Deal Dead at the Diligence Table: Anthropic Walks Away From Its $6 Billion Decart Buyout

On the morning of September 8, 2026, Bloomberg reported that Anthropic PBC has decided against pursuing its acquisition of Israeli AI startup Decart AI. On paper, it is a two-line story: a buyer walked away from a deal. In context, it is one of the most telling M&A data points the AI industry has produced this year — the collapse of what would have been the Claude maker’s largest acquisition ever, killed at the last stage before paperwork: due diligence.

What happened

According to people familiar with the matter, Anthropic had been exploring a purchase of Decart for roughly $6 billion, performed full due diligence on the company, and ultimately walked away. Both companies declined to comment on the talks, and Bloomberg notes the companies could still pursue other collaboration opportunities — a diplomatic caveat that leaves the door open to a licensing deal or partnership without a change of control.

The numbers explain why this deal mattered. Decart last raised money in May 2026: a $300 million round led by Radical Ventures at a valuation of nearly $4 billion, up from $3.1 billion in August 2025. Anthropic’s offer therefore implied a premium of roughly 50% over the most recent mark — a generous price by any standard, and a signal of how badly Anthropic wanted what Decart had built.

Who Decart is — and why Anthropic wanted it

Decart, founded in 2023 by Israeli brothers Dean and Orian Leitersdorf together with Moshe Shalev, occupies a strategically uncomfortable seat in the AI stack: it sits between the models and the GPUs. The company’s core product line is an inference-optimization engine — software that extracts dramatically better latency, throughput, and total cost of ownership from existing hardware, including Nvidia GPUs, AWS Trainium, and other accelerators. Reporting around the deal talks cited figures on the order of an 8x improvement in AI inference throughput, and a claim that the technology could push gross margins toward 77% for a buyer that operates its own models at scale.

For Anthropic, the attraction is straightforward. The company has been ramping up spending on computing capacity as it develops products and prepares for a potential IPO. Claude Code and agentic products have driven skyrocketing inference demand, and Anthropic has scrambled to line up cloud computing deals with SpaceX, Google, and others to keep up. Buying the team that knows how to squeeze more tokens per dollar out of every accelerator is the kind of vertical integration that shows up directly in gross margin — the single most scrutinized line item in any AI lab’s IPO filing.

Under the plan that died this week, Decart’s team would have joined Anthropic’s inference and performance organization. It would have been Anthropic’s fifth acquisition of 2026, and by a wide margin its largest.

Why the deal died

Neither side is saying what the diligence turned up, and that silence is itself informative. The reporting that has emerged points to a cluster of plausible factors rather than a single smoking gun.

First, the structure. The offer was reportedly built primarily in stock rather than cash. For Decart’s shareholders, that means their payout was really a leveraged bet on Anthropic’s IPO — and the IPO timing has been sliding, with the marketing window recently slipping into mid-October. A stock-heavy deal against a moving IPO date makes everyone’s risk models nervous.

Second, the dilution math. A pre-IPO company issuing billions in stock to fund an acquisition invites a straightforward question from prospective public-market investors: does this deal add more value than the dilution subtracts? If diligence could not answer that question convincingly, walking away is the disciplined outcome, not a failure.

Third, the competitive backdrop is stranger than it looks. Nvidia — already on Decart’s cap table as an investor, alongside Atreides Management, Valor Equity Partners, Adobe Ventures, and eBay (which is also a customer) — had reportedly tabled an earlier bid that was deemed more financially attractive. Decart’s founders passed on it in favor of Anthropic’s offer. With Anthropic now out, Nvidia gets a second shot at a company it evidently already wanted, possibly at a lower price.

What it means for the market

The uncomfortable lesson for every AI startup negotiating an acquisition right now: the ‘agreed’ stage is no longer the finish line. When a marquee buyer kills a deal at a ~50% premium after diligence — not before it — the market has to reprice the probability that any announced AI M&A actually closes. Targets will demand stronger protections, and buyers will quietly discount the value of exclusivity during the diligence window.

For Anthropic, the walk-away is a statement of discipline at a delicate moment. The company is spending aggressively on compute and heading toward a public filing with a mid-October marketing window. Abandoning its largest-ever deal rather than forcing it through suggests management believes the IPO story is stronger without Decart inside it — either because of what diligence found, or because of what the acquisition would have cost the equity story.

And for Decart, this is not a death sentence but a reset. The company is capitalized (over $680 million raised to date), backed by strategic investors, and operates in the one layer of the stack where demand keeps outrunning supply: making expensive silicon do more work. A second Nvidia bid below $6 billion would confirm that Anthropic’s exit was a genuine re-pricing event rather than a timing hiccup — and would hand Nvidia the inference-optimization asset it tried to buy once already.

The bigger picture

Strip away the dollar figures and this story is about a maturing market. A year ago, the dominant AI acquisition narrative was speed: labs buying teams and technology at whatever price preserved momentum. The Decart collapse shows the pendulum swinging back toward ordinary financial gravity — diligence that can kill a deal, IPO math that constrains stock-based offers, and sellers who discover that a handshake at a 50% premium is worth exactly nothing until the wire clears.

That is, on balance, healthy. An industry where $6 billion deals die quietly at the diligence table is an industry whose capital allocators are still capable of saying no. The next data point to watch is Anthropic’s S-1: if a Decart collaboration appears in it, this was a restructuring. If nothing does, it was a clean kill — and Nvidia’s move, or lack of one, will tell us what Decart is really worth.