One of the most valuable startups on earth is about to spend billions not on a smarter model, but on a way to make the models it already has cost less to run.

Anthropic rarely buys anything. So when word broke on August 13 that the maker of Claude was in talks to acquire the Israeli startup Decart AI for about $6 billion, the number was not the only surprising part. The target was. Decart is not a rival lab racing to a smarter chatbot. It builds two things: "world models" that try to simulate physical reality, and — more quietly — software that squeezes more work out of the same silicon by helping AI chips run more efficiently. According to Fortune, citing Bloomberg's reporting, it is that second, less glamorous capability that Anthropic wants most.

If the deal closes — and the people describing it caution that talks could still fall through — it would be Anthropic's largest known acquisition, landing just as the company prepares for a hotly anticipated public offering. The Decart team, per that same reporting, would fold into Anthropic's inference and performance organization, the group responsible for the unsexy but decisive question of how much it costs to answer each user's prompt. That placement tells you what this is really about.

What the Anthropic Decart acquisition actually buys

Start with the mechanics, because they explain the price. Every time you send a message to Claude, a chip somewhere runs a calculation and bills Anthropic for the electricity and the hardware time it took. Multiply that by the surging volume of an assistant now embedded in enterprises, coding tools, and consumer apps, and the cost of "inference" — running the model, as opposed to training it — becomes the dominant line item on the bill. Frontier labs have spent three years competing to build the smartest model. The next phase of the war is being fought over who can serve that intelligence most cheaply.

Decart sits precisely on that seam. Founded in 2023 by three Israeli engineers — brothers Dean and Orian Leitersdorf and Moshe Shalev — the company trains its systems on text plus millions of hours of video to learn how the physical world behaves. Its consumer-facing Lucy model can take a live video feed of a person and render, in real time and high resolution, that person appearing to try on a dress or a bag, solving a problem that long frustrated fashion e-commerce, where realistic fabric has always looked fake. The e-commerce platform eBay is both an investor and a customer. The technical feat underneath Lucy — modifying a live video stream instantly, without lag — is what one person close to the talks called a signal of rare infrastructure talent, according to Fortune's account.

That talent, not the try-on demo, is the asset. Anthropic is not buying a virtual fitting room. It is buying a team that has proven it can make expensive computation run fast and cheap, and pointing that team at its own inference stack.

The valuation math frames the urgency. Decart raised $300 million in May, in a round led by Radical Ventures with Nvidia, Adobe Ventures, Atreides Management, and Valor Equity Partners joining, alongside earlier backers Sequoia, Benchmark, and Zeev Ventures. That round valued the company at almost $4 billion, as the Wall Street Journal reported, up from $3.1 billion just nine months earlier. A $6 billion price is roughly a 50% premium on a valuation that was itself only weeks old. Anthropic is paying up, and paying fast.

The margin problem hiding behind the growth story

To understand why efficiency is suddenly worth $6 billion, look at Anthropic's own numbers. The company is growing at a pace that strains belief. According to CNBC, Anthropic was on track for roughly $10.9 billion in revenue in the second quarter of 2026 — more than double the $4.8 billion it posted in the first — and a first operating profit in the neighborhood of $559 million.

Read that again. Revenue doubled in a single quarter, and the company barely edged into the black. That is the tell. When your top line is compounding this fast and your operating margin is still razor-thin, the constraint on profitability is not demand. It is the cost of every unit you sell. In software, the classic dream is near-zero marginal cost. In frontier AI, marginal cost is the enemy: each additional query burns real compute, and the smartest models burn the most. A lab that can shave even a fraction off inference cost does not just save money — it changes whether the entire business model works at IPO scrutiny.

Buying Decart is a direct assault on that constraint. If Anthropic's existing data-center footprint can be made to absorb more demand without proportionally more spending, the company defends its margins precisely when public investors will start demanding them. The Decart deal is, in that light, less an acquisition than a hedge against its own success.

What it means for each side of the table

For Anthropic, the logic is vertical integration under margin pressure. It has spent heavily — alongside OpenAI, it has committed to tens and potentially hundreds of billions in data-center capacity, per Fortune. Owning a team that makes that capacity go further is cheaper than renting the same efficiency indefinitely, and it keeps a scarce skill in-house rather than in a rival's. The world-model technology is a bonus optionality: a foothold in simulation, robotics, and video that Anthropic has not seriously staked out.

For Decart, a $6 billion exit into a soon-to-be-public buyer is a clean, rich outcome in a market where world-model startups are plentiful and durable moats are scarce. Its founders spoke in July at the Raise AI conference in Paris about reinventing "every different aspect of the economy in the next 18 to 24 months," in CEO Dean Leitersdorf's words to Bloomberg. Selling now trades that ambition for certainty — and for the compute of a partner that can actually deploy the tech at scale. Notably, Calcalist reported Decart had also been in talks with SpaceX, a claim Elon Musk dismissed on X as "fake news."

For Nvidia, there is a quiet irony. Nvidia backed Decart in May, and Decart's whole pitch is helping chips do more with less — which, taken to its logical end, means buyers needing fewer of Nvidia's chips per unit of output. The chipmaker profits from the deal as an investor while funding the very efficiency that could soften demand for its hardware. That tension will define the next stage of the AI economy.

For every other lab and enterprise buyer, the message is that the competitive frontier is moving. The bragging rights of a higher benchmark score matter less when the market takes model quality as roughly given and starts asking who can deliver it profitably. Inference efficiency is becoming the new battleground, and the labs are buying their way to it.

The takeaways for operators

Three lessons carry beyond this one deal.

First, in any business built on a per-use cost, the moment growth outruns margin, the smartest capital goes toward the cost side, not the demand side. Anthropic could have spent $6 billion on more chips or more researchers. It reportedly chose to spend it on making what it already has cheaper to run. When you cannot easily raise price, you defend the business by lowering unit cost — and you are often willing to pay a premium to do it quickly.

Second, the most valuable acquisition is sometimes the least visible one. The headlines will fixate on Lucy's virtual dressing room. The balance sheet cares about the inference stack. Read acquisitions by where the team lands, not by what the demo shows.

Third, the higher the promise, the higher the burn — and eventually someone has to answer for the burn. Anthropic is buying discipline before Wall Street forces it on them.

The bigger picture

For three years the story of AI has been about scale: bigger models, bigger clusters, bigger rounds. The Decart talks hint at the chapter after scale, where the question shifts from "how smart can we make it" to "how cheaply can we serve it." That is the arc every transformative technology follows — railroads, electricity, cloud computing — from a land grab to an efficiency grind, from building the thing to affording the thing.

Anthropic's reported bet says the grind is starting now, before the IPO bell, before the models even stop improving. The company that wins the next phase may not be the one with the cleverest model. It will be the one that can run intelligence at the lowest cost per token — and quietly bought the team that knew how, back when everyone else was still counting parameters.

Frequently Asked Questions

What is Anthropic reportedly acquiring Decart for?

According to Fortune, citing Bloomberg's reporting, Anthropic is in talks to buy the Israeli AI startup Decart for about $6 billion. The talks were not finalized as of August 13, 2026, and could still fall through. It would be Anthropic's largest known acquisition.

Why does Anthropic want Decart?

Decart builds "world models" that simulate physical reality and, more importantly for Anthropic, software that helps AI chips run more efficiently. That efficiency technology could let Anthropic's existing infrastructure absorb more demand and lower its cost of running Claude, according to Fortune's reporting. The Decart team would join Anthropic's inference and performance organization.

How does the $6 billion price compare to Decart's valuation?

Decart raised $300 million in May 2026 at a valuation of almost $4 billion, up from $3.1 billion in August 2025, per Wall Street Journal reporting cited by Fortune. A $6 billion acquisition price represents roughly a 50% premium over that recent round.

What does the Anthropic Decart acquisition say about the AI industry?

It signals that competition among frontier AI labs is shifting from building the smartest model toward serving intelligence at the lowest cost. As Anthropic prepares for an IPO with revenue doubling quarter over quarter but thin operating margins, controlling inference cost — not just model quality — is becoming the decisive advantage.

Editor's note — sources: Fortune / Bloomberg on the Decart talks, valuation, founders, and Lucy model; CNBC on Anthropic's Q2 2026 revenue and first operating profit. Decart's May round valuation is per Wall Street Journal reporting as cited by Fortune. Deal terms are as reported and not final.

Share this article: