Amazon (AMZN.US) CEO: AWS cloud computing capacity is fully booked until 2027, and a large amount of capacity has already been committed by customers for 2028

According to Zhitong Finance APP, Amazon (AMZN.US) CEO Andy Jassy stated on Friday that, with the continuously rising demand for artificial intelligence (AI) infrastructure, Amazon AWS's cloud computing capacity is essentially sold out through 2027, and a significant amount for 2028 has already been reserved by customers. In a video podcast, Jassy said Amazon is striving to keep up with the growth in demand from AI labs and enterprise clients, whose appetite seems insatiable.

Earlier this year, Amazon revealed its plan to invest approximately $220 billion in capital expenditure by 2026. Jassy remarked that, given the 2026 cloud computing capacity is already sold out, the company “could easily invest even more, and it would not be an unwise use of capital.”

He said: “Capacity for 2027 is basically already sold out. We always reserve a certain amount of capacity to meet on-demand needs and for our inference services like Bedrock or EC2, but all bulk capacity is essentially sold out. A considerable portion of 2028’s capacity has also already been sold.” EC2 is an AWS service providing cloud servers for running applications; Bedrock enables customers to access various AI models to build generative AI applications.

Deutsche Bank Preview of Amazon Q3 Results: Raises AWS Growth Outlook, 2024 Expenditure May Rise to $330 Billion

Amazon plans to announce its third quarter results on October 29th. On October 20th, Deutsche Bank released a preview of Amazon’s Q3 earnings, projecting that strong AWS cloud business demand, robust advertising business, and a rebound in e-commerce metrics after promotional period overlaps will support a performance close to the upper end of Amazon’s revenue guidance. The bank maintains its “Buy” rating and a $325 price target.

Deutsche Bank considers AWS a key driver for future upward guidance revisions. However, the improved growth outlook comes with increased investments: the bank raised its 2027 capital expenditure forecast for Amazon to around $330 billion, above the market consensus of roughly $290 billion.

Based on its survey of AWS, advertising, and retail business, Deutsche Bank estimates Amazon’s total Q3 revenue at around $202.8 billion, near the company’s upper guidance limit; operating profit at about $26.9 billion, 1.4% above guidance and 3.5% above market consensus; and earnings per share at about $2.05, approximately 4.7% higher than consensus.

Deutsche Bank raised its forecast for AWS’ year-over-year Q3 revenue growth from 40% to 41%, up from 37% in Q2. This acceleration comes despite a higher year-ago comparison base by about 2.8 percentage points than in Q2. The bank forecasts that Q4 growth will rise further to 44%.

According to these forecasts, Q3 revenue would be about $46.5 billion and Q4 about $51.1 billion, about 2% and 4% above market consensus, respectively. However, Deutsche Bank also warns that buy-side expectations for AWS Q3 growth are already around 42%, higher than its own 41% forecast.

In terms of profit margins, Deutsche Bank estimates AWS’ Q3 operating margin at 37.1%, versus 34.6% a year ago; with Q4 at 38.1%, versus last year’s 35%. Excluding a one-off $600 million gain in Q2, margins were about 37.9%, compared to 32.9% the previous year.

The improved demand is evident in both committed spending and actual consumption. According to conversations between Deutsche Bank and Aran Khanna, co-founder and CEO of cloud cost management company Archera, from June to September, the growth rate of customer committed spending for the next 12 months rose from 35% to nearly 38%, while actual usage growth climbed from 38% to 40%. The former reflects future demand commitments, while the latter is closer to current business consumption; both accelerating in tandem, supporting revenue growth.

Compute Capacity Remains a Bottleneck! Power Shortages Leave Billions Worth of Chips Gathering Dust in Warehouses

The bottleneck remains compute capacity. According to Amazon internal documents and a staff meeting recording recently obtained by Business Insider, AWS is temporarily storing Nvidia GB300 racks in “intermediate warehouses” because, although there is room for more racks at the target data centers, there is not enough available power to bring them online.

Internal emails indicate that in August alone, Amazon expected to receive about $5 billion worth of GB300 racks, and will continue to receive about $3 billion each month, keeping this pace “for the foreseeable future.”

At an internal staff meeting in September, AWS CFO John Felton jokingly told Prasad Kalyanaraman, the executive in charge of AI infrastructure, “You’ve made our company’s free cash flow go negative, congratulations.” Prasad Kalyanaraman did not dispute this and stated that the company’s free cash flow is expected to remain negative for the “next several years.” Notably, when Amazon released its Q2 earnings this July, its rolling free cash flow had already turned significantly negative, to minus $7.6 billion—the first time in many years.

The impact of this situation on the market is already apparent. Chips sitting idle in warehouses generate no computing power, leading to persistent shortages of online compute capacity. GPU leasing prices remain high, resulting in a classic inflationary bottleneck—large amounts of capital, labor, and energy are consumed in advance, while output is delayed for months or even years.

The root cause of the chip backlog lies in severely lagging power infrastructure. Prasad Kalyanaraman revealed in the internal meeting that, due to shortages of pipefitters and electricians in West Texas, Amazon has had to fund construction of worker accommodations so crews can live on-site for two to three years until project completion.

Prasad Kalyanaraman also disclosed that in 2023, he promised Andy Jassy that AWS would overcome capacity constraints by 2027. Now, Amazon expects demand in 2027 will still outstrip available capacity.

Channel feedback shows AWS is gradually rolling out inference capacity to address customer backlog, but, like its peers, still faces significant capacity constraints. In this resource-constrained environment, AWS is reportedly pushing GPU-dependent customers to sign broader and longer-term procurement agreements, and to use its in-house developed compute and networking hardware where applicable.