What did Amazon's CEO say about selling chips directly?
Amazon CEO Andy Jassy said on the company's Q1 2026 earnings call that Amazon may begin selling its Trainium AI chips directly to third parties — not just offering them through AWS cloud services. That would put Amazon in direct competition with chip companies like Nvidia, according to About Amazon.
Jassy framed the move as a natural next step given the scale Amazon has already reached. "If our chips business was a stand-alone business, and sold chips produced this year to AWS and other third parties (as other leading chips companies do), our annual run rate would be ~$50 billion," he said.
How big is Amazon's chip business right now?
Amazon's chip business currently carries an annual revenue run rate of over $20 billion. That figure reflects chips used internally across AWS — not direct third-party sales.
In Q1 2026, the business saw nearly 40% quarter-over-quarter growth. Year-over-year growth is running at triple-digit percentages. Jassy said the business is now one of the top three data center chip businesses in the world.
Which chips is Amazon selling — and how do they compare?
Trainium is Amazon's custom AI training chip, designed to compete with Nvidia GPUs on price and performance. Here's where each generation stands:
| Chip | Status | Price-Performance vs. Prior Gen |
|---|---|---|
| Trainium2 | Largely sold out | ~30% better than comparable GPUs |
| Trainium3 | Nearly fully subscribed | 30–40% better than Trainium2 |
| Trainium4 | Much already reserved | ~18 months from broad availability |
Trainium3 started shipping at the start of 2026. Trainium4 is still roughly 18 months from broad availability, yet much of its capacity has already been reserved.
You might also like
Who has committed to using Amazon's Trainium chips?
Amazon has secured over $225 billion in total revenue commitments for Trainium. The two largest customers are Anthropic and OpenAI, both of which have signed multi-year, multi-gigawatt deals. Uber has also committed to Trainium.
On the CPU side, Meta has committed to using tens of millions of Graviton cores. Graviton is Amazon's custom CPU chip, which Jassy says delivers up to 40% better price-performance than x86 processors. It is now used by 98% of the top 1,000 EC2 customers.
Amazon Bedrock — Amazon's managed AI model service — runs most of its inference on Trainium. Almost 80% of Fortune 100 companies use Bedrock, and over 125,000 customers use it in total.
What is Amazon SVP Peter DeSantis saying about AI's future?
Amazon AI Senior Vice President Peter DeSantis spoke at VivaTech 2026 in Paris on June 17. He argued that AI's most important breakthroughs have not yet happened, as reported by Times of AI.
DeSantis said current AI systems still need "a couple more orders of magnitude of development" before they fully transform work and innovation. He pointed to three specific gaps: real-time human interaction, deep reasoning, and tighter hardware-software integration.
He described the relationship between chip design and model development as a "flywheel." Better chips enable better models, which in turn reduce costs and improve efficiency. He said this loop is underrated across the AI industry. DeSantis also noted that AI systems need to operate on a 40-millisecond clock to match human interaction speeds — a target that current systems have not yet met.
DeSantis has spent 27 years at Amazon and was involved in the company's development from the early days of EC2. He also helped lead the acquisition of Annapurna Labs.
Why does the hardware-software gap matter for AI builders?
DeSantis's core argument is that chip designers and model developers working in isolation slow down progress. When model teams don't know what the next chip generation can do, they can't design for it — and gains are lost at launch.
Here's what we know so far: Amazon is betting that tight coupling between its Trainium hardware and the models trained on it is a structural advantage, not just a marketing claim. That's the logic behind both the flywheel argument and the scale of Trainium pre-orders.
For builders and founders tracking the AI chips market, the Trainium roadmap is now a credible alternative to Nvidia's GPU stack — especially given the price-performance numbers Jassy cited publicly.
Amazon's position also matters for the custom silicon race playing out across the industry. Competitors are moving fast, and Amazon's $225 billion in Trainium commitments signals that hyperscaler-designed chips are no longer a niche option.
The Nvidia vs. Amazon chip dynamic is also shifting. Nvidia has long dominated AI training hardware, but Amazon's Trainium2 sell-out and Trainium3 near-full subscription suggest real demand — not just internal captive use.
For context on how other chip players are positioning, see coverage of Nvidia's Texas factory and Nvidia's Vera CPU push into new markets.
What happens next?
The most concrete near-term milestone from the sources: Trainium4 is approximately 18 months from broad availability, and much of its capacity is already reserved. Amazon has not announced a formal date for launching a standalone chip sales business, but Jassy's Q1 2026 earnings comments put the idea on the record publicly for the first time.
0 Comments
Log in to comment
Not a member yet? Join the community
Pick a meme
KlipyHave a great take?
Drop your email — we'll send a magic link so you can post it. No password.
Not a member of the community? Join today.
Join the community →