Amazon is in talks to acquire Decart, an AI startup whose technology helps developers move workloads between different AI chips, for roughly $7 billion, the Wall Street Journal reported on Sunday, citing people familiar with the matter. According to the Journal, a deal could come together soon — a striking turn for a startup that came close to being absorbed by Anthropic barely two months ago.
The report was quickly picked up by Reuters, Bloomberg and Barron's on Sunday evening, giving the story a rare multi-outlet footprint for deal talk that has not yet been confirmed by either company. Neither Amazon nor Decart has publicly commented on the negotiations. For more context on this story, see our ongoing more AI stories.
What Decart Actually Does
Decart's business sits at two layers of the AI stack. The company develops software that makes it easy for developers to switch between different AI chips — an increasingly valuable capability as training and inference workloads get shuffled between Nvidia GPUs and a growing field of in-house silicon from Amazon, Google and others.
The startup also builds AI models that can generate and simulate real-world 3D environments, a category often described as world models that researchers see as a stepping stone toward AI systems that understand physics and spatial relationships rather than just text and images. In practice, that means software that can conjure interactive, photoreal settings on demand — useful for training robots, building simulations, and powering games or virtual production without expensive on-location capture.
That combination — hardware-agnostic infrastructure software plus frontier-adjacent world models — made Decart a target before. The company was valued at nearly $4 billion in May, which means the roughly $7 billion figure the Journal cited would nearly double its valuation in under six months.
The Trainium Connection
The deal logic is not hard to reconstruct. Amazon is progressing with development of its own AI chip, Trainium3, in direct competition with Nvidia's data-center dominance, and the Journal reported that Amazon had given Decart early access to the chip back in 2025 — a sign the two companies have been collaborating for more than a year. Software that makes it painless to port workloads onto new silicon is exactly what a chip vendor trying to woo developers away from CUDA needs, and buying that capability outright would deepen the moat around Trainium.
For Amazon Web Services, which rents AI compute by the hour, every customer that can shift between chips without rewriting code is a customer less locked into Nvidia's pricing. Chip-switching software is, in effect, an anti-lock-in tool — and Amazon would now own one.
Anthropic Walked Away First
Decart's acquisition saga has already had one act. In August, Anthropic was reported to be in talks to acquire the company for $6 billion, and subsequent reporting said the talks had progressed toward $7 billion after Decart's founders spurned a richer offer from Nvidia. Then in September, Bloomberg reported that Anthropic walked away from the deal entirely.
If Amazon closes at or near $7 billion, it would validate the price that reportedly scared Anthropic off — and reinforce a pattern in which the largest cloud and compute players, rather than the model labs themselves, are doing the buying. Amazon has been steadily adding AI capability through smaller acquisitions in recent years, but a purchase of this size would rank among its biggest in years.
The timing also lands amid a broader consolidation wave in AI infrastructure, with Nvidia reportedly exploring a major investment in or acquisition of open-weight startup Reflection AI, and SoftBank seeking as much as $100 billion from Gulf investors for an AI acquisition fund, according to the Financial Times.
What to Watch
Talks can fall apart, and neither company has confirmed them. But the Journal's sourcing describes a deal that could come together soon, and the strategic rationale on Amazon's side is unusually clean: Decart's software lowers the switching cost away from Nvidia, its world-model research complements Amazon's simulation and robotics ambitions, and the reported price is one Amazon can absorb without straining a balance sheet already funding one of the industry's largest data-center build-outs.
The antitrust angle is unlikely to be decisive at this size, but the deal would still draw scrutiny simply for what it consolidates: a neutral layer that today serves many chips would sit inside one of the very vendors it helps developers reach. Whoever ends up owning the on-ramp, the off-ramp questions follow.
For the AI industry, the deeper signal is that the market for AI infrastructure middleware is consolidating into the hands of the hyperscalers — the very companies that most benefit from developers being able to move freely between chips, as long as the moving happens inside their own walls.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
