Google and Amazon plan to sell custom AI chips directly, raising competitive pressure on Nvidia’s ecosystem.
Nvidia’s AI chip powerhouse is facing a new kind of pressure, and it could come from the very customers that helped build its lead.
For years, Nvidia stock has benefited from a simple demand story: the AI industry needs its chips.. But as Google and Amazon reported earnings. both pointed to ambitions to move beyond supplying their custom silicon only through their own cloud services.. Until now. customers could access Google’s TPUs and Amazon’s Trainium chips through those platforms rather than owning the hardware themselves.
In this context, the strategic shift is significant because it changes who controls the customer relationship. If large cloud players begin offering chips more directly, Nvidia’s position stops looking like an unavoidable default and starts looking like one option among several.
Nvidia’s standing remains strong, and both Google and Amazon have said they will continue to work with Nvidia.. Still. Amazon’s CEO Andy Jassy suggested a path toward selling full racks of Trainium chips beyond its own cloud “over the next couple of years. ” framing it as part of a broader transformation in how AI compute is delivered.. Google signaled an even more immediate direction. with its CEO Sundar Pichai saying the company plans to provide TPU chips to a select group of customers in their own data centers this year. while noting that most revenue from those sales may arrive later.
This is where the tension with Nvidia could deepen: Google and Amazon are both large buyers of Nvidia chips today. using them as building blocks inside their cloud offerings.. The move toward direct chip delivery could gradually reduce how much of that demand is funneled through Nvidia. even if cooperation continues.
Breaking into a dominant chip ecosystem is rarely straightforward.. Analysts cited by Misryoum noted that Nvidia has built an ecosystem that extends beyond hardware into software and ongoing support. making adoption easier for many enterprises.. They also argued that selling chips is not the same as providing access. because buyers often need onboarding. education. and service layers that go with the product. not just the silicon itself.
There is also the issue of customization.. Cloud giants tend to tailor their systems for their own infrastructure. which can make their chip offerings feel less like a mass-market commodity and more like a specialized solution.. Meanwhile. the broader AI chip market is not necessarily zero-sum: some AI developers increasingly use more than one supplier. depending on performance needs. cost. and availability.
A final factor adding momentum to this competition is the evolving focus inside AI workloads.. As inference becomes more central to how models are used after training. both Google and Amazon are positioning their custom chips around those practical deployments.. For Nvidia. the challenge is not only competing on chips. but staying embedded in the full stack as customer preferences and infrastructure strategies shift.
This matters for the industry because custom silicon offerings can reduce reliance on a single supplier and reshape bargaining power between chip leaders and the largest cloud platforms.. Over time, the market could become more diversified, with buyers weighing multiple pathways to build and run AI systems.