Anthropic Eyes $6 Billion Acquisition of Infrastructure Startup Decart

If it closes, the deal would be Anthropic's largest acquisition to date and a quiet admission that compute efficiency has become a board-level priority.

The asset underneath the price tag

Decart sells software, not chips. Its value proposition is straightforward: squeeze more usable performance out of whatever silicon is already in the rack — NVIDIA GPUs, TPUs, Amazon's Trainium and Inferentia. Anthropic, like every frontier lab, has been throttled by the GPU supply curve. Buying a workload optimizer is cheaper than waiting on new fab capacity or outbidding hyperscalers for H200 allocations. If you cannot buy more flops, rent them better.

Following the money

The deal lands in a peculiar stretch of Anthropic's cap table. The company closed a $65 billion Series B in 2026 — a figure that already strains credibility and will need to be defended in the S-1. Spending $6 billion of that war chest on a private startup, rather than on compute contracts or balance sheet repair ahead of a rumored October listing, is a signal investors should parse carefully. Bloomberg Deals reporter Ryan Gould framed the number as "modest" relative to the wider AI M&A landscape. "Modest" is doing heavy lifting. For a pre-IPO company, every dollar deployed is a line item a public-market investor will discount.

There is also a narrative angle. Anthropic has historically run a disciplined M&A book — small, targeted, almost invisible. A nine-figure infrastructure play changes that posture. It tells prospective public investors: we are willing to spend for control of the cost curve, not just for model capability. Bloomberg AI reporter Shirin Ghaffary noted that Decart's cross-platform optimization is the differentiator that makes this asset defensible rather than redundant.

The reality check

Decart was also linked to SpaceX earlier in the week; Elon Musk publicly denied those talks. That detail matters less for what it says about Decart and more for what it confirms about the market — scarce AI infrastructure assets are being bid up across the board, and rumor control has become its own form of competitive intelligence.

The sober read: even if Anthropic manages to run its chips cheaper, the real test is whether the cost-efficiency story survives due diligence from public-market investors who have already watched AI labs promise margin discipline and deliver the opposite. A $6 billion acquisition can dress up the S-1. It cannot, by itself, fix the unit economics.