Add to Google Preferred Sources
Anthropic has decided against acquiring AI startup Decart AI after conducting due diligence, people familiar with the matter said. The two companies had been in talks over a deal valued at roughly $6 billion, first reported in August, though no agreement was ever finalized. The firms may still pursue other collaboration opportunities. Decart develops software that improves chip efficiency and lowers AI training and operating costs. Anthropic rarely makes large acquisitions and has instead focused spending on computing power ahead of a highly anticipated IPO, with people familiar saying the company aims to raise as much as or more than SpaceX. Anthropic has separately expanded into life sciences, acquiring Coefficient Bio earlier this year and launching tools such as Claude Science and the Model Hardware Standard.
Key Elements
Anthropic has decided not to go ahead with an acquisition of Decart AI, the startup developing software that makes chips run more efficiently, people familiar with the discussions said. The decision ends, for now, what would have been the largest acquisition in Anthropic’s history, valued at roughly $6 billion.
The Claude chatbot maker had been examining a purchase of Decart and completed a due diligence review before ultimately stepping away from the transaction, according to the people, who asked not to be identified because the matter is private. The two companies could still explore other forms of collaboration, one of them said. Representatives for both Anthropic and Decart declined to comment.
News of the possible acquisition first surfaced in August, when Bloomberg News reported that Anthropic was in talks to buy Decart for about $6 billion. At the time, no agreement had been reached, and people close to the situation cautioned that negotiations could still collapse. The latest development confirms those earlier warnings.
Decart, an Israeli-founded startup, builds software that helps chips operate more efficiently, which in turn lowers the cost of training and running AI models. That capability has made it an attractive target for companies racing to control the infrastructure costs that define the economics of modern AI.
Founded in 2023 by Dean Leitersdorf and Moshe Shalev, the company employs about 100 people and develops models of its own alongside the optimization work. Lucy edits live video in real time, while Oasis generates simulated environments used to train and test robotics and autonomous-driving systems. The optimization layer is built to extract more performance from a wide range of chips rather than a single vendor, which is why the company drew interest from chipmakers and cloud providers alike.
The price under discussion represented a steep escalation. Decart raised $300 million in May at a valuation of about $4 billion, a round that brought Nvidia in as a new investor and lifted total funding to roughly $450 million. A purchase at $6 billion would have marked the company up by half in the space of three months. Calcalist reported that Nvidia had itself held advanced talks to buy Decart before those discussions collapsed when higher offers emerged.
| Financing step | Amount raised | Valuation |
|---|---|---|
| First round (Sequoia Capital) | $21 million | Not disclosed |
| Follow-on, one month later | $25 million | About $500 million |
| Later round | Not disclosed | $3.1 billion |
| May 2026 (Radical Ventures) | $300 million | About $4 billion |
| August 2026 talks with Anthropic | Not applicable | About $6 billion |
Note: Total funding reached roughly $450 million before the takeover discussions began. The $6 billion figure was the level under negotiation and was never agreed.
For Anthropic, the decision to walk away aligns with a broader pattern. The company rarely pursues large acquisitions, preferring instead to pour capital directly into computing power. That spending supports product development and customer service as the firm prepares for a Wall Street debut that has generated significant anticipation across the <a href="https://bitcomme.com/hang-feng-technology-innovation-co-ltd-announces-first-half-2026-financial-results/” title=”Hang Feng Technology Innovation Co., Ltd. Announces First Half 2026 Financial Results”>technology industry.
Its dealmaking has tended instead toward small, team-focused transactions. In August 2025 the company took on most of the Humanloop team without acquiring that startup’s assets or intellectual property, a distinction Anthropic confirmed at the time.
People familiar with Anthropic’s IPO preparations have said the company aims to raise an amount comparable to or exceeding what SpaceX pulled in through its own public listing. That ambition underscores the scale of the resources Anthropic expects to need as competition intensifies among frontier AI developers.
That is a demanding benchmark. SpaceX sold 555.6 million shares at $135 each in June, raising $75 billion in the largest initial public offering on record and valuing the rocket company at $1.77 trillion, according to its filing with the Securities and Exchange Commission. The total climbed to roughly $86 billion once underwriters took up their additional allotment. Anthropic itself was valued at $965 billion in a $65 billion funding round completed in May, and filed confidential listing paperwork in June.
