This Wednesday, Amazon’s (AMZN) cloud business AWS and NVIDIA (NVDA) announced an expanded partnership, with AWS planning to deploy an additional 2 million NVIDIA graphics processing units (GPUs) across its global infrastructure between 2027 and 2028. This commitment roughly triples the total procurement volume agreed upon just five months ago, raising the total committed number from over 1 million to more than 3 million. Following the announcement, market reactions were mixed: NVIDIA, which reported earnings this week, saw its stock surge nearly 9% on Thursday, while Amazon’s stock edged down about 1.5%, trading at approximately $256 as of press time.
As early as March at NVIDIA’s GTC conference, AWS announced plans to add over 1 million NVIDIA GPUs starting in 2026. However, actual demand over the past several months has exceeded prior expectations. According to the latest statement on Wednesday, the additional 2 million GPUs will cover NVIDIA’s Blackwell Ultra, Rubin, and Rubin Ultra series chips. NVIDIA Founder and CEO Jensen Huang stated that the two companies have built one of the most powerful growth engines of the AI era, and current demand continues to outpace all forecasts. Beyond chip procurement, the two companies also plan to build AI factories for the U.S. government, including the deployment of 100,000 GPUs on secure AWS infrastructure. At the same time, Amazon’s own chip division, Trainium, will also collaborate with NVIDIA’s custom high-bandwidth memory technology.
In fact, Amazon’s capital expenditure had already been on an upward trajectory prior to this. The company estimated around $200 billion in capital spending for 2026 in February and April, but during the July earnings call, CEO Andy Jassy raised that figure to approximately $220 billion, citing rising memory costs. As of the second quarter of 2026, Amazon’s net capital expenditures for property and equipment purchases (net of sales and incentive rebates) over the past 12 months totaled $169 billion, an increase of $66.1 billion year-over-year, primarily directed toward AI. During the same period, operating cash flow grew 33% to $161.4 billion, but due to substantial investments, free cash flow turned from a positive inflow of $18.2 billion in the same period last year to a negative outflow of $7.6 billion. Jassy explicitly stated that even at this spending level, capacity in 2026 and 2027 will still not meet all demand, and demand for 2028 is already considerable. With the latest GPUs scheduled for delivery in 2027-2028, this high level of investment will extend well beyond 2026.
Behind the robust capital spending is real revenue support. AWS revenue accelerated from 28% year-over-year growth in the first quarter to 37% in the second quarter, reaching $42.2 billion, its fastest growth rate since 2021. More critically, its contract backlog is rapidly accumulating, with AWS’s undelivered contract backlog reaching $496 billion in the second quarter, growing at triple-digit rates year-over-year. AWS’s AI business annualized revenue run rate has also surpassed $25 billion, maintaining triple-digit year-over-year growth. Faced with such strong demand signals, Amazon’s significant incremental hardware commitment appears well-grounded.
