Amazon reportedly nears $7.5bn deal for AI start-up Decart
Amazon is close to a deal to buy Decart, a three-year-old AI lab whose software makes AI models run faster on chips including Amazon's Trainium, for about $7.5bn, the Financial Times reported on 21 November 2026.

Listen to this articleListen
Amazon is closing in on a deal to buy Decart, a three-year-old artificial intelligence lab, for about $7.5bn, the Financial Times reported on 21 November 2026. The FT said the acquisition would help Amazon Web Services, Amazon’s cloud business, compete with Nvidia in the race for computing power.
Decart makes AI models that generate and edit video live, and software that makes AI models run faster on chips from Nvidia, Google and Amazon. Decart already runs its live video models on Trainium3, Amazon’s latest AI chip.
What does Decart make?
Decart makes two families of live AI models, Lucy for editing video and Oasis for simulating worlds, and DOS, the software stack they run on. Lucy changes a live video stream as it plays, so a presenter can try on an outfit from a catalogue or hold a product that is not in the room. Decart says Lucy edits live video in under 40 milliseconds at up to 100 frames a second, and its latest version, Lucy 2.6, went live on 5 November 2026. Oasis generates interactive simulations of the physical world for robots and other physical AI; Oasis 3 arrived in June 2026.
DOS, the Decart Optimization Stack, sits underneath both. Decart says DOS 2.0, launched in May 2026, runs on Nvidia’s graphics processors, Google’s TPUs and Amazon’s Trainium, and delivers more than 1,600 tokens a second for AI agents against an industry average of about 200. Decart licenses DOS to cloud providers and AI labs. It was set up in September 2023, is led by its chief executive and co-founder, Dean Leitersdorf, and has a little over 100 staff in Tel Aviv, New York and San Francisco.

Why would Amazon want Decart?
Decart runs its live video models on Trainium3, Amazon’s own AI chip, and Amazon picked out its results when it launched Trainium3 servers on 2 December 2025: real-time video four times faster at half the cost of graphics processors. Trainium is made by Annapurna Labs, Amazon’s chip unit, and Amazon says its Trainium3 servers let customers train and run AI models faster and at lower cost. On Amazon’s Trainium customer page, Decart’s chief executive, Dean Leitersdorf, says early testing cut the delay before its models respond “from 40 ms to 10 ms”.
The two companies already work together. Decart’s May 2026 funding announcement named Amazon as one of its strategic customers and described a commercial agreement and joint sales effort with AWS. In it, Nafea Bshara, a vice-president at Annapurna Labs, said Decart’s engineers “were among the first in the world to optimize on Trainium3”.

How much has Decart raised?
Decart has raised $453m in four rounds since October 2024, so a $7.5bn price would be more than 16 times the money investors have put in. Its largest round, $300m in May 2026, was led by Radical Ventures, with Nvidia among the new investors. Sequoia Capital led its first round and Benchmark its second.
| Date | Round | Raised |
|---|---|---|
| October 2024 | Seed, led by Sequoia Capital | $21m |
| December 2024 | Series A, led by Benchmark | $32m |
| August 2025 | Funding round | $100m |
| May 2026 | Funding round, led by Radical Ventures | $300m |
| Total | $453m |

Decart will be at Amazon’s cloud conference next week
Decart is due to show its models at AWS re:Invent, Amazon’s annual cloud conference in Las Vegas, which runs from 30 November to 4 December 2026. It will be at booth 1930 and will run live demonstrations at the AWS Startup Plaza during the conference.
Sources
- Financial Times: Amazon closes in on $7.5bn deal to buy AI start-up Decart, 21 November 2026ft.com
- Amazon: Trainium3 UltraServers, 2 December 2025aboutamazon.com
- AWS: Trainium customersaws.amazon.com
- Decart: Decart raises $300M, 18 May 2026decart.ai
- Decart: about the companydecart.ai
- Decart: Lucy 2.6, 5 November 2026decart.ai
- Decart: eventsdecart.ai
- Decart: home pagedecart.ai