The outcome carries weight well beyond Silicon Valley. Anthropic opened a Tokyo office in October 2025, its first in the Asia-Pacific region, and signed a memorandum with Japan’s AI Safety Institute covering cooperation on AI evaluation methods. The company has said its run-rate revenue across the region grew more than tenfold over the preceding year, and has pointed to Rakuten, Nomura Research Institute and Panasonic as Japanese customers building on Claude.
In April, NEC became the first Japanese company to join Anthropic as a global partner, agreeing to roll out Claude to about 30,000 group employees and to co-develop secure, industry-specific agents for finance, manufacturing and local government using the Claude Cowork desktop application. That work runs on Claude Opus 4.7 and the Claude Code agent inside NEC’s own transformation offerings. Commitments on that scale rest on the same computing capacity that Anthropic has been prioritizing over acquisitions.
The stalled Decart deal does not signal a retreat from dealmaking altogether. Anthropic has been quietly building out its presence in adjacent fields, particularly the life sciences. In April, the company reportedly acquired Coefficient Bio in a stock transaction valued at approximately $400 million. TechCrunch later confirmed the deal had closed, citing people close to the matter, though the exact figure was not verified.
Coefficient Bio was a young company, roughly eight months old at the time of the acquisition, with a team of about ten people. Its two founders previously worked at Genentech’s Prescient Design unit, where they focused on computational drug discovery. The team was expected to join Anthropic’s healthcare and life sciences division.
That acquisition reflected a growing interest among frontier AI companies in applying their models to biological research. Anthropic has continued to move in that direction through a series of product and infrastructure announcements.
In late June, the company launched Claude Science, a research workbench that lets scientists search literature, analyze data, and call specialized models within a single environment. The platform connects to Nvidia’s BioNeMo tools and supports models such as Evo 2, Boltz-2, and OpenFold3. The design mirrors the approach Claude has taken in software development: organizing scattered tools into one continuous workflow rather than rebuilding every component from scratch.
On Aug. 27, Anthropic introduced a research preview of the Model Hardware Standard, or MHS. The project began through a collaboration with the Howard Hughes Medical Institute’s Janelia Research Campus and aims to standardize how AI systems interact with laboratory equipment. Many labs still rely on devices controlled through outdated script interfaces or proprietary graphical tools, which makes it difficult for models to coordinate experiments across multiple machines.
MHS is designed to reduce that friction by providing a unified description and control interface. The standard is not tied exclusively to Claude, a deliberate choice that Anthropic hopes will encourage adoption by equipment makers and research institutions. The company’s broader strategy appears to be positioning Claude as the entry point through which researchers launch tasks and coordinate tools, even when those tools come from other providers.
A recent job posting from Anthropic signals that the company wants to accelerate this push. The listing describes a role focused on acquisitions, investments, and partnerships with AI-focused biotech companies. Target areas include drug discovery, clinical development, medical documentation, laboratory automation, and healthcare data infrastructure.
The posting asks candidates to distinguish between AI-native businesses with durable competitive advantages and companies that are simply wrappers around existing models. That requirement suggests Anthropic is looking for teams with mature experimental processes, proprietary data, and customer delivery capabilities rather than superficial AI integrations.
Still, the company faces meaningful hurdles in translating model capability into real-world laboratory results. Connecting equipment is only the first step. Automated experiments do not guarantee sound scientific conclusions, and issues such as sample variability, measurement error, and experimental design still require human judgment.
The collapse of the Decart talks leaves Anthropic without the chip-efficiency software it had been evaluating. But the company’s spending on computing infrastructure continues, and its expansion into life sciences shows no sign of slowing. Whether the two firms ultimately find another way to work together remains an open question, according to the people familiar with the matter.
For Decart, the end of acquisition discussions means continuing to operate independently at a moment when demand for AI infrastructure optimization is rising sharply. The startup’s technology addresses one of the industry’s most pressing problems: the enormous cost of training and serving large models.
Investors and competitors will be watching whether Anthropic revisits the deal or shifts its attention to other targets. The company’s IPO ambitions give it both the incentive and, potentially, the resources to pursue large transactions in the future. But for now, the $6 billion acquisition that briefly captured the AI world’s attention is off the table.
Once added, BigGo Finance appears first in Google Search Top Stories, so you get the broadest, most up-to-the-minute, and most comprehensive global financial news first.
