Amazon Decart Deal: $7 Billion Bet on AI Chip-Switching Software

What happened

Amazon is negotiating to buy Decart, a two-year-old AI startup, in a deal reportedly worth about $7 billion. The Wall Street Journal first reported the talks on Sunday, October 11, citing people familiar with the matter.116 A deal could close soon unless the talks fall apart or another buyer steps in.6 Neither company has confirmed anything. Amazon declined to comment outside business hours, and Decart did not respond to requests for comment.16

The news follows a strong session for Amazon stock. Shares closed Friday at $262.43, up 3.29%.4 They then rose about 0.4% in overnight trading after the report came out.3 Retail traders on Stocktwits were described as "bullish," a reading that hasn't changed over the past week.3

The headline number gets the attention, but the business logic behind it matters more. This looks less like a talent grab or a model play and more like Amazon buying the missing software piece for its custom-chip strategy.

What Decart actually sells

The coverage agrees on two product lines. The first is software that lets developers move AI workloads between different chips.58 The second is a set of "world models" that generate and simulate 3D environments.3

One outlet describes the company's technology as a single stack. A performance layer called DOS sits underneath two models, Lucy and Oasis.13 Lucy can edit live video in real time, according to Decart's website.6 Oasis generates simulated environments for training and testing robotics and self-driving systems.6 Oasis first drew international attention as a game that creates video on the fly as users play it.19

The reports weight these lines differently. Infrastructure-focused coverage centers on the chip-switching software. It frames the deal as Amazon's bid to make its Trainium chips a credible alternative to Nvidia.37 General-news coverage puts more emphasis on real-time video and robotics simulation.12 My reading is that the infrastructure layer is what justifies the price, and the world models are the upside.

The Trainium connection

Amazon and Decart already work together. AWS gave Decart early access to its Trainium3 chip in 2025 after meeting the founders and coming away impressed.3 Decart's CEO later used Trainium3 to demo the Lucy model.4 One report says AWS joined Decart's customer base in May and calls it a strategic customer of the startup.19

That history explains the strategy. AWS is building Trainium3 to compete with Nvidia's processors.5 The hard part of challenging Nvidia is often the software, not the silicon. Developers have built years of code around one hardware ecosystem, and moving that code is expensive. Software that lowers that switching cost for customers leaving Nvidia GPUs fits directly into AWS's sales pitch.4 Several outlets make the same point: competition between chipmakers now depends on software compatibility and developer tools as much as raw performance.713

The ownership structure adds an irony. Nvidia joined Decart's $300 million round in May, which Radical Ventures led.6 One report says Nvidia had earlier pursued its own deal for Decart before Anthropic became the preferred bidder in August.19 If Amazon wins, a company Nvidia backed could end up making it easier to move workloads off Nvidia hardware.

There's also a tension inside Amazon's own pitch. Decart's value comes from portability across chips. If Amazon tunes the software too heavily toward Trainium, it risks losing the developer trust that made the tool appealing in the first place.13 How Amazon handles that balance will show whether it bought a neutral tool or a funnel into AWS.

The valuation jump and the Anthropic detour

The price stands out. PitchBook data put Decart at nearly $4 billion in May.6 At around $7 billion, Amazon would be paying a big premium to that round.4 Since its 2023 founding, Decart has raised more than $400 million from investors including Sequoia Capital, Benchmark and Radical Ventures.7 Other accounts put total funding at about $450 million.1913

How the sale unfolded matters too, and outlets frame it differently. Bloomberg reported that Anthropic explored a deal worth about $6 billion in August. Then, last month, it decided not to proceed after due diligence.6 At least one outlet describes Amazon's interest as topping Anthropic's earlier $6 billion offer.18 That framing makes it sound like a bidding war. The more careful accounts say Anthropic walked away rather than lost.1620

The difference matters. A buyer that drops out after due diligence suggests at least one sophisticated acquirer saw too little value at $6 billion. One analysis argues that Anthropic, a model developer, may have valued Decart differently than a cloud provider whose chip business could benefit directly.13 That seems the most plausible explanation. Decart is likely worth more to the company that sells the compute than to a company that only buys it.

The reports say nothing on Decart's revenue, so whether $7 billion is justified on fundamentals can't be judged from what's public. Even the more enthusiastic analyses concede that Decart still has to prove customers will pay at scale.13

Lessons for AI founders

For founders, Decart is an unusually clear example of a winning exit strategy in the current AI market.

Solve a problem big companies are stuck on. Decart didn't try to win the frontier-model race. It went after a bottleneck: as AI workloads grow, running across different chips becomes more valuable.3 Software that frees companies from a single hardware platform matters when compute supply and cost determine how models get built and deployed.7

Turn a supplier into a customer, and then a buyer. Early Trainium access led to a showcase demo on Amazon hardware. That gave Decart a working relationship with exactly the company that now wants to own it.413 For founders, the playbook is to make yourself essential to a major platform's roadmap before asking it to write a check.

Have more than one product. The world models give Decart a second story in robotics and simulation, beyond chip software. That timing helps: AWS reportedly released an open-source Physical AI Toolchain for robotics just a week earlier.10 Having more than one strategic angle makes a startup worth more to more possible buyers, and Decart has drawn interest from Nvidia, Anthropic and Amazon.19

Hire for credibility early. Decart was founded by Israeli engineers Dean Leitersdorf, Orian Leitersdorf and Moshe Shalev.7 It is based in San Francisco and does most of its R&D in Tel Aviv.3 Within about three years, it reached a reported $7 billion price on research-heavy infrastructure work rather than consumer traction.

Where coverage disagrees on Amazon's finances

Some market commentary paints conflicting pictures of Amazon's finances. Several GuruFocus pieces describe the company as cash-flow-negative and unprofitable.1518 That claim is hard to square with other reported numbers. Market data shows trailing net income of about $135 billion, a P/E of 21.11 and slightly positive levered free cash flow of about $3.22 billion.1 The more defensible reading is that heavy AI capital spending is squeezing free cash flow while profits stay strong.1

Against that backdrop, $7 billion is a large deal but not a strain on Amazon's balance sheet. Amazon holds about $123 billion in cash.1 For comparison, it paid about $6.5 billion for MGM in 2022 and announced a deal for Globalstar worth about $11 billion this year.19

Some Wall Street analysts have trimmed their outlooks ahead of earnings. Cantor Fitzgerald cut its price target to $310 from $320 and kept an Overweight rating.1 Wells Fargo lowered its target to $329 from $338.9 Amazon reports earnings around October 29.1 AMZN is up about 13.7% year to date, slightly behind the S&P 500's 14.1%.3

The bottom line

The deal isn't done. Final terms, regulatory review and Trainium adoption are the variables to watch.4 The strategic logic is still clear. The AI chip race is moving into software, and Amazon seems willing to pay well above Decart's last valuation to own a tool that makes switching away from Nvidia easier. If the deal closes and Trainium adoption picks up, $7 billion could look cheap. If it doesn't, Amazon will have paid a large premium for a promising team and a set of interesting world models.