- Amazon is hunting for a way out of Nvidia dependency, and it thinks Decart is the key. The San Francisco startup, founded in 2023 by three Israeli engineers, has built software that lets developers write code once and run it on any AI chip—Nvidia, Amazon’s Trainium, Intel, whatever. That abstraction layer is becoming priceless in a world where every company is scrambling to secure chip supply. Amazon is in advanced talks to buy Decart for around $7 billion, a 75% premium over its May valuation of $4 billion. The jump in valuation is revealing: chip abstraction software is now worth billions because it reduces customer lock-in to specific hardware. If the deal closes, Amazon gets a tool to make its custom Trainium chips more attractive to developers who would otherwise have to rewrite code to switch from Nvidia.
- Decart is exactly what Amazon needs. AWS gave the startup early access to Trainium3 in 2025 after being impressed by its founders. The relationship makes sense: Decart’s software solves Amazon’s hardest problem—convincing developers that using Trainium instead of Nvidia GPUs won’t lock them into Amazon infrastructure. With Decart software, a developer can write once and run on Trainium, Nvidia, Intel, or anything else. That flexibility is the missing piece in Amazon’s chip strategy. Nvidia has dominance because it has developer momentum, not necessarily technical superiority. Decart removes that switching cost and makes Trainium competitive on economics alone. For a company scrambling to reduce Nvidia’s chokehold on AI infrastructure, that’s invaluable.
- Decart is also building world models—Lucy and Oasis are systems that can simulate 3D environments in software. That’s a separate business line that could become huge if AI systems start needing environmental simulation for training or testing. Amazon sees that upside too. The startup raised $400 million from top-tier VCs including Sequoia, Benchmark, and Radical Ventures, which means it had no shortage of admirers. The $7 billion price tag says Amazon is willing to pay a premium to lock in a strategic asset before competitors realize what Decart does.
- The timing matters. Amazon has custom chips, Facebook has custom chips, Google has custom chips, and Microsoft is exploring them. Everyone is trying to break Nvidia’s grip. But custom chips only win if developers adopt them, and developers won’t adopt them if migration costs are high. Decart solves that. The deal could close soon according to people familiar with the talks, which suggests Amazon sees this as urgent. With OpenAI revenue under pressure and AI infrastructure spending being questioned, Amazon is doubling down on the economic case: use Trainium with Decart software and get comparable performance at lower cost than Nvidia.
What Happened?
Amazon is in advanced negotiations to acquire Decart, a San Francisco-based AI software startup, in a deal valued at approximately $7 billion. Decart develops software that enables developers to write AI code once and run it across different AI chips, including Nvidia GPUs, Amazon’s custom Trainium chip, Intel processors, and others. Founded in 2023 by three Israeli engineers—brothers Dean and Orian Leitersdorf and Moshe Shalev—Decart was previously valued at nearly $4 billion in May. The company has raised more than $400 million from investors including Sequoia Capital, Benchmark, and Radical Ventures. AWS provided Decart early access to Amazon’s Trainium3 custom AI chip in 2025 after being impressed by its founders. Beyond chip abstraction, Decart is developing world models called Lucy and Oasis, systems capable of generating and simulating real-world 3D environments. A deal could close soon, though talks could still collapse or another buyer could emerge. Amazon’s custom chip Trainium competes with Nvidia’s dominant GPUs in the market for specialized AI computing hardware.
Why It Matters?
The Decart acquisition would represent Amazon’s biggest strategic move yet to break developer lock-in to Nvidia. Nvidia’s market dominance stems not from technical superiority alone but from developer momentum: once developers write code optimized for Nvidia, switching is expensive. Decart removes that switching cost. For Amazon, which invested billions in custom chip development, this is the missing piece. A developer can now use Trainium and keep the option to switch to Nvidia (or vice versa) with minimal code changes. That’s the flexibility that wins in competitive markets. At $7 billion, Amazon is paying a steep premium, but the strategic value justifies it. Every dollar Amazon spends on infrastructure efficiency reduces customer cloud costs, which makes AWS more competitive. For the broader market, the deal signals that chip abstraction software has become a strategic asset. If other companies realize what Decart does, competition for similar software could drive valuations even higher. The world model development (Lucy, Oasis) adds another dimension: if those systems become critical to AI training, Decart becomes even more valuable. Finally, the deal says something about confidence in custom chips. Amazon wouldn’t pay $7 billion to reduce Nvidia dependency if it didn’t believe Trainium would win real market share.
What’s Next?
Watch for the deal close announcement. If it happens, expect Amazon to immediately announce Decart availability as part of AWS AI services—positioning chip abstraction as a core AWS capability. That moves the narrative from “use Trainium or Nvidia” to “use whichever chip makes sense, and Decart handles the switching.” If the deal falls apart, watch for other buyers. Google, Microsoft, Meta, or even OpenAI might be interested in chip-agnostic software. Track Decart’s software adoption post-deal. If major AI companies start using Decart to reduce Nvidia dependency, that validates Amazon’s bet. If adoption stays niche, the $7 billion price tag looks expensive. Monitor Trainium adoption in the marketplace. If Trainium gains real market share as a Nvidia alternative, that proves chip abstraction works. If adoption stays flat, it suggests developers still prefer Nvidia regardless of software flexibility. Watch for similar acquisitions from Google, Meta, or Microsoft pursuing the same chip abstraction strategy. If multiple companies try to acquire abstraction software, that confirms the strategic importance. Finally, keep an eye on Nvidia’s response. If Nvidia starts building chip-agnostic compatibility itself or acquires similar companies, that shows the threat is real.
Affected Tickers and Coins: AMZN | NVDA | GOOGL | META
Source: Wall Street Journal
