Amazon in Talks to Buy Decart for $7 Billion: Why Chip-Portability Software Is the New Prize

The report in brief

On Sunday, October 11, 2026, the Wall Street Journal reported, citing people familiar with the matter, that Amazon is in talks to buy San Francisco-based AI startup Decart in a deal that could be valued at around $7 billion. Reuters, Bloomberg, Yahoo Finance, SiliconANGLE, PYMNTS and Israel’s CTech all followed within hours. As of October 12, neither Amazon nor Decart had publicly confirmed the talks, and no definitive agreement had been announced. According to the reporting, a deal could be wrapped up soon if negotiations do not collapse or another suitor does not emerge.

The story is interesting well beyond the price tag. It sits at the intersection of three forces shaping the AI industry: scarce computing capacity, the race among chip vendors, and the strategic value of software that sits between models and hardware.

Who is Decart?

Decart was founded in 2023 by Israeli engineers including brothers Dean and Orian Leitersdorf and Moshe Shalev. Its core product, as described in the reporting, is software that makes it easier for developers to switch between different AI chips. The company has also developed models, including Lucy for real-time video editing and Oasis for simulated environments used in robotics and autonomous-driving tests.

Funding and valuation history tells the story of how quickly the category has heated up. Decart raised $300 million at a roughly $4 billion valuation in May 2026. A reported $7 billion price would represent a substantial step up in about five months. Amazon Web Services had previously given the startup early access to its custom Trainium3 chip, so the two companies already have a working relationship.

Why chip portability is worth billions

For most of the current AI boom, the practical default has been to build on Nvidia GPUs and the CUDA software ecosystem. That created two problems for everyone else.

Supply. Demand for compute exceeds the supply of any single vendor’s processors. Labs and enterprises that can run on more than one kind of chip can use whatever capacity is available.

Cost and bargaining power. If your stack only runs well on one vendor’s hardware, you accept that vendor’s pricing. Portability turns hardware into a competitive market.

Software that lets developers move models and workloads across chips, or optimizes them for non-default accelerators, therefore has a double payoff. It gives customers flexibility, and it gives alternative chip vendors a way to attract workloads without asking developers to rewrite everything.

Amazon has invested heavily in custom silicon, including Trainium for training and Inferentia for inference. Hardware alone does not win developers; software ergonomics do. Buying a company whose product reduces the friction of running on Trainium could improve utilization of Amazon’s chips relative to third-party GPUs. Market commentary described the logic as buying optimization software that makes Amazon’s chips more attractive to developers who want portability. That is an analytical reading of the strategy, not something either company has said.

The Anthropic chapter

Earlier reports said Anthropic had explored acquiring Decart for about $6 billion before walking away in September after due diligence. Tech Times framed it as leaving a compute-cost gap unsolved, while Startup Fortune described Amazon as circling the startup after other interested parties had stepped back. The reasons for Anthropic’s decision have not been officially stated, and it would be a mistake to over-read it. Due diligence can end talks for price, integration, technology, or strategy reasons.

What the sequence does show is that multiple well-capitalized buyers considered the asset valuable enough to run serious diligence. That is a meaningful data point about how AI labs and cloud providers value anything that lowers inference cost.

What the market and press emphasized

Coverage clustered around four angles:

1. The price. Roughly $7 billion for a company founded in 2023 stood out, especially against the May valuation.
2. The Nvidia angle. Outlets such as WSJ have written about Amazon’s custom chips posing another threat to Nvidia; a Decart deal would fit that storyline.
3. The Israeli tech angle. CTech highlighted Amazon emerging as a potential buyer of an Israeli-founded startup, adding to a long pattern of large exits for Israeli infrastructure companies.
4. The AI-lab compute race. Yahoo Finance and others tied the story to labs hunting for cheaper, faster compute.

For search interest, Google’s U.S. daily trending list on October 12 featured sports and entertainment names, not this story. As with most deal news, interest is concentrated among founders, investors and infrastructure engineers searching for specifics.

Lessons for startup founders

Strategic buyers pay for removing friction

Decart is valuable not because it makes the biggest model but because it makes a painful problem less painful for customers and for chip vendors. Founders should ask whether their product removes a bottleneck that a larger company would otherwise have to remove itself.

Neutrality is an asset and a liability

A chip-agnostic tool is attractive because it works across vendors. Once it is owned by a cloud provider with its own silicon, customers on other platforms may question its neutrality. If you are building in this category, think early about what you would lose in an acquisition and whether the exit path or the independence path creates more long-term value.

Multiple bidders create leverage

The reported interest from more than one buyer demonstrates the value of running a process rather than waiting for a single inbound offer. Keep strategic relationships warm with several potential acquirers, and document technical differentiation clearly enough to survive diligence.

Diligence is real

A failed diligence process, such as the one reported with Anthropic, does not necessarily damage a company’s value, but it can change negotiating dynamics. Keep your data room clean, your IP assignments documented and your customer contracts transferable.

Timing windows are short

Valuations in hot categories can move quickly in both directions. A company that raised at around $4 billion in May being discussed at about $7 billion in October illustrates how fast the window can open. It also shows how fast it can close if the strategic logic changes.

Lessons for operators and buyers of AI infrastructure

• Audit your chip dependence. Know which parts of your stack are tied to one vendor’s libraries and how costly a migration would be.
• Test alternatives deliberately. Run pilot workloads on non-default accelerators to learn real performance and cost differences rather than relying on vendor claims.
• Watch contract terms. If a tool you rely on is acquired, review change-of-control, support and pricing clauses.
• Plan for consolidation. Independent tooling vendors in this layer may become part of cloud providers. Build exit plans for critical dependencies.

Risks and open questions

• The deal may not happen. Reported talks collapse often, and neither company has confirmed these.
• Regulatory review. A transaction of this size involving a major cloud provider could draw antitrust attention, especially given broader scrutiny of Big Tech acquisitions.
• Integration. Keeping founders and engineers motivated after acquisition is difficult, especially in a market with intense competition for AI talent.
• Customer reaction. Customers using Decart’s tools across multiple clouds will want assurances about future support.

Frequently asked questions

Has Amazon confirmed the deal? No. Amazon and Decart had not publicly confirmed the talks as of October 12, 2026.

How much is the deal reportedly worth? Around $7 billion, per the Wall Street Journal.

What does Decart do? It builds software that eases switching between AI chips and has developed video and simulation models such as Lucy and Oasis.

Why did Anthropic walk away? Reports say it explored a roughly $6 billion deal and ended talks in September after diligence; the specific reasons have not been officially stated.

Why does Amazon care? Chip-portability software could make its Trainium processors easier to adopt, per market commentary.

Bottom line

Whether or not the Amazon-Decart talks produce a signed agreement, the report captures where value is accumulating in AI: not only in models and chips, but in the software that lets workloads move between them. Founders in adjacent categories should treat strategic buyers as a real path. Operators should treat chip flexibility as risk management.

This article summarizes press reports of unconfirmed talks and is not investment advice.

Sources consulted: The Wall Street Journal, Reuters, Bloomberg, SiliconANGLE, PYMNTS.com, CTech, Startup Fortune, Tech Times, Yahoo Finance and Google Trends (US daily trending searches, October 12, 2026).