Amazon CEO Andy Jassy’s annual shareholder letter reads something like a Kendrick Lamar diss track, if the rapper was a corporate-speak-talking CEO and not a poetic Pulitzer-prize winning musician.
Meaning, you have to know the history to understand all of the competitors Jassy takes aim at, alongside cute personal stories about his unrealized dream of being a sportscaster and watching hockey games with his dad.
Of course, Jassy doesn’t throw the gauntlet down directly. He takes a more nuanced approach. For instance, in his challenge to Nvidia, he writes, “We have a strong partnership with NVIDIA, will always have customers who choose to run NVIDIA” and will always support these chips in its cloud.
But he also says: “Virtually all AI thus far has been done on NVIDIA chips, but a new shift has started.” AWS customers, he says, “want better price-performance” meaning Amazon’s own home-grown Trainium AI chips.
Jassy says demand is so high for this chip that capacity for the newest one, Trainium3, is nearly sold out. Remarkably, he says that capacity is also nearly sold out for Trainium4, which is still 18 months away from being available.
This means that Amazon’s chip business has hit a $20 billion annual revenue run rate. But if Amazon were a chipmaker that sold its wares to others, it would be at $50 billion ARR, he postulates.
Granted, Nvidia did $215.9 billion in actual revenue last year. Nvidia may not be shaking in its boots, yet. Still, Jassy presents Trainium as a formidable up-and-comer.
Techcrunch event
San Francisco, CA
|
October 13-15, 2026
Jassy didn’t spare Intel either. He points out that AWS’s homegrown Graviton CPU, a competitor to the Intel x86 architecture, “is now used expansively by 98% of the top 1,000 EC2 customers,” aka some of the biggest companies in the world. Two companies even asked to “buy all of our Graviton instance capacity in 2026,” he writes (emphasis his). “We can’t agree to these requests given other customers’ needs, but it gives you an idea of the demand.”
He promised that Amazon’s Starlink competitor, Amazon Leo, scheduled to launch in mid-2026, is already succeeding, too. It’s won contracts from Delta Airlines, AT&T, Vodafone, Australia’s National Broadband Network, and NASA, among others.
Interestingly, he also said Amazon could be looking at selling robotics one day. It may turn all the data from its 1 million warehouse robots into “robotics solutions” for industrial uses and consumers, he wrote. Is there an Amazon humanoid in our future? We’ll see. He talked up other Amazon businesses, too, like same-day delivery, groceries, and drones.
But mostly, Jassy tried to make the case for the hundreds of billions of dollars of capital expenditures he’s committed. In February, he announced plans to spend $200 billion in 2026 on capex, mostly building out AWS data centers. That’s more than any of the other major tech companies, which are also spending big on capex. Jassy’s pitch to shareholders makes sense, considering Amazon’s stock plunged to below $200 a share and hasn’t recovered.
“We’re not investing approximately $200 billion in capex in 2026 on a hunch,” he wrote, using as an example that his deal with OpenAI included the model maker pledging to spend $100 billion on AWS. Of course, there are those who doubt OpenAI will meet all of its spending promises.
In a nod to that, Jassy insists that beyond OpenAI, “there are several other customer agreements completed (and unannounced), or deep in process,” lined up to buy the AWS capacity.
We’ll have to wait and see. Those who cause a bubble are never the ones who see (or admit to) its existence. “I’ve followed the public debate on whether this technology is over-hyped, whether we’re in ‘a bubble.’” But he declares in this letter that, for Amazon at least, this isn’t the case.
Correction: This story has been updated to reflect that the estimated ARR of Amazon’s custom chip business includes multiple chips beyond Trainium.