Anthropic Walks Away From $6B Decart Acquisition

Anthropic has pulled out of a reported $6 billion deal to buy the Israeli chip-efficiency startup Decart, ending weeks of advanced negotiations after due diligence, according to Bloomberg, which broke the story on 8 September. It would have been the largest acquisition in Anthropic's history. It is now the largest one that did not happen.

For anyone building on Claude, the interesting part is not the gossip. It is what the deal was supposed to fix, and the fact that it still needs fixing.

What Decart does and why Anthropic wanted it

Decart builds software that makes AI chips run more efficiently. In practice that means more tokens out of the same silicon, which pushes down the cost of both training and serving models. Anthropic, like every frontier lab, is compute-constrained: demand keeps outrunning the capacity available to meet it.

The company is about three years old and has raised roughly $450 million, including a $300 million round in May at a valuation near $4 billion, per Globes. A $6 billion offer was a meaningful premium on that mark set four months earlier.

Buying an efficiency layer outright, rather than licensing it, would have given Anthropic a cost advantage its competitors could not simply go and buy next quarter. That is the strategic logic, and it is sound. The deal died on the details.

Why it fell apart

Nobody involved has published a full account, and the reporting is sourced rather than official, so treat the reasons as reported rather than confirmed. Three explanations recur across coverage:

• Due diligence surfaced something. The most common framing, and the least specific. Weeks of diligence ended with Anthropic declining at a price it had previously been willing to discuss.

• IPO timing. Announcing a $6 billion acquisition mid-process would mean amending the S-1 and inviting fresh SEC scrutiny at an inconvenient moment.

• A corporate move to the US. Globes reported in a follow-up that a required restructuring, relocating the Israeli company's corporate domicile, complicated the transaction alongside diligence findings.

Globes also reports that Decart's founders had earlier turned down a higher offer from Nvidia. If accurate, that reframes the story: this was not a startup desperate for an exit, and the price was not the binding constraint.

The companies may still work together, with Anthropic becoming a customer or an investor instead of an owner. That is the normal landing spot for a collapsed acquisition where the technology still solves a real problem.

What this actually means if you build on Claude

Very little changes this week. Nothing about pricing, rate limits, or model availability moves because a deal did not close. But the deal existing at all tells you something worth holding onto.

Frontier labs are spending seriously to attack inference cost, because inference cost is the thing standing between current pricing and the pricing they want to offer. When a lab is willing to pay $6 billion for an efficiency layer, the honest read is that cheap capacity is not solved and is not close to solved.

The practical implications for anyone with a product on top of a frontier API:

1. Do not plan a business model on the assumption that token prices fall on a schedule. They have fallen a lot and will probably fall more, but the path is lumpy and driven by exactly this kind of deal, which sometimes does not happen.

2. Keep your provider abstraction thin but real. A collapsed acquisition is a reminder that the cost curve you depend on is someone else's roadmap, not yours.

3. Watch capacity language, not just price. Rate limits and tier gating tell you more about a lab's compute position than its price list does.

If you want the general version of that reasoning, we wrote about what an AI funding round actually means for customers and what to do when a startup whose tools you rely on gets acquired. A deal that collapses is the same analysis with the sign flipped: the strategic gap the acquisition was meant to close is still open.

The pattern worth noticing

This is the second time in recent weeks that an AI acquisition story ended somewhere other than an acquisition. Cognition publicly said it was not for sale shortly after its own funding round. Consolidation in this market is being reported far more often than it is being completed.

That is a useful prior for reading the next one. Reported talks are not deals. Deals are not closings, which is worth remembering the next time a headline arrives, and is half of reading AI news usefully. And a lab's willingness to spend billions on a capability is better evidence about its constraints than any blog post it publishes about efficiency.

FAQ

Did Anthropic officially announce it was cancelling the Decart deal? No. The reporting is sourced, starting with Bloomberg on 8 September and followed by Israeli outlets including Globes and Calcalist. Neither company has published a formal statement confirming the terms or the reasons.

Does this affect Claude pricing or availability? No. The deal was never announced as closed, so nothing about the current API, pricing tiers, or rate limits changes as a result of it not closing.

What was Decart valued at before this? Around $4 billion, set by a $300 million funding round in May 2026, per Globes. The reported Anthropic offer of roughly $6 billion was well above that mark.

Could the two companies still work together? Reportedly yes. Coverage suggests Anthropic could become a customer or an investor in Decart rather than an acquirer, which is a common outcome when diligence kills an acquisition but the underlying technology still fits.