interesting to essential, which is why the company attracted attention from the industry’s biggest names in the first place.
The episode offers a glimpse of how the AI industry’s consolidation will work in practice. The appetite for acquisitions is real, and the prices are large, but so is the caution that comes with approaching public markets. Anthropic’s willingness to spend months on due diligence and then say no suggests that discipline is returning to a sector where deals were once announced with more enthusiasm than analysis. The company has made its largest purchase of the year only by deciding not to.
