SEATTLE —— Amazon is in talks to acquire Decart, a San Francisco-based artificial-intelligence startup, in a deal valued at roughly $7 billion, according to a report from Dow Jones Newswires published early Monday (Morningstar). The company's software tackles one of the thorniest problems in the AI boom: helping developers move workloads between different AI chips without having to rebuild their code from scratch.
The talks, reported overnight, have not been finalized and could still fall apart. But the reported price tag alone makes the potential deal one of the largest AI software acquisitions of the year — and a clear signal of where the industry's battle lines are moving. The fight is no longer just about who trains the biggest model; it is about who controls the layer that decides which hardware those models run on.
Table of Contents
- What Decart actually does
- Why the price tag is so high
- The chip-independence wars
- Amazon's acquisition playbook
- What could still go wrong
- Frequently Asked Questions
What Decart actually does
Decart builds an abstraction layer for AI compute: software that lets developers write once and deploy across different AI accelerators — Nvidia's GPUs, Amazon's own Trainium and Inferentia chips, and other alternatives — with far less friction than today's hand-tuned migrations. In practice, that means an AI company that trained on one vendor's hardware can shift some of its inference workloads to cheaper silicon without weeks of re-engineering.
That capability matters because AI chips are no longer interchangeable commodities. Each vendor ships its own software stack, and switching is expensive. Decart's pitch is to make the chip beneath the software fungible — which, if it works, puts the startup squarely at the center of every cloud provider's cost-management strategy.
Why the price tag is so high
Seven billion dollars is a striking figure for a company that sells developer tooling rather than a consumer-facing model. But in the current market, chip-portability software is a scarcity asset. Every major cloud provider is desperate to reduce its exposure to a single chip supplier — partly for cost, partly for bargaining leverage, and partly because regulators are watching the concentration of the AI supply chain.
Amazon, through AWS, already designs its own Trainium and Inferentia chips, and it is racing to capture a share of the AI-chip boom that has made custom silicon one of the most competitive corners of the industry. Owning the software layer that makes switching chips painless would accelerate that strategy: AWS customers could adopt Amazon's cheaper in-house chips for inference workloads without paying the usual migration tax.
The chip-independence wars
Amazon is not alone in this logic. The last year has seen a flurry of moves aimed at loosening the industry's dependence on any one chip architecture. AMD bought World Labs in an $8.2 billion deal (Chronicle), and cloud providers are all quietly building portability tooling of their own. A Decart acquisition would let Amazon leapfrog years of internal work — and deny the same capability to rivals, since an Amazon-owned Decart would have little reason to optimize for competing clouds.
That denial effect may be as valuable as the technology itself. Whoever controls the most widely used switching layer gets a seat at every hardware negotiation in the industry.
Amazon's acquisition playbook
Amazon has a history of paying up for infrastructure-level assets and folding them into AWS: Annapurna Labs, which became the foundation of Amazon's chip efforts, was acquired for a fraction of this price and now underpins billions in annual AWS revenue. If Decart follows the same path, the $7 billion figure would start to look like an early down payment on margin — AWS margins on compute are far healthier when the compute runs on Amazon-designed silicon.
The company has also been reshuffling its consumer AI leadership (Chronicle), a reminder that its AI investments now span both ends of the stack: devices in homes and the infrastructure beneath them. A Decart deal would fill a conspicuous gap in the middle.
What could still go wrong
Several hazards stand between the current talks and a closed deal. First, the price: startup valuations in AI infrastructure are running hot, and Amazon's board will need convincing that Decart's technology is defensible rather than a feature the cloud giants will eventually build themselves. Second, regulators: a $7 billion acquisition of a nascent but strategically important software layer will almost certainly draw antitrust scrutiny in Washington and Brussels, especially given the ongoing attention on cloud competition. Third, integration risk: developer tooling companies live and die on trust with their users, and an acquisition by one cloud provider can alienate the multi-cloud customers the product was built for.
Negotiations are ongoing, and terms may change. If the deal closes, expect it to be framed as an AWS play rather than a headline consumer product — the real prize is cheaper, more flexible compute, not a new chatbot.
Frequently Asked Questions
What does Decart do?
Decart is a San Francisco AI startup whose software makes it easier for developers to switch between different AI chips without rewriting their code. It effectively decouples AI workloads from the specific hardware they run on, reducing dependence on any single chip vendor.
How much is Amazon reportedly paying?
The talks value Decart at around $7 billion, according to Dow Jones Newswires. The discussions are ongoing and a final agreement has not been reached.
Why would Amazon want this company?
The software would make it far easier for AWS customers to move workloads onto Amazon's own Trainium and Inferentia chips, lowering costs and reducing dependence on a single chip supplier. It also strengthens AWS's position in the industry-wide race for chip independence.
Will regulators approve the deal?
That is far from certain. Large cloud acquisitions now routinely attract antitrust scrutiny, and a deal giving one provider control of chip-portability software could concern regulators in the US and EU. If talks result in an agreement, expect a lengthy review process.
