Amazon Soars Despite Raised CapEx Guidance: What Does This Earnings Report Signify?

Amazon’s latest earnings call sent a strong signal: demand for data centers from hyperscale cloud providers is far from peaking, benefiting leading stocks in the data center infrastructure sector.

On July 30, after the US market close, Amazon CEO Andy Jassy announced during the Q2 earnings call that the cash capital expenditure guidance for 2026 was raised by $20 billion to $220 billion, representing a year-over-year increase of 71.4%.

More notably, Jassy explicitly stated that even with such substantial investment, capacity this year will still be insufficient to meet all demand, and he expects this supply-demand gap to persist through 2027.

According to Zhuifeng Trading Desk, Barclays promptly released a research report stating that the above remarks effectively refute previous market concerns about the overconstruction of AI infrastructure, and that demand for hyperscale infrastructure may continue to grow even after the current AI training cycle matures.

However, Barclays pointed out that changes in the macroeconomic environment and waning AI enthusiasm are the primary risks facing the valuations of these stocks. Fluctuations in the US dollar trend, interest rate levels, and energy costs could all impact the earnings and valuation performance of related companies.

Supply-Demand Gap Persists; Capacity Constraints Extend Through 2026–2027

Wall Street News mentioned that Andy Jassy expects the company to further increase its investment in artificial intelligence, with capital expenditures projected to reach $220 billion this year. In February, Amazon had stated that capital expenditures for the year would reach $200 billion, and maintained this forecast unchanged in April.

Andy Jassy indicated that rising memory prices have pushed up Amazon’s capital expenditure expectations. He believes that Amazon’s spending spree is unlikely to slow down in the short term. Andy Jassy said:

Even so, our capacity this year will still not be able to meet all demand. I believe this situation will also exist in 2027.

According to Barclays’ analysis, this judgment confirms that demand from hyperscale cloud providers remains exceptionally strong, and the entire industry is still operating in a capacity-constrained environment. This supports sustained demand for data center development, powered land, and wholesale capacity in 2027 and beyond.

Regarding the pace of expansion, Amazon stated it is proceeding with capacity construction as planned. The target disclosed in previous quarters was to double capacity and power capacity by the end of 2027 compared to 2025 levels, and current progress is in line with expectations.

Demand Visibility Extends Unusually Far into 2028; Enterprise Inference Applications Remain in Early Stages

Amazon’s assessment of future demand extends beyond 2027. Jassy pointed out that demand signals from 2028 are already "striking," emphasizing that enterprise customers are still in a very early stage of large-scale AI inference usage in their existing production applications.

According to Barclays’ report interpretation, combined with the judgment that enterprise inference adoption is still in its early stages, these comments suggest that demand for hyperscale infrastructure may continue to grow even after the current AI training cycle matures.

This structural long-tail characteristic of demand provides data center operators with longer-term revenue visibility.

In terms of capital allocation logic, Jassy characterized data centers as long-duration infrastructure assets that transcend