stocksoncrypto

Amazon is exploring selling its in-house AI chips, including AWS Trainium, directly to data centres and enterprise customers.
Amazon $AMZN runs a three chip AI stack: Trainium for training, Inferentia for inference and Graviton CPUs for general compute.
Trainium has been restricted to AWS infrastructure until now, the reported plan would extend availability to external buyers.
This would mark Amazon's shift from a cloud provider using custom chips to a vertically integrated AI compute supplier, offering both cloud services and the hardware that powers them.
Nvidia's business model depends on every company needing to buy GPUs from Nvidia, regardless of which cloud they use. If Amazon starts selling Trainium directly to data centres, a company no longer needs AWS or Nvidia.
They can buy Amazon's chips and run them on their own infrastructure. That removes Nvidia from the transaction entirely.
Amazon's Trainium AI accelerator has already attracted major customers including OpenAI, Anthropic and Uber, all accessing it via AWS.
Anthropic has committed to up to 5 gigawatts of current and future Trainium capacity.
OpenAI has agreed to about 2 gigawatts. These are workloads that would otherwise likely run on Nvidia GPUs.
Amazon said in April that the Trainium chip has generated over $225 billion in revenue commitments. Trainium2 has largely sold out.
Trainium3 began shipping at the start of 2026 and is nearly fully subscribed. Much of Trainium4, still about 18 months from broad availability, has already been reserved.
That level of demand shows real customers are choosing Trainium over Nvidia GPUs even while restricted to AWS only.
Andy Jassy said in his April shareholder letter that the chip business is generating roughly $20 billion annualized across Trainium, Graviton, and Nitro, growing at a triple-digit percentage rate year over year.
He said that if the chip business sold externally the same way Nvidia does, it would represent a $50 billion annual run-rate business on its own.
Peter DeSantis, the AWS executive overseeing the chip effort, confirmed the talks, citing demand for AI compute to stay under local control, especially Europe's push for sovereign infrastructure. He dismissed concerns about cannibalizing AWS cloud revenue, citing heavy underconsumption of AI compute industry-wide.
This follows Alphabet's move in April, when CEO Sundar Pichai said Google would start making its TPUs available to outside data centres through a limited customer program.
Two of the three largest cloud providers on earth are now independently moving to sell their own AI chips outside their own cloud walls. That is the part that should worry Nvidia most.
It is not one competitor. It is the entire hyperscaler layer simultaneously deciding they no longer need to be Nvidia's customer exclusively.
Nvidia $NVDA is still the dominant force today.
Data center revenue rose 92% year over year to a record $75.2 billion last quarter, and Amazon itself remains one of Nvidia's largest customers, committing to deploy more than 1 million Nvidia GPUs starting in 2026.
The two largest cloud companies on earth are now actively building the infrastructure to compete with Nvidia directly, using customers Nvidia currently depends on.
#StocksOnCrypto