Anthropic's $6 billion acquisition talks for Decart AI

Decart's public identity rests on Lucy, a real-time video model ^[2], and Oasis, which simulates 3D training environments for robotics and self-driving systems ^[7]. But neither product may be the real point of the deal. The piece Anthropic is reportedly after is the Decart Optimization Stack, or DOS, software that raises chip utilization across Nvidia GPUs, Amazon's Trainium, and Google's TPUs ^[2]- and it's this team that would fold into Anthropic's inference and performance organization if the deal closes ^[3]. That integration pattern, closer to an acqui-hire than a product acquisition, echoes how Anthropic absorbed smaller teams like Humanloop and Bun earlier in 2026 ^[3].

The engineers behind DOS carry an unusual resume. Co-founders Dean Leitersdorf, his brother Orian, and Moshe Shalev are all veterans of Israel's Unit 8200 military intelligence unit, and Dean earned his Technion PhD at 23 - the youngest in the university's history - with a dissertation that won the ACM PODC Dissertation Award ^[4]. For a company about to answer to public shareholders, that pedigree matters as much as any demo reel: one analyst framing of the deal credits the team with having 'demonstrated 8x throughput gains at commercial scale' ^[1], meaning Anthropic isn't betting on a promise, it's buying a result that's already been shown to work.

Strip away the sticker price and the numbers argue Anthropic is purchasing capacity, not consumer-facing products. Decart's DOS reportedly pushes Model FLOPS Utilization above 80% on Amazon Trainium chips, against an industry norm of 40-50% ^[1]. Analysts have described that gain as operationally equivalent to buying $10 billion to $15 billion in additional hardware without buying any additional hardware ^[1]- a meaningful offset against a 2026 compute bill estimated around $19 billion ^[1].

Set that against Anthropic's other 2026 infrastructure commitments - roughly $30 billion of Microsoft Azure capacity, a $50 billion data-center buildout with Fluidstack, and up to 5 gigawatts of additional Amazon compute under a $100-billion-plus AWS deal ^[6]- and a $6 billion outlay for efficiency software looks like the cheaper lever, not the flashier one. It also lines up with surging demand for Claude that Anthropic's existing infrastructure is straining to absorb ^[5], which is the practical reason a compute-constrained lab would rather buy software that stretches its current chips than simply order more of them.

Not everyone reads this as disciplined infrastructure economics. Decart markets Lucy as a real-time video model that can alter what a camera captures, aimed at uses like e-commerce, advertising, live streaming, and gaming ^[2], and Oasis as a generator of synthetic 3D training environments for robotics and autonomous-vehicle systems ^[7]. That framing drew open pushback in AI-skeptic corners of Reddit, where the debate centered on whether 'world models' is a rigorous technical category at all - invoking Yann LeCun's JEPA work as a counterpoint to what critics dismissed as generative video filler - and treated the acquisition as another data point for broader AI-industry excess.

That tension is worth sitting with rather than resolving: the efficiency story Anthropic and its analysts are telling public-market investors is a very different story from the one a technically skeptical community is telling about the same products. It's also a reminder that the deal's headline assets (Lucy, Oasis) and its actual stated purpose (the optimization stack) are, by Anthropic's own reported priorities, not the same thing - which is exactly the gap the skeptics are pointing at.

If completed, the roughly $6 billion price tag would make Decart Anthropic's largest acquisition to date and its fifth of 2026 ^[5]- a striking pace for a company that filed a confidential S-1 with the SEC in June and could go public as early as next month or early October ^[6]. The price also represents roughly a 50% premium over the near-$4 billion valuation Decart commanded just three months earlier in its Series B ^[8], a jump that reflects both Decart's own momentum and how much strategic value Anthropic is willing to pay to secure it before a public listing invites more scrutiny of every dollar spent.

None of it is locked in - multiple outlets caution the talks could still collapse ^[5]- but the size of the premium and the pace of the pursuit together signal how urgently Anthropic wants this efficiency story locked down before its IPO roadshow begins.