Analysis
Anthropic is in talks to acquire Decart, an Israeli AI infrastructure startup, for roughly $6 billion -- what would be Anthropic's largest acquisition on record, according to Bloomberg. The talks are not finalized and could still fall apart.
The reported price already tells its own story: it's roughly 50% above the $4 billion valuation Pulse covered Decart setting in a $300 million raise this past June.
Decart was founded by Dean Leitersdorf, now CEO, and Moshe Shalev, chief product officer, both veterans of Israel's Unit 8200 intelligence corps -- a background shared by a disproportionate number of Israeli AI infrastructure founders. The company has raised roughly $450 million total, and its public identity has centered on real-time video generation and 'world model' research, the kind of interactive-simulation technology that gets compared to Google DeepMind's Genie or OpenAI's Sora. What acquirers are reportedly paying for in this deal, though, is a different part of Decart's stack: software that improves how efficiently AI chips are used during training and inference, letting existing GPU fleets absorb more workload without adding hardware.
“A $6 billion cash-or-stock acquisition this close to a public listing is a real balance-sheet and governance decision, not a routine tuck-in.”
## Why the price doesn't match the public product That distinction matters for how to read the price. Decart's closest public competitors in generative video -- Runway, Pika, and Luma AI -- have raised at valuations in the low-to-mid billions without attracting acquisition interest at this scale, suggesting the video product alone doesn't explain a $6 billion number.
Anthropic's own infrastructure needs are the more likely driver: the company is running on more than $47 billion in annualized revenue and is reportedly targeting a public listing that could value it near $2 trillion, and a lab growing that fast is almost certainly compute-constrained more often than it would like. Software that stretches an existing GPU allocation further is worth a premium to a company in that position, even if the acquired startup's headline product is unrelated.
This would be Anthropic's largest acquisition to date, in a year where AI-native acquirers have increasingly bought infrastructure and efficiency companies rather than pure research teams -- a shift from the 2024-2025 pattern of labs primarily acquiring talent and model IP. It also lands during an unusually sensitive stretch for Anthropic, which is simultaneously working with Morgan Stanley, Goldman Sachs and JPMorgan on IPO preparations. A $6 billion cash-or-stock acquisition this close to a public listing is a real balance-sheet and governance decision, not a routine tuck-in.
The bear case: talks falling through is a real possibility Bloomberg's own reporting flags, and Decart's most valuable technology -- efficient inference at scale -- is notoriously hard to verify from the outside until it's running in production on someone else's fleet. Paying a 50% premium for a claim that hasn't been stress-tested at Anthropic's scale is a bet, not a proven return.
Watch for confirmation of final deal terms and whether the structure is cash, stock, or a mix -- and whether Decart's leadership, including Leitersdorf, stay on to run the unit or exit after close, which will say a lot about whether Anthropic is buying the team or just the technology.
