Amazon CEO Andy Jassy says Trainium beats Nvidia on value. Here's what that means for Canadian investors with money in the market

"Virtually all AI thus far has been done on Nvidia chips, but a new shift has started," Jassy wrote. "We have a strong partnership with Nvidia, will always have customers who choose to run Nvidia, and we will continue to make AWS the best place to run Nvidia. However, customers want better price-performance. We've seen this movie before."

The "movie" Jassy is referencing is Amazon Web Services' (AWS) experience unseating Intel in cloud computing. In 2018, Amazon developed its own central processing unit (CPU) chip, called Graviton (1). "In the CPU space, virtually all of the workloads ran on Intel chips until we invented Graviton in 2018," he wrote. "Graviton, which has up to 40% better price-performance than other x86 processors, is now used expansively by 98% of the top 1,000 EC2 customers."

"The same story arc is unfolding in AI," he added.

The AI chip business: By the numbers

Jassy laid out the Trainium roadmap in detail. Amazon's second-generation AI chip, Trainium2, delivered about 30% better price-performance than comparable GPUs and "has largely sold out (1)." Trainium3, which began shipping at the start of 2026, offers another 30% to 40% improvement over its predecessor and is nearly fully subscribed. A significant portion of Trainium4 — still about 18 months from broad availability — has already been reserved by customers.

Amazon's custom chips business — which includes Graviton, Trainium and its Nitro networking cards — has hit an annual revenue run rate above US$20 billion (C$27.8 billion) (1). That figure has doubled from the US$10 billion (C$13.9 billion) Amazon disclosed alongside its Q4 2025 earnings and is growing at triple-digit percentages year over year, according to the company's CEO.

"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 ~US$50 billion (C$69.6 billion)," Jassy wrote.

For context: Nvidia reported record full-year revenue of US$215.9 billion (C$300 billion) for its fiscal year ending January 2026, with data centre sales climbing 75% year-on-year to US$62.3 billion in Q4 alone (2).

At scale, Jassy wrote, Trainium is expected to save Amazon "tens of billions of capex dollars per year, and provide several hundred basis points of operating margin advantage versus relying on others' chips for inference (1)." Amazon Bedrock, AWS's fast-growing inference service, already "runs most of its inference on Trainium."

"Our chips business is on fire," Jassy wrote, adding it "changes the economics for AWS, and will be much larger than most think."

Tired of high commissions eating your returns? Compare Canada’s top discount brokerages and switch to a $0-commission platform today.

Who's actually using Amazon's AI chips?

The strongest evidence for Jassy's claims may be the customer list. Anthropic, the AI safety company behind the Claude AI assistant, is the most prominent Trainium customer (3). In April 2026, Anthropic revealed it is using 500,000 Trainium2 chips as part of Project Rainier — a massive AI compute cluster spread across multiple data centres — and that its models would scale to more than 1 million Trainium2 chips for training and inference (4).

According to a TechCrunch tour of Amazon's Trainium lab published in March, there are now 1.4 million Trainium chips deployed across all three generations, and Anthropic's Claude runs on over 1 million of the Trainium2 chips deployed (5).

OpenAI has also signed on in a commitment to AWS at a cost of more than US$100 billion, which Jassy referenced in the letter, making Amazon the exclusive infrastructure provider for parts of OpenAI's workloads (6). Apple is reportedly testing the chips as well to drive smart glasses similar to Meta's Ray-Ban Meta, according to TechCrunch (7).

The relationship with Anthropic goes beyond a typical cloud customer arrangement. Anthropic engineers had direct input into the chip's instruction set architecture, and Amazon agreed to open up its instruction set. This removed a pain point that Anthropic's engineers had experienced with Nvidia GPUs, where the company tries to obscure that information to keep competitors from seeing it, Semafor reported last month (8).

Nvidia's position — and its challengers

Despite Jassy's enthusiasm, Nvidia's position in AI is still considerably stronger. The company holds an 81% market share by revenue for data centre chips, according to IDC research (9). Bloomberg also reports Nvidia's CEO Jensen Huang recently projected US$1 trillion in cumulative AI chip revenue through 2027 (10). Nvidia's Blackwell Ultra chips are also the clear performance leader on several benchmarks (11).