Amazon in talks to acquire Decart for ~$7B

Five months ago, Decart was a $4 billion company closing a $300 million funding round ^[4]. Today it is reportedly being discussed at close to $7 billion ^[1]. In between, Anthropic began due diligence on a roughly $6 billion offer by August 2026, in what would have been its largest acquisition to date, then abandoned the deal in September with no public explanation ^[3]. Decart had also been negotiating with Nvidia before another bidder entered the process, and SpaceX and Nebius have separately been reported as interested parties alongside Amazon ^[7]. That sequence of walk-aways and new entrants, rather than a single escalating auction, is what pushed the implied price from $4 billion to $7 billion in under half a year. On X, one analyst thread flagged this same trajectory directly, arguing the price roughly tripled in five months for a company that had not shipped a mainstream consumer product, and describing the outcome, in the post's own words, as less a single buyer's valuation than 'a bidding war for the two founders and their inference layer.'

Decart's headline product, DOS, is infrastructure: software that lets AI workloads run efficiently across different chip types, including Nvidia GPUs, Google TPUs, and Amazon's own Trainium chips ^[2][5]. The performance case is concrete. On Amazon's Trainium chips, Decart's optimization reportedly took video processing from about 26 frames per second on GPU to close to 101 frames per second ^[6], and the company's DOS 2.0 release is reported to let AI agents process over 1,600 tokens per second, roughly eight times the industry average ^[4]. Decart also builds two 'world models' - Lucy, a real-time video generation and editing model, and Oasis, which generates simulated 3D environments for robotics and autonomous-driving training ^[5]. Crucially, this was not a cold approach: Amazon had already invested in Decart's $300 million round and was using its technology as a customer before talks began ^[7], and Decart had already appeared as a keynote customer on the AWS stage, demonstrating real-time visual intelligence live. The relationship, in other words, predates the deal by months.

AWS has long sold itself as the cloud where customers can freely choose between Nvidia GPUs, Google TPUs, or Amazon's own Trainium silicon. Owning Decart's chip-optimization layer could meaningfully boost Trainium adoption, but it sits awkwardly next to that pitch, since the same software that makes switching chips easy for customers also becomes a lever Amazon could use to steer workloads toward its own hardware ^[2]. Decart's own founder has publicly argued the technology is chip-agnostic by design - 'customers don't care what chip we use,' he has said, framing Amazon's advantage as currently being about cost and speed rather than lock-in ^[6]. Reddit discussion of the deal picked up on a related angle: that Decart's value lies partly in helping Amazon reduce reliance on Nvidia's CUDA ecosystem, though one commenter noted Trainium is still fabricated by TSMC, so the move doesn't fully resolve the underlying chip supply bottleneck.

Nothing here is finalized. Amazon and Decart have both declined to comment, and the reported $7 billion figure is not a confirmed offer or final valuation ^[1][2]. That caution is reinforced by recent history: Anthropic's own ~$6 billion attempt to buy Decart got as far as due diligence before collapsing ^[3], so a reported price in talks is clearly not the same as a signed deal. Community reaction has been correspondingly mixed rather than celebratory - some Reddit discussion questioned whether the wider SpaceX-and-Nebius bidding-war narrative is being overstated, and there was speculation, unverified, about potential personal financial interests among executives involved. Combined with the fact that Anthropic has already walked away from a serious Decart deal once, the most defensible read right now is that this is a live negotiation with real strategic logic behind it, not a done deal.