Amazon has dramatically scaled its commitment to Nvidia hardware. In a span of just five months the cloud giant moved from an agreement to deploy more than one million Nvidia GPUs to a total exceeding three million chips. The latest addition of two million units covers Nvidia’s Blackwell Ultra, Rubin and Rubin Ultra architectures with deliveries scheduled for 2027 and 2028.
But the move reveals far more than a simple hardware transaction. It signals the ferocious pace at which demand for artificial intelligence compute continues to outstrip even the most aggressive forecasts. And it underscores a stubborn reality. Even as Amazon pours resources into its own silicon designs it cannot do without the market leader’s processors.
The announcement landed during Nvidia’s quarterly earnings call in late August. Nvidia CFO Colette Kress laid out the details. “AWS is deploying an additional two million GPUs starting this quarter through the second quarter of FY29,” she said. She added that Amazon would adopt Nvidia’s full physical AI stack to power its fleet of warehouse robots.
Nvidia CEO Jensen Huang struck an optimistic tone. “NVIDIA and AWS have built one of the great growth engines of the AI era, and demand is running ahead of every forecast,” he stated according to the TechCrunch report.
The expansion builds directly on plans unveiled at Nvidia’s GTC conference in March. Back then Amazon Web Services committed to more than one million Nvidia GPUs starting in 2026. Demand exceeded those expectations almost immediately. Startups. Enterprises. AI labs. Governments. All wanted more capacity than anticipated.
Financial terms remain undisclosed. Analysts point to GPU prices that can reach tens of thousands of dollars per unit. The expanded order likely carries a value measured in tens of billions of dollars. Nvidia’s supply commitments already stretch years into the future. This deal locks in significant capacity.
One hundred thousand of the new GPUs will support dedicated AI factories for U.S. government workloads. These facilities will operate on AWS infrastructure cleared for Impact Level 6 security classifications. One of the highest standards for national security applications. The partnership with Nvidia extends to building these secure environments from the ground up.
Yet the agreement reaches beyond raw compute. Nvidia’s Vera CPUs will arrive in AWS data centers. Some will pair with Rubin GPUs. Others will run standalone. The companies plan deeper work on networking. On open models. On data processing. And on robotics platforms. Amazon’s vast warehouse operations stand to benefit directly.
This reliance on Nvidia persists even as Amazon accelerates its custom chip efforts. The company’s Trainium AI accelerators, Graviton CPUs and Nitro networking chips have reached a $25 billion annualized revenue run rate. Amazon has secured more than $225 billion in revenue commitments for Trainium. Major deals with Anthropic and OpenAI form part of that total.
Still the hyperscaler keeps writing large checks to Nvidia. Bloomberg noted that Amazon, Microsoft and Google all develop alternative chips yet continue as Nvidia’s largest customers. The reason is simple. Nvidia’s GPUs power training and inference for most leading AI models. Switching costs run high. Performance advantages remain compelling.
Amazon’s capital spending reflects the scale of its ambitions. The company now projects roughly $220 billion in capital expenditures for 2026. Much of that money will build data centers and install the very GPUs now on order. Memory costs have driven some of the increase. Power infrastructure adds another layer of expense.
Recent developments show Amazon hedging its bets. In early September the company struck a deal with Qualcomm that could involve up to $60 billion in AI data center chips and related products. The agreement focuses on inference workloads where power efficiency matters most. Reuters reported that Qualcomm received a warrant allowing Amazon to purchase about $4 billion in stock as part of the arrangement.
Qualcomm aims to carve out a bigger role in data centers as demand for inference hardware grows. Amazon’s move suggests it wants multiple suppliers. No single vendor can satisfy every need at the required scale and cost.
The Nvidia expansion also ties into Amazon’s government push. Last year the company announced plans to invest up to $50 billion in AI and supercomputing infrastructure for federal agencies. That effort includes purpose-built capacity across classified environments. The new GPUs and Nvidia partnership will accelerate those capabilities.
Industry watchers see the deal as validation of Nvidia’s continued dominance. Even with competition from custom silicon the chipmaker’s order book stays full. Blackwell and Rubin generations together target more than $1 trillion in revenue through 2027 according to some projections cited in recent coverage.
Yet questions linger about long-term economics. Hyperscalers including Amazon, Microsoft, Alphabet and Meta collectively plan around $760 billion in capital spending this year. Returns on these massive AI infrastructure bets remain uncertain. Training ever-larger models delivers diminishing performance gains. Inference costs must come down for widespread adoption.
Amazon executives argue demand already exists. AWS revenue grew 37 percent in the second quarter with operating income up 63 percent. Customers aren’t waiting for lower prices. They pay premiums for immediate access to the latest hardware.
The company’s internal GPU management offers a window into the pressure. In 2025 retail operations faced shortages that delayed projects. Amazon responded with centralized allocation pools and stricter approval processes. Those measures helped. But the experience showed how quickly compute hunger can strain even the largest organizations.
So Amazon buys more Nvidia chips. It develops its own accelerators. It partners with Qualcomm for specialized inference silicon. The strategy spreads risk while maximizing capacity.
Deliveries for the latest order stretch into 2028. That gives both companies time to refine liquid cooling designs, networking fabrics and software optimizations. Success will depend on turning those raw GPUs into efficient, profitable AI services that customers renew year after year.
For now the message is clear. Demand keeps surging. Amazon keeps ordering. And Nvidia remains the primary beneficiary.
Amazon’s $Billions GPU Bet: Tripling Nvidia Orders to Feed AI Hunger first appeared on Web and IT News.