Key Takeaways
- Nvidia still leads the AI‑chip market with its GPUs, CUDA software, and NVLink ecosystem, but custom accelerators are eroding its share.
- Amazon’s in‑house AI chips (Trainium for training, Inferentia for inference) have surged from 1.4 % of cumulative AI‑chip shipments in Q1 2024 to 7.5 % by Q4 2025, a more‑than‑five‑fold increase.
- CEO Andy Jassy disclosed that Amazon’s chip business is now at an annual run‑rate above $20 billion and growing at triple‑digit year‑over‑year rates; as a stand‑alone seller it could generate roughly $50 billion in revenue.
- Demand for the next‑generation Trainium3 is so strong that nearly all of its 2026 capacity is already booked, and early interest in Trainium4 suggests continued upside.
- Amazon’s custom Graviton CPUs are now used by 98 % of its top 1,000 AWS customers, up from zero in 2018, positioning the chip line as a core differentiator for AWS.
- By deploying its own chips, Amazon expects to save tens of billions of dollars in capex annually, improve AWS margins, and accelerate earnings despite its planned $200 billion capital‑expenditure spend for the year.
The AI‑Chip Landscape: Nvidia’s Dominance and Emerging Challengers
Nvidia (NVDA 1.96 %) remains the clear leader in the artificial‑intelligence chip arena. Its all‑purpose GPUs, coupled with the ubiquitous CUDA software platform and the high‑speed NVLink fabric, have become the default choice for large‑scale data‑center customers training and inferencing AI models. Yet the market is not a monopoly. A growing share of compute demand is flowing toward custom AI accelerators built by hyperscalers and specialist chip firms. These purpose‑built silicon solutions aim to deliver better performance‑per‑dollar for specific workloads, offering cloud giants a lever to cut costs and differentiate their platforms.
Amazon’s Chip Business: From Niche to a Multi‑Billion‑Dollar Engine
Among the hyperscalers, Amazon has turned its custom‑silicon effort into a standout growth story. The company’s AI‑focused chips—Trainium for training and Inferentia for inference—have seen their share of cumulative AI‑chip shipments jump from 1.4 % in Q1 2024 to 7.5 % by Q4 2025, according to estimates from Epoch AI. That trajectory represents a more‑than‑quintupling of Amazon’s presence in just two years.
Leadership Insight: Jassy’s Perspective on Scale and Opportunity
Amazon’s leadership has been vocal about the momentum. CEO Andy Jassy noted that the firm’s AI services are expanding at a triple‑digit rate, and he revealed that the chip business itself has reached an annual run‑rate exceeding $20 billion in Q1. He elaborated, “If it were a stand‑alone business selling chips directly to other cloud providers, Jassy said it would generate $50 billion in revenue.” This framing underscores the sheer economic scale that Amazon’s custom silicon could achieve if it were offered to external customers.
Third‑Party Sales: A Potential Next Step
While Amazon currently uses its chips primarily to power AWS, Jassy hinted at the possibility of selling Trainium and Inferentia to third parties. He observed that only a handful of major AI labs—such as OpenAI or Anthropic—might be interested, and that dealing with direct cloud rivals remains unlikely. Nevertheless, a partnership with a leading AI research organization could dramatically lift production volumes for Trainium and Inferentia, turning Amazon’s internal advantage into an external revenue stream.
Custom CPUs: Graviton’s Role in the Agentic AI Wave
Beyond AI accelerators, Amazon’s Graviton line of central processing units is gaining traction. Meta recently committed to deploying tens of millions of Graviton CPU cores, citing the newest generation as purpose‑built for the demands of agentic AI. Jassy shared that the custom CPU is now used by 98 % of Amazon’s top 1,000 AWS customers, a remarkable rise from zero adoption in 2018. As reasoning models and autonomous agents proliferate, Graviton’s efficiency could become a decisive factor for workloads that rely heavily on general‑purpose compute alongside AI accelerators.
Trainium3 and Trainium4: Capacity Already Locked In
The latest iteration, Trainium3, began shipping this year, and demand is already outstripping supply. Jassy reported that “nearly all of its 2026 capacity [is] sold as of the end of March.” Looking ahead, Trainium4, slated for release in roughly 18 months, is already attracting very strong interest, indicating that Amazon’s pipeline remains robust and that the company can continue to scale production to meet escalating AI workloads.
Financial Impact: Capex Savings, Margin Expansion, and Earnings Acceleration
The strategic shift to proprietary silicon is not merely a technical endeavor; it carries concrete financial benefits. In his shareholder letter, Jassy wrote, “At scale, we expect Trainium will save us tens of billions of capex dollars per year.” This projection is especially relevant given Amazon’s announced plan to spend $200 billion on capital expenditures in the current fiscal year. By sourcing chips at near‑cost rather than paying retail prices for Nvidia GPUs, Amazon anticipates improved margins on AWS revenue. The combination of lower capex, higher hardware efficiency, and expanding AI‑service uptake should translate into strong earnings acceleration for shareholders.
Broader Implications: A Diversifying AI‑Chip Ecosystem
Amazon’s progress illustrates a broader trend: the AI‑chip market is diversifying away from a single‑source dominance. While Nvidia’s GPUs will likely remain essential for many training tasks, the rise of hyperscaler‑specific accelerators—Google’s TPU, Microsoft’s Maia, Meta’s MTIA, and Amazon’s Trainium/Inferentia—signals that cloud providers are seeking cost‑effective, workload‑tailored silicon. This shift could reshape purchasing patterns, encourage more competition, and ultimately drive down the total cost of AI infrastructure across the industry.
In summary, Amazon’s custom chip business has evolved from a modest internal project to a multi‑billion‑dollar growth engine that threatens to erode Nvidia’s grip on AI hardware. With soaring demand for Trainium and Inferentia, rapid adoption of Graviton CPUs, and clear pathways to external sales and massive capex savings, Amazon is well positioned to continue expanding its silicon footprint—and to reap substantial financial rewards in the process.
https://www.fool.com/investing/2026/04/30/meet-the-company-thats-quintupled-its-share-of-ai/