Amazon commits $25 billion to custom AI chip development, accelerating hyperscaler vertical integration into silicon.
Amazon's in-house processors have grown fast enough to make the company a serious rival to Nvidia, even though Amazon remains one of Nvidia's largest customers. The company reported in its latest earnings that its custom chips business has reached an annual revenue run rate of $25 billion. This milestone comes as Nvidia prepares to enter the CPU segment, where Amazon already has an established foothold.
The growth is concentrated in two product lines. Graviton, Amazon's general-purpose CPU, and Trainium, its AI accelerator, are both delivering triple-digit growth. By contrast, Nvidia reported quarterly revenue of $96 billion in its most recent period, underscoring the significant scale gap Amazon still faces.
The central question is whether cheaper custom chips will cannibalize demand for Nvidia's premium hardware or expand the overall market. Evidence so far suggests expansion. Yet Amazon's stated interest in possibly spinning its chip work into a standalone business indicates ambitions that extend beyond serving its own cloud infrastructure.
**The Dual Position**
Amazon Web Services, the world's largest cloud provider, sells access to Nvidia's top GPUs to customers running AI workloads. CEO Andy Jassy underscored this commitment on the company's recent earnings call: "We'll continue making AWS the best place to run Nvidia chips." Simultaneously, AWS offers its own silicon on the same platform, allowing customers to choose Nvidia hardware for some workloads and Amazon's chips for others, often within the same cloud account. Few companies hold such a dual position in the industry.
**The Competitive Angle**
Amazon is not competing on raw performance. Its strategy centers on price. Graviton offers at least 30 percent better price performance than competing CPUs. Trainium is marketed as a lower-cost option for training and running AI models. For cost-conscious customers moving from experimentation to large-scale deployment, this positioning carries real weight. In his shareholder letter earlier this year, Jassy signaled that Amazon's chip operation could eventually become a standalone business—a notable shift from its historical focus on building hardware primarily to support its own services.
**The Lineup**
AWS introduced its first Graviton processor in 2018 and later added Trainium for AI training workloads. Both are designed in-house and available primarily through AWS rather than sold as standalone parts to external buyers. Amazon has also built large Trainium-based computing clusters for AI developers, including Anthropic, reflecting these chips' growing role in frontier model development.
**The CPU Front**
The timing of Nvidia's CPU push matters significantly. CPUs are the general-purpose chips found in every computer and are increasingly viewed as central to agentic AI, in which software systems autonomously carry out multistep tasks. Many in the industry regard agentic AI as the next major growth driver. This positions Nvidia's upcoming CPU in direct competition with Graviton, which already commands an established customer base within AWS. For the first time, Nvidia faces a segment where the newcomer position belongs to them.
**Pressure on Inference**
Amazon has argued that a wider array of chip options should drive down AI inference costs—the process of running trained models to produce answers. Lower prices would benefit customers and could pressure a company like Nvidia, whose business depends on premium pricing. Amazon's value-focused approach targets customers seeking cheaper alternatives.
**Why Nvidia Remains Dominant**
Despite Amazon's momentum, there is little evidence that its gains come at Nvidia's expense. Two factors explain this. First, demand for AI computing remains robust enough that no single supplier can meet it. As companies deploy AI in real products, that demand shows signs of holding. Second, the chips are not direct substitutes. Customers requiring maximum performance for certain workloads still choose Nvidia, while price-sensitive projects may run on Amazon hardware. Many large buyers use both.
The likeliest near-term scenario is continued growth for both companies. Amazon may not displace Nvidia at the market's top, but its breadth across CPUs and accelerators, combined with direct access to millions of cloud customers, makes it arguably a more consequential challenger than traditional chipmakers such as AMD.