Amazon aims to sell Trainium AI chips beyond its own cloud

Amazon Pushes to Sell Trainium AI Chips Beyond Its Own Cloud

Amazon is moving closer to selling its proprietary Trainium AI chips directly to outside companies, with Peter DeSantis, the company’s senior vice president overseeing AI, custom silicon, and quantum computing, reinforcing that ambition at VivaTech 2026 in Paris this week.

DeSantis, who took the helm of Amazon’s combined AI and chip division in December 2025, appeared at the conference on Tuesday, where he discussed the company’s strategy of co-designing chips and AI models as a competitive advantage over rivals. In a CNBC interview the same day, DeSantis outlined Amazon’s goal of competing directly with frontier AI labs within a year.timesofai

From Cloud-Only to Chip Seller

The push to sell Trainium hardware externally builds on statements CEO Andy Jassy made in his April 2026 shareholder letter. “There’s so much demand for our chips that it’s quite possible we’ll sell racks of them to third parties in the future,” Jassy wrote. During Amazon’s first-quarter earnings call in late April, Jassy provided a clearer timeline, saying there was “a good chance” the company would begin offering Trainium racks outside AWS “over the next couple of years”.qz

Amazon’s chip unit — spanning Trainium AI accelerators, Graviton processors, and Nitro networking cards — now carries an annual revenue run rate exceeding $20 billion, growing at triple-digit percentages year over year. Jassy estimated the unit’s standalone value could reach $50 billion if it sold to both AWS and external customers.aboutamazon

Challenging Nvidia on Its Own Turf

DeSantis’s emphasis at VivaTech on the “flywheel” between chip design and model development underscores Amazon’s argument that it occupies a unique position in the AI supply chain. Unlike companies that simply buy GPUs, Amazon designs its own silicon through subsidiary Annapurna Labs and trains its own Nova models on that hardware — a vertical integration DeSantis frames as essential for the next wave of AI breakthroughs.timesofai

Trainium2 offers roughly 30 percent better price-performance than comparable GPUs and is “largely sold out,” according to Amazon, while Trainium3, which began shipping in early 2026, delivers an additional 30 to 40 percent improvement and is “nearly fully subscribed”. Major customers already using Trainium through AWS include Anthropic, which runs its Claude models on over one million Trainium2 chips, and Uber, which began trialing Trainium3 in April.networkworld

What Comes Next

No specific external clients or timeline for direct chip sales have been publicly named. But the trajectory is clear: Amazon is positioning itself not merely as a cloud provider that happens to make chips, but as a semiconductor company in its own right — one prepared to compete with Nvidia for customers who want to deploy AI hardware in their own data centers.businessinsider