Amazon's Chip Business Is Now a $20 Billion Threat to Nvidia
Amazon has quietly built a chip business big enough to matter. Its custom silicon, the Trainium AI accelerators, Graviton processors and Nitro chips, now runs at more than 20 billion dollars a year, and CEO Andy Jassy says it could be worth 50 billion and may be sold to outside customers. It is one of the most credible challenges to Nvidia yet.
Amazon has quietly built a chip business big enough to matter. Its custom silicon, the Trainium AI accelerators, Graviton processors and Nitro chips, now runs at more than 20 billion dollars a year, and CEO Andy Jassy says it could be worth 50 billion and may be sold to outside customers. It is one of the most credible challenges to Nvidia yet.
The point is to cut out the Nvidia tax. Amazon designs Trainium specifically to train and run AI models on its cloud at lower cost, and it says the chips offer roughly 30 percent better value than comparable Nvidia hardware. When a company runs as many AI workloads as Amazon, even a modest saving per chip adds up to billions, and owning the design means it is not at the mercy of one supplier.
Demand backs up the ambition. Trainium2 has largely sold out, and the newer Trainium3, shipping this year with another big efficiency gain, is nearly fully subscribed. Amazon has lined up huge multi-gigawatt commitments from frontier labs, including capacity for OpenAI and a vast deal with Anthropic that involves spending around 100 billion dollars on AWS over a decade. The chips are being used, not just announced.
Selling them externally would be the bigger shift. Until now these chips powered Amazon's own cloud, but Jassy hinting that Amazon could sell them to other customers would turn an internal cost-saver into a direct Nvidia competitor. Google has made similar moves, and a world where the cloud giants sell their own AI chips is a real threat to Nvidia's near-monopoly. The buyers are becoming sellers.
The honest caveat is that Nvidia is still far ahead. Its chips remain the default for the most demanding AI work, its software ecosystem is years deep, and custom silicon tends to handle specific tasks rather than the whole market. A 20 billion dollar business is large but small next to Nvidia's scale, and selling chips externally is a hard, unproven step. This is pressure on Nvidia, not a dethroning.
So Amazon has turned in-house chips into a real business and is eyeing an even bigger one, adding to the pile of companies trying to break Nvidia's grip. A 20 billion dollar run rate, a 50 billion ambition, and a hint at selling to all comers. The race to own AI silicon is widening fast. Watch whether Amazon actually sells its chips outside its own cloud.