The AI wave has generated robust demand for cloud computing power, driving Amazon’s (AMZN) cloud business growth expectations steadily higher, with institutions raising their forecasts for performance and capital expenditures. However, shortcomings in supporting infrastructure such as power infrastructure and construction labor are constraining hardware deployment, and the large volume of procured AI servers cannot be brought online in a timely manner. The supply-demand imbalance in computing power is difficult to reverse in the short term, and hefty capital investment continues to weigh on corporate cash flow levels.
Amazon Chief Executive Officer Andy Jassy revealed in a video podcast program that, affected by the rapid growth in demand for artificial intelligence infrastructure, AWS cloud computing large-capacity capacity has essentially been sold out through 2027, and a substantial amount of capacity for 2028 has already been pre-booked by customers. Jassy stated that computing power demand from AI laboratories and enterprise customers remains persistently strong, and enterprises are doing their utmost to catch up with the pace of demand growth. The company previously planned to invest approximately $220 billion in capital expenditures in 2026, and Jassy believes that, against the backdrop of cloud capacity already being sold out in 2026, continued additional investment is reasonable. AWS will reserve a small amount of capacity for on-demand business, the Bedrock inference service, and the EC2 service, among which EC2 can provide cloud servers for applications and Bedrock supports customers in calling various AI models to build generative AI applications.
Amazon will release its third-quarter financial report on October 29, and Deutsche Bank published a preview report on October 20, expressing optimism about strengthening AWS demand, stable advertising business, and recovering e-commerce business, and expecting the company’s revenue to approach the upper end of its guidance, while maintaining a Buy rating and a $325 target price. The bank regards AWS as the core source of earnings upgrades and simultaneously raised its capital expenditure forecast, expecting Amazon’s capital expenditure in 2027 to be approximately $330 billion, higher than the market consensus. Calculation data shows that Amazon’s third-quarter revenue is expected to reach $202.8 billion, with operating profit and earnings per share both higher than market expectations; AWS third-quarter revenue growth forecast has been raised to 41%, and fourth-quarter growth is expected to further climb to 44%, with profit margins improving simultaneously. Based on information from exchanges with the co-founder of cloud cost management company Archera, customer committed spending and actual computing power consumption are accelerating in tandem, providing support for cloud business growth.
The core bottleneck in computing power supply is not chip procurement, but power infrastructure. According to internal documents and meeting recordings, Amazon is storing Nvidia GB300 racks in transit warehouses; data center machine room space is sufficient, but power supply capacity is insufficient to support bringing the equipment online. Internal emails show that in August Amazon planned to receive approximately $5 billion worth of GB300 racks, and would continue to receive approximately $3 billion worth of equipment each month thereafter, a pace that will not change in the short term. AWS’s chief financial officer mentioned at an internal meeting that heavy investment has already turned free cash flow negative, and the relevant responsible person also admitted that free cash flow may remain negative for several years to come. Amazon’s trailing twelve-month free cash flow had already turned negative to $7.6 billion in the second quarter. The progress of power infrastructure construction is lagging, and the problem of labor shortage is equally prominent; the company is building worker accommodation areas in West Texas to ensure construction progress. The team previously promised to alleviate capacity tightness in 2027, but now expects demand in that year to still exceed available capacity. At the current stage, AWS is releasing inference capacity in batches, while guiding GPU customers to sign longer-cycle procurement agreements and promoting self-developed hardware to ease the pressure of computing power shortages.
