What happened
CryptoBriefing reported Wednesday that Anthropic, the San Francisco AI lab behind the Claude model family, is in talks to acquire Decart, an Israeli startup focused on making large model training and inference cheaper. The reported price is around $6 billion. Neither Anthropic nor Decart has issued a public statement confirming or denying the negotiations, and the report notes the talks are still in progress.
The deal has not been signed. If it closes at the reported figure, it would sit near the top of the private AI M&A league table and mark Anthropic's most aggressive vertical integration move since its founding. Anthropic's cap table already includes Google and Amazon as strategic backers, and the company has been burning through compute credits at a pace that has forced Silicon Valley to rethink what a mature AI lab's cost structure looks like.
Why it matters
Compute is Anthropic's single largest expense. Every basis point shaved off training and inference cost flows straight to unit economics on Claude API calls and enterprise contracts. Buying Decart outright, rather than licensing the tech, tells you Anthropic wants that efficiency layer inside the tent and away from OpenAI and Google.
The price also matters. $6 billion for a startup whose product is a wrapper on cost reduction implies Anthropic sees the payback measured in single-digit years of compute savings, not decades. That's a real signal about how expensive frontier training has become.
For the crypto side of the room, the read-through is narrower but real. Decentralized compute networks and AI-adjacent tokens have spent 2026 pitching themselves as the low-cost alternative to hyperscaler capex. A well-funded incumbent absorbing an efficiency specialist tightens that pitch.