Analysis
Bloomberg reported that Anthropic has ended talks to acquire Decart, an Israeli AI startup, for roughly $6 billion -- a deal that would have been Anthropic's largest acquisition to date. People familiar with the matter said the two sides had progressed deep into due diligence before Anthropic pulled out; the exact reason has not been disclosed, though Anthropic and Decart may still pursue a smaller commercial partnership instead of a full acquisition. Pulse has tracked Anthropic's spending and dealmaking closely all year -- this would have been its largest bet yet.
Founded by Israeli engineers Dean Leitersdorf, Orian Leitersdorf and Moshe Shalev, Decart's core product is Oasis, a real-time "world model" that generates interactive video -- effectively a playable game environment rendered frame-by-frame by a neural network at 20 frames per second with no perceptible latency. The company has separately built chip-efficiency technology that reduces the cost of running large AI workloads, which was reportedly the specific asset Anthropic wanted. Decart's funding history:
- Seed, Oct. 2024 -- $21M, led by Sequoia Capital, with Zeev Ventures
- Series A, 2025 -- $100M at a $3.1B valuation, adding Benchmark and Aleph VC
- Series B, May 2026 -- $300M led by Radical Ventures, with Adobe Ventures, Toyota Ventures and Valor Equity Partners, lifting the valuation to roughly $3.9B and total funding above $450M
“The company has separately built chip-efficiency technology that reduces the cost of running large AI workloads, which was reportedly the specific asset Anthropic wanted.”
Why Anthropic wanted this
Anthropic's economics are dominated by inference and training compute costs, and any technology that meaningfully improves utilization on existing GPU and TPU fleets is worth a premium -- which is presumably why Anthropic was willing to pay roughly 54% over Decart's last private mark. Decart isn't alone in that category: Cerebras and Groq compete on custom silicon for inference speed, while on the video-generation side that Decart's Oasis model also touches, Runway, Luma AI and Google's own Genie project are building comparable real-time world-model technology. Anthropic buying Decart would have combined a compute-efficiency asset with a video/world-model team in one deal -- unusual scope for an AI lab whose M&A history has otherwise skewed toward smaller, narrower acqui-hires.
The counterweight
It's tempting to read a diligence-stage collapse as a red flag on Decart specifically, but deals fall apart after diligence for mundane reasons too -- price renegotiation, key-employee retention terms, or overlapping technology Anthropic decided it could build in-house faster than it could integrate.
Anthropic itself just closed a $65 billion Series H and is preparing an IPO now expected in mid-October, so capital discipline on a cash-and-stock acquisition, even for a lab this well-funded, is not automatically a signal that something is broken at Decart.
What to watch
Decart's board now has to decide whether to run a fresh process with other potential acquirers -- Google, Meta and Nvidia have all been active buyers of AI infrastructure and video-generation talent this year -- or raise a Series C independently at a valuation that has to clear the $3.9 billion bar investors already paid in May. Either path tests whether the diligence findings that scared off Anthropic show up again with the next bidder.
