Everything About AI
NextFin News - Anthropic has walked away from talks to acquire artificial intelligence startup Decart AI in a deal that would have been valued at about $6 billion, people familiar with the matter said on Tuesday, ending a courtship of just under four weeks that would have been the Claude maker's largest acquisition ever. The collapse comes as Anthropic positions itself for a public listing after a confidential IPO filing in June and a $65 billion funding round that valued the company at $965 billion.
The reversal is the clearest sign yet that the breakneck merger-and-acquisition pace inside frontier artificial intelligence is running into a new constraint: the discipline of an imminent stock-market debut. A roughly $6 billion price tag — about 50% above the nearly $4 billion valuation Decart secured in a May funding round — would have been the kind of premium that public-market investors dissect line by line in a prospectus. Walking away costs Anthropic little in cash; walking away with the deal signed could have cost it credibility at a sensitive moment.
The Deal That Was, and the Technology at Stake
When the talks first surfaced in mid-August, the logic was straightforward. Decart builds software designed to make AI chips run more efficiently, lowering the cost of training and operating large models. For a company spending tens of billions of dollars a year on data centers and processors, that capability is not a nice-to-have — it is a direct lever on the single biggest line item standing between rapid growth and profitability.
Under the proposed arrangement, Decart's team would have joined Anthropic's inference and performance organization, according to people familiar with the negotiations. The startup, founded in 2023 by former members of Israel's Unit 8200 intelligence unit, employs roughly 100 people across Tel Aviv, San Francisco and New York. Its product stack spans three lines: DOS, an inference and training stack built for agents and reasoning models; Lucy, a world model for immersive experiences; and Oasis, a world model aimed at physical AI and robotics.
Decart had been re-rated quickly. It raised $300 million in May at a valuation near $4 billion, up from $3.1 billion in August 2025, with Radical Ventures leading the round and Nvidia joining as a new investor alongside Sequoia Capital, Benchmark, Adobe Ventures and individual backers including former Nvidia scientist Andrej Karpathy. A $6 billion acquisition would have handed early shareholders a roughly 50% premium on the most recent mark and, according to Israeli business coverage of the deal, created at least two new billionaires on paper.
But the company's revenue story was less polished than its valuation. Reports covering the Israeli technology sector noted that Decart's graphics and gaming models struggled to generate stable revenue, forcing frequent product shifts, while the company spent tens of millions of dollars on computing resources and server space. That gap between a soaring private-market valuation and an unproven commercial track record is precisely the kind of tension that intensifies under IPO scrutiny.
"Once systems can understand and operate within the physical world, the range of what becomes possible expands dramatically — from robotics and autonomous systems to entirely new forms of commerce and live experience," Decart co-founder and chief executive Dean Leitersdorf said in May when announcing the Series B round. "The most immediate test is what happens when you put it in front of a game designer, an advertiser or an engineer building a robot — and watch what they reach for."
Neither Anthropic nor Decart commented on the end of the talks. Anthropic declined to comment when the deal was first reported; Decart did not respond to requests for comment at the time.
Why the Math Changed: The IPO Effect on AI M&A
The most important number in this story is not $6 billion. It is $965 billion — and the revenue trajectory behind it.
On June 1, Anthropic confidentially submitted a draft registration statement on Form S-1 to the Securities and Exchange Commission for a proposed initial public offering. "This gives us the option to go public after the SEC completes its review," the company said in a public announcement. Days earlier it had closed a $65 billion Series H round at a $965 billion post-money valuation, surpassing OpenAI to become the world's most valuable private artificial intelligence company.
The growth underneath that valuation is what makes a $6 billion acquisition suddenly look expensive in disclosure terms. The company's annualized revenue run-rate has climbed from about $9 billion at the end of 2025 to more than $47 billion in May and topped $65 billion by the end of July, according to a person familiar with the matter. In the second quarter, revenue exceeded $11.5 billion, up from $4.73 billion in the first quarter, and the company posted positive adjusted operating income, according to preliminary financial information reviewed in reporting on the company — figures the sources cautioned were preliminary and could still be revised.
That sequence changes the cost-benefit analysis of a splashy acquisition in a subtle but powerful way. As a private company racing toward a listing, Anthropic's every move is now being underwritten by the market it hopes to enter. A $6 billion deal — 0.6% of its own valuation, small in relative terms — would still appear in the prospectus as a scrutiny magnet: What was paid? What was acquired? What revenue backs it? How much of the price is goodwill? For a company that has only just turned corner on adjusted profitability, adding a large, hard-to-value intangible asset weeks before an expected autumn debut invites questions management may not want to answer.
There is also a timing dimension. With the IPO potentially weeks away — reports have pointed to a September or October window — closing a complex cross-border acquisition involving Israeli tax considerations, employee equity rollovers and integration into a U.S. parent adds execution risk at the worst possible moment. A delayed or renegotiated deal during the SEC review window would be a distraction; a collapsed deal after signing would be worse. Walking away now is the cleanest outcome available.
This is the cyclical read of the event, and it is the stronger one in the near term. M&A among frontier AI labs has been a function of abundant private capital and scarce assets. When the capital is still abundant but the clock changes — from "grow at any cost" to "show the public market a defensible path" — deal appetite contracts first at the margin, where premiums are richest and revenue is thinnest. Decart, with its 50% premium and unproven revenue, sat exactly at that margin.
The Structural Read: The Chip-Efficiency Problem Has Not Gone Away
It would be a mistake to read the walk-away as a rejection of Decart's technology. The underlying problem Anthropic was trying to buy its way out of is structural, and it is not going anywhere.
Training and running frontier models is, at its core, a compute problem. Every percentage point of efficiency gain in how well software utilizes graphics processors, tensor processing units or custom accelerators translates directly into lower capital expenditure or higher margin. Anthropic has been assembling a diversified compute portfolio to manage this: long-term arrangements for Amazon Web Services' custom Trainium chips, plans to use Google's tensor processors, a compute deal with SpaceX, and reported discussions to lease capacity from Meta. It is also building a 15,000-square-meter development center in London employing 200 people.
Buying a chip-efficiency specialist is the fastest route to marginal gains that would otherwise require years of in-house systems research. The strategic logic was sound. What failed was the price and the timing — not the thesis. That distinction matters for the rest of the sector. Nvidia, which had held advanced talks to acquire Decart before halting that process after Anthropic's offer emerged, remains a buyer in the space. Amazon and others have surfaced as potential acquirers. The demand for efficiency technology is structural; the willingness to pay a 50% premium for it three weeks before an IPO is cyclical.
There is a second-order implication here that the market has not fully priced. The same companies walking away from acquisitions are the same ones committing tens of billions of dollars to data centers. That is not a contradiction — it is a ranking of control. Capital expenditure on your own infrastructure is an asset you own, depreciate on your terms, and can point to in a prospectus as a moat. A $6 billion acquisition of a pre-revenue software shop is an intangible that auditors and analysts will probe. When capital is cheap and the listing clock is distant, M&A wins. When the clock is close, owned capex wins. The money is not leaving the AI build-out; it is being redirected toward assets that survive disclosure.
The Counter-Thesis: Maybe This Is Just Noise
The strongest case against reading too much into the collapse is the simplest: talks fall apart all the time, especially at this stage of negotiation. Due diligence may have surfaced technical concerns, valuation gaps, or integration issues that have nothing to do with IPO positioning. Decart had multiple suitors; Anthropic may simply have decided the technology was not as differentiated as hoped, or that building it in-house would be cheaper. The company has historically been selective about large acquisitions, and a fast deal cadence alone raises the probability of at least one false start.
This counter-thesis has real force. One collapsed negotiation does not establish a regime change in AI M&A. If anything, the continued interest from Nvidia and others suggests the asset class remains hot, and a different buyer could still pay a premium. The cyclical-vs-structural call could be wrong in both directions: the pullback could be narrower than argued (just this deal), or broader (a genuine cooling across the sector).
The signal that would prove the "IPO discipline" thesis wrong is specific and observable: if Anthropic, or a peer preparing for a public listing such as OpenAI, signs a comparable chip-efficiency or infrastructure-software acquisition at a premium of 40% or more above the target's last private valuation within the next six months — and does so without materially revising its own IPO timeline — then the walk-away was deal-specific noise, not a structural shift in pre-IPO behavior. Conversely, if two or more pre-IPO frontier labs pass on premium acquisitions in that window while continuing to announce billion-dollar capex commitments, the discipline read is confirmed.
What Comes Next: Three Horizons
Short term (weeks): Expect quiet. Anthropic is unlikely to announce a replacement acquisition before its IPO, and Decart will likely return to a sale process or fundraising with a narrower set of buyers. The immediate market signal is muted because both companies are private — there is no ticker to move. The affected parties are the startup's employees and investors, who now face a reset of exit expectations after a year of rapid re-rating.
Medium term (the IPO window): Watch the S-1 as it becomes public. The prospectus will reveal Anthropic's capital-expenditure trajectory, its profitability path, and any mention of in-sourcing the efficiency capabilities it sought to buy. If capex guidance rises while acquisition activity stays dormant, the redirection thesis holds. Base case: Anthropic proceeds with a listing in the autumn at a valuation in the high hundreds of billions, having substituted owned compute partnerships for bolt-on M&A. Upside case: strong enterprise demand and a calm market lift the debut toward the roughly $2 trillion range that has been discussed in private-market conversations. Downside case: weak public-market appetite for unprofitable growth or a broader AI valuation reset forces a delay or a lower valuation.
Long term (structural): The chip-efficiency arms race continues regardless of who owns Decart. Whether through acquisition, in-house development, or partnerships with the chipmakers themselves, every frontier lab will keep chasing marginal efficiency gains because the economics of inference demand it. The winner in this race is not necessarily the company that buys the best startup; it is the one that integrates efficiency into its stack fastest and can show public investors a credible path from capex to margin.
The deal that did not happen tells you more about the market than the deal that did. Anthropic is not stepping back from the AI build-out — it is stepping back from a price, at a moment when the bill for that price would soon be payable in public scrutiny. The capex keeps flowing; only the shortcuts are being closed.
Explore more exclusive insights at nextfin.ai.
