Amazon is signaling a potentially transformative shift in the artificial intelligence landscape. CEO Andy Jassy revealed that the company is considering selling its custom AI chips externally—a move that could position Amazon as a direct competitor to Nvidia and Advanced Micro Devices.

The implications extend beyond product strategy. This is a sign that the battle for AI infrastructure is entering a new phase—one where cloud providers are no longer just customers of chipmakers, but emerging rivals.

AWS Growth Highlights the Scale of AI Demand

Amazon’s push into AI hardware is grounded in surging demand within its cloud business. Amazon Web Services (AWS) has reached an AI revenue run rate exceeding $15 billion as of early 2026, reflecting rapid enterprise adoption of machine learning and generative AI tools.

Jassy emphasized that growth could be even faster if not for capacity constraints. Despite adding 3.9 gigawatts of infrastructure capacity in 2025—and planning to double that by 2027—AWS is still struggling to meet demand.

This supply-demand imbalance is a critical signal. It suggests that AI infrastructure is not just growing—it is outpacing even the largest providers’ ability to scale, creating opportunities for vertical integration.

Custom Chips Move From Internal Advantage to External Opportunity

Amazon has long developed its own silicon, including the Graviton CPU line and Trainium AI accelerators, to optimize performance and reduce reliance on third-party suppliers.

Now, that strategy appears to be evolving. The possibility of selling these chips externally indicates that Amazon sees its hardware not just as a cost-saving tool, but as a revenue-generating product in its own right.

Demand for these chips is already strong. The Trainium2 processor is reportedly sold out, while the newer Trainium3 is nearly fully subscribed shortly after launch. Additionally, major AWS customers have expressed interest in securing large portions of Amazon’s chip capacity.

This level of demand suggests that Amazon’s chips are competitive on performance and cost—two factors that are increasingly critical in AI deployment.

Price-Performance Pressure on Nvidia and AMD

Jassy’s comments also highlight a key tension in the AI market: cost. While Nvidia remains the dominant player in high-performance AI chips, its products come at a premium price.

Amazon’s strategy appears focused on improving price-performance ratios, offering customers more cost-efficient alternatives without sacrificing capability. If successful, this could erode Nvidia’s pricing power and reshape competitive dynamics across the semiconductor industry.

For AMD, which has been positioning itself as a challenger to Nvidia, Amazon’s entry introduces another layer of competition—this time from a vertically integrated cloud giant with massive internal demand.

Capacity Constraints Signal a Structural Shift

One of the most telling aspects of Amazon’s strategy is the ongoing capacity constraint. The fact that AWS cannot fully meet demand—even with aggressive expansion—underscores the scale of the AI boom.

It also explains why Amazon may move toward selling chips externally. By expanding its role in the supply chain, the company can both alleviate internal bottlenecks and capture additional value from the broader AI ecosystem.

This reflects a broader industry trend: companies are increasingly seeking control over critical infrastructure layers to secure growth and margins.

Outlook: A New Competitive Landscape in AI Infrastructure

Looking ahead, Amazon’s potential move into chip sales could significantly alter the competitive landscape. If cloud providers begin competing directly with traditional semiconductor companies, the boundaries between hardware, software, and infrastructure will continue to blur.

The success of this strategy will depend on execution—particularly Amazon’s ability to scale production, maintain performance competitiveness, and manage relationships with existing chip suppliers.

For now, the message is clear: the AI race is no longer just about software or models—it is about controlling the entire stack.