Anthropic in Talks to Acquire Decart for $6bn
Anthropic is reportedly in talks to acquire Decart, an Israeli AI startup, for roughly $6bn, according to reports attributed to Reuters. If the deal goes through, it would be Anthropic’s biggest acquisition to date. The negotiations are not final, and both companies have declined to comment publicly.
What makes this story notable is not just the price tag, but the target: Decart does not build chatbots, but rather software that optimizes chip performance, reducing the cost of training and running AI models. This is particularly relevant in an era of resource-conscious spending, where cheap Chinese models consistently undercut Western labs, and investors have become wary of spiraling compute costs.
Decart was founded in 2023 by Dean Leitersdorf, Orian Leitersdorf, and Moshe Shalev, and while it also showcases impressive generative video and world models, its core focus lies in chip efficiency. This aligns with Anthropic’s current priorities, which are heavily influenced by the need to manage compute costs as they prepare for an IPO.
The reported deal value represents a significant premium over Decart’s valuation of $3.1bn a year ago and $4bn earlier this year. Anthropic, on the other hand, has its own massive valuation of around $965bn, as indicated by its confidential S-1 filing with the SEC in early June. This acquisition would be Anthropic’s fifth of 2026, following at least four smaller ones, and is likely strategically timed ahead of their public listing.
The logic behind this move is clear: Anthropic aims to secure essential compute infrastructure while controlling costs. By acquiring a company focused on chip efficiency, they can maximize the performance of their hardware investments and potentially reduce long-term operational expenses.
This strategy isn’t limited to Anthropic; other tech giants like Nvidia, SpaceX, and Amazon were also reportedly interested in Decart. The market recognition of chip efficiency as a strategic asset underscores its growing importance in the AI landscape.
Paradoxically, this shift towards cost-efficiency challenges the long-held belief among frontier labs that larger models and data centers are the only path to better AI.