Amazon Eyes a $7 Billion Decart Deal After Anthropic and Nvidia Walked Away From the Same Startup

Amazon is circling a startup that two of the biggest names in AI already tried to buy and let go. The company is in talks to acquire Decart AI for around $7 billion, according to people familiar with the matter who spoke to the Wall Street Journal. The deal could wrap up soon unless talks collapse or another suitor steps in.

Neither Amazon nor Decart has confirmed anything. But the price tag alone says a lot. Decart is three years old and has never disclosed its revenue.

The suitors who walked

Amazon is not the first to knock. Anthropic discussed buying Decart for roughly $6 billion over the summer, then walked away in September after due diligence. Israeli outlet Globes reported a sticking point: the founders resisted Anthropic's push to move much of the Tel Aviv team to San Francisco.

Nvidia held its own acquisition talks earlier. It settled for a stake instead, investing in Decart's $300 million round in May 2026 at about a $4 billion valuation. Amazon joined that round as a strategic customer.

So the startup's value has nearly doubled in five months, and the two companies that looked closest are now watching from the side. Anthropic's absence is the strange part. It is one of Amazon's most important partners, with Amazon having invested around $18 billion and promising up to $15 billion more tied to milestones. Anthropic also runs its models on Amazon's Trainium chips.

What Decart actually sells

Decart was founded in 2023 by brothers Dean and Orian Leitersdorf along with Moshe Shalev. It is headquartered in San Francisco, with research and development in Tel Aviv.

Its core product is unglamorous and very valuable: software that makes AI models run faster on chips from Nvidia, Google and Amazon. Decart licenses that technology to cloud providers and AI labs. Industry estimates put the speedup at up to about 8x, depending on the workload and hardware.

The company also builds its own models. Lucy edits live video in real time, and Oasis generates interactive worlds. Those are the flashy demos. The optimization layer underneath is what the big buyers want.

Why Amazon cares

The real target is Nvidia's grip on AI computing. Most AI developers write for Nvidia's CUDA software, which makes switching to rival chips painful. Amazon has spent years pushing its own Trainium processors as an alternative, and software that makes models perform well on Trainium is exactly what it needs to pry customers loose.

Decart has already shown what that looks like. At re:Invent in December 2025, Amazon said Decart's real time video ran 4x faster on Trainium at half the cost of GPUs, after Decart moved its models onto Trainium3.

The race to own the full AI stack, from chips to the software that runs on them, is moving faster than the rules around it. Microsoft's president recently argued that AI is less regulated than milk, even as the biggest companies keep stacking up acquisitions and custom silicon. Microsoft itself unveiled its Maia 300 inference chip the same week as the Decart reports, aiming to make it the primary chip for Microsoft products from 2027.

The open question

The biggest unknown is what happens to Decart's customers who do not run on Amazon hardware. Today its software supports Nvidia and Google chips too. Whether that continues under Amazon ownership is unclear, and it matters to every lab and cloud provider currently licensing the technology.

An Amazon that owns the tool making rival chips fast would have a powerful lever. It could keep the neutral product running and collect fees from competitors, or it could tilt the best optimizations toward Trainium. For now, the talks are unconfirmed and the deal is not done. If it does close, Amazon will have paid roughly $7 billion for a company founded in 2023.

Some offers on this page may be paid placements or contain affiliate links.