Think about what happens when you try switching from an iPhone to Android. It is never just about buying new hardware. Your apps, your paid subscriptions, your photo libraries, and your muscle memory are all locked into Apple’s garden. Leaving feels exhausting, so you stay put and pay whatever Apple asks for the next upgrade. That is precisely how Nvidia conquered artificial intelligence. For two decades, Nvidia did not just sell chips; it gave away CUDA, a proprietary programming platform launched in 2006. AI developers built their code, compiled their math, and tuned their models directly on top of Nvidia’s private software libraries. If you want to run those models on someone else's silicon, you cannot just flip a switch. You have to rewrite the underlying code from scratch. Now, Amazon is reportedly preparing to take a sledgehammer to that wall. According to the Financial Times, the tech giant is in advanced talks to acquire Decart, a three-year-old AI startup, for roughly $7.5 billion in cash, stock, and earnouts. An agreement could be announced as soon as the coming week, though the discussions could still collapse. Neither company has confirmed the talks, which follow earlier reports of a lower valuation. To understand why Amazon would cut a ten-figure check for an unproven startup, look at what both companies actually sell. Amazon runs AWS, the world's largest rented-computing empire, where companies host websites, store data, and train massive AI models. Decart sells portability. Its flagship product, the Decart Optimization Stack, acts as a universal adapter for AI models. It allows code to hop smoothly between Nvidia graphics chips, Google processors, and Amazon’s custom Trainium silicon. Instead of rewiring the entire house just to plug in a different appliance, Decart’s software translates the current on the fly.
The Numbers That Matter
Price LIVE$246.82
Market Cap$2.67T
Forward P/E22.6x
Total Revenue (TTM)$797.5B
52-Week Low$196.00
52-Week High$287.20
Analyst ConsensusStrong Buy (62 analysts)
Analyst Target Mean$326.51
Price refreshes live · All other figures as of November 20, 2026
Nvidia sells the high-performance sports car, but it built the road so that only its wheels fit the asphalt. Decart repaves the road. We are already seeing the first cracks in that pavement. In its recent public disclosures, AI lab Anthropic revealed that bringing a model up on rival AMD chips used to take six to twelve months and dozens of engineers. Its latest Claude Mythos model did it in about three days by writing almost all of the code autonomously. That shift changes the economics of the entire cloud buildout. Amazon is not trying to become a traditional merchant chipmaker. It is buying the software layer that turns everyone else’s silicon into a generic commodity. If software can strip away CUDA's friction, then the massive price premium Nvidia charges stops looking like an impenetrable moat and starts looking like an avoidable expense. That dynamic is worth more to Amazon than to anyone else on Earth, because Amazon designs its own chips. Amazon's in-house silicon business—spanning Graviton, Nitro, and Trainium—is already massive. The company says the unit surpassed a $25 billion annualized run rate earlier this year. Chief Executive Andy Jassy has pointed to more than $225 billion in revenue commitments for Trainium, arguing the business would be worth roughly $50 billion if priced like an independent supplier. During an on-stage demo at AWS re:Invent, Decart showcased real-time video running on Amazon's Trainium3 hardware, generating frames four times faster than graphics processors at half the cost. For equity investors, this deal cuts straight to the core of Amazon's valuation. Amazon trades at roughly 22 times forward earnings while guiding toward a staggering $220 billion in cash capital expenditures this year. Wall Street tolerates that historic spending spree because AWS generates a sparkling 38.7% operating margin. Every time an AWS client trains a model on an in-house Trainium chip instead of renting an expensive Nvidia processor, Amazon keeps that margin for itself. Lowering the switching cost for customers converts a costly hardware purchase into high-margin proprietary software. The bear case, however, writes itself. Amazon would be paying an implied multiple comfortably north of 150 times estimated revenue for a utility whose entire value rests on being completely neutral. Decart generated only low tens of millions in estimated revenue, primarily from specialized semiconductor consulting projects. Cloud customers embraced Decart precisely because they did not want to be held hostage by any single vendor. Stick that software inside AWS, and independent AI labs will immediately question its neutrality. Furthermore, chip portability is a double-edged sword: if AI code moves effortlessly across hardware, AWS's rented servers become just as replaceable as Nvidia's chips. Translation layers also routinely sacrifice throughput, and Nvidia still commands an estimated 70% to 75% of the accelerator market. If Decart's technology fails to scale outside of Amazon's walled garden, this reported $7.5 billion deal quickly turns into an eye-watering acqui-hire for three founders. What we are watching closely is whether Amazon keeps Decart's stack fully optimized for Nvidia and Google silicon if a transaction closes. If Amazon preserves open cross-chip support, it is establishing itself as the default software tollbooth for flexible enterprise computing.

Disclosure: The Signal holds no position in AMZN. Positions may change. This is not financial advice.