Anthropic is reportedly in talks to buy Decart for $6bn, its biggest deal yet

Anthropic is reportedly in talks to acquire Decart, an Israeli AI startup, for roughly $6bn, according to reports attributed to Reuters.

If it happens, it would be the largest deal in Anthropic’s history. The talks are not final, both companies declined to comment, and the whole thing could still fall apart.

What makes the story worth pausing on is not the price but the target. Decart does not build a rival chatbot. It builds software that squeezes more work out of the same chips, cutting the cost of training and running AI models.

In an era of thrift-maxxing, where cheap Chinese models keep undercutting Western labs and investors have grown wary of bottomless compute bills, that is a pointed thing to spend $6bn on.

Decart was founded in 2023 by three Israeli engineers, Dean Leitersdorf, Orian Leitersdorf, and Moshe Shalev. Alongside its efficiency work, it has a flashier side in generative video and world models, showcased in its Oasis demo back in October 2024.

But it is the unglamorous chip-efficiency layer, the part that quietly lowers the electricity bill, that fits Anthropic’s current preoccupations rather than the eye-catching video tricks.

The price tag reflects a rapid re-rating. Decart was valued at $3.1bn a year ago, in August 2025, then nudged towards $4bn this May after a $300m round led by Radical Ventures.

A $6bn deal would mark a premium of roughly 50% over that most recent mark, which is a brisk climb for a two-year-old company.

Anthropic, for its part, is operating on a different scale entirely. Its own valuation has ballooned to somewhere around $965bn after its most recent mega-round, and in early June it confidentially filed an S-1 with the SEC, the paperwork that precedes a public listing.

This would be its fifth acquisition of the year, following at least four smaller ones, but by far the biggest, and the timing so close to a listing is unlikely to be a coincidence.

The logic is not hard to read. Anthropic is locked in the same compute arms race as everyone else, and the cost of that race is the number every prospective public-market investor will scrutinise most closely.

Buying a company whose entire purpose is to make each chip go further is a way of shoring up the infrastructure and controlling costs before the books go on display.

It is a rare and rather European sort of moment, an AI leader spending big precisely so that it can spend less, and betting that thrift, not sheer size, is what the market now wants to see.

That instinct is already visible elsewhere in Anthropic’s dealmaking. It recently signed a multi-year GPU deal with CoreWeave to run Claude at production scale, part of a wider scramble to lock in capacity on terms it can live with.

An efficiency acquisition would be the other half of the same equation, wringing more from the hardware it is busy securing.

Anthropic was not the only suitor circling, either. Nvidia, SpaceX, and Amazon were all reportedly interested in Decart, which tells you the market has decided efficiency is now a strategic asset in its own right rather than a nice-to-have.

When the chipmaker itself, a rocket company, and a cloud giant are all sniffing around the same startup, the thesis is no longer niche.

There is a tidy irony in all of this. The frontier labs spent years insisting that the path to better AI ran through ever-larger models and ever-larger data centres.

Now, with an IPO in view and a cost-conscious market rewarding thrift, one of the loudest champions of scale is reportedly willing to pay a record sum to get smaller, cheaper, and leaner.