By Ammar Sabilarrohman 
In a move that signals a seismic shift in the artificial intelligence hardware landscape, Amazon Web Services (AWS) is reportedly exploring a strategy that would bring it into direct, open competition with Nvidia. For years, the cloud giant has relied on its proprietary "Trainium" chips to power internal workloads and offer unique, cost-effective compute options to its cloud customers. Now, Amazon is actively in talks to sell these AI-optimized chips to third-party companies, potentially disrupting the industry’s current hierarchy and challenging Nvidia’s near-monopoly on the high-performance AI accelerator market.
The Strategic Shift: From Cloud Exclusive to Hardware Vendor
The narrative surrounding AWS’s chip strategy has evolved rapidly. Historically, AWS maintained a "walled garden" approach regarding its custom silicon. By keeping Trainium chips exclusive to its cloud environment, Amazon ensured that customers remained tethered to its ecosystem, creating a lucrative "waterfall effect" of revenue. When a client uses Trainium, they aren’t just paying for compute cycles; they are also utilizing AWS’s integrated suite of storage, networking, security, and monitoring services.
However, in a recent interview with Bloomberg, Peter DeSantis, Amazon’s senior vice president of AWS utility computing, confirmed that the company is currently evaluating the feasibility of selling its Trainium chips directly to other organizations for use in their own data centers. While the company maintains that these discussions are in the early stages, the pivot marks a fundamental change in Amazon’s philosophy: transitioning from a pure-play cloud provider to a merchant silicon vendor.
A Chronology of Ambition
The path to this potential shift has been marked by a deliberate and aggressive expansion of Amazon’s internal hardware capabilities.
- The Foundation: AWS began its foray into custom silicon years ago with the Graviton series of general-purpose CPUs, which proved that Amazon could design chips that outperformed industry standards in terms of energy efficiency and cost-to-performance ratios.
- The Trainium Breakthrough: Recognizing the surging demand for AI training, Amazon introduced Trainium, a specialized chip designed to reduce the cost and power consumption associated with training large language models (LLMs).
- The Shareholder Signal (April 2026): The most significant public indicator of this shift came in early April 2026, when Amazon CEO Andy Jassy released his annual shareholder letter. Jassy noted that if the company’s internal chips were treated as a standalone business, their production capacity would command an annual run rate of approximately $50 billion. He explicitly teased the possibility of selling "racks of them" to third parties, a statement that shifted the industry’s perception of AWS from a consumer of Nvidia GPUs to a burgeoning powerhouse of chip design.
- The Current Phase: As of June 2026, AWS executives have moved from hypothetical musings to active dialogues with potential partners, marking the beginning of a formal evaluation process regarding supply chain logistics and market viability.
Supporting Data: The Scale of the Challenge
To understand the magnitude of this potential disruption, one must look at the numbers. Nvidia, the current undisputed king of the AI hardware market, is operating at an extraordinary scale, with a revenue run rate approaching $326 billion.
Amazon’s estimated $50 billion run rate for its chip business, while significant, does not threaten to topple Nvidia overnight. However, it places Amazon in the same weight class as industry stalwarts like Intel, which has historically dominated the CPU market. By entering the market, Amazon is not merely selling a chip; it is providing a proven, high-performance alternative to Nvidia’s H100 and Blackwell architectures at a time when companies are desperate for supply chain diversification.
The Supply Chain Bottleneck
A critical factor in this transition is manufacturing capacity. AWS already struggles to keep up with internal demand for its chips. Jassy noted in April that current Trainium capacity is sold out almost instantly, and projections for the next generation—Trainium4—show similar demand levels well before the hardware even hits the market.
To scale, Amazon would need to secure significantly higher allocations from TSMC, the world’s premier semiconductor foundry. This creates a high-stakes standoff: Nvidia has recently supplanted Apple as TSMC’s largest customer, leaving little room for competitors to maneuver. For Amazon to succeed as a merchant chip vendor, it must demonstrate that it can secure the manufacturing throughput required to serve both its massive cloud footprint and a new cohort of external enterprise clients.
Official Responses and Corporate Stance
AWS has been careful to frame this potential shift as an evolution of its customer-centric model rather than a reactionary move against Nvidia. In a statement provided to TechCrunch, AWS spokesperson Doron Aronson reiterated the company’s official position: "While we’ve historically declined requests to sell chips directly, Andy noted it’s quite possible we’ll sell racks of them to third parties in the future."
This measured language suggests that Amazon is conducting a rigorous cost-benefit analysis. The company is weighing the immediate, high-margin revenue of selling hardware against the long-term, compounding revenue of keeping those customers within the AWS cloud ecosystem. The "waterfall effect"—where the hardware acts as a gateway to secondary services—remains a powerful incentive to keep chips exclusive.
Implications for the Industry
The entry of Amazon into the merchant chip market carries profound implications for the future of AI infrastructure.
1. The Diversification of AI Hardware
For years, the industry has been characterized by "Nvidia-or-bust" infrastructure. By introducing a high-performance alternative, AWS provides enterprise customers with leverage. If Amazon can provide a cost-effective, readily available alternative to the GPU-heavy architectures currently favored by most data centers, it will inevitably force a competitive response from Nvidia, potentially leading to lower prices or accelerated innovation cycles.
2. The Cloud-Hardware Convergence
We are witnessing a blurring of lines between cloud service providers (CSPs) and semiconductor designers. Just as Nvidia has moved into CPU territory—with CEO Jensen Huang recently outlining a $200 billion market opportunity in specialized CPUs—Amazon is moving into the silicon foundry ecosystem. The market is becoming a battle of "full-stack" players: companies that control the hardware, the software stack, and the infrastructure platform.
3. Impact on OpenAI and Anthropic
The relationship between model builders and hardware providers is also shifting. AWS has recently integrated OpenAI’s models into its cloud service, joining its existing partnership with Anthropic. As Amazon designs its chips to be increasingly optimized for these specific models, the "Trainium" brand could become the standard for the next generation of generative AI, effectively creating a "walled garden" that is optimized for specific architectural needs rather than just general-purpose compute.
Conclusion: A New Arena for Competition
The potential decision by Amazon to sell Trainium chips to third parties is a strategic masterstroke that forces a recalibration of the AI hardware market. While it remains to be seen if Amazon can navigate the complex supply chain constraints of TSMC and maintain its own cloud capacity requirements, the intent is clear: the company is no longer content to let Nvidia hold the keys to the kingdom of AI acceleration.
As Amazon prepares to move from a consumer to a supplier of high-end silicon, the industry is entering a new, more volatile phase of competition. For enterprises, the prospect of having a viable, high-performance alternative to Nvidia’s dominant GPUs offers a much-needed path to supply chain resilience. For Nvidia, the move represents a direct challenge to its dominance in the data center, turning one of its biggest customers into one of its most formidable competitors. The "Cloud Wars" are no longer just about software and service; they are, quite literally, being forged in silicon.
Lina Irawan 
Nila Kartika Wati 
Nana 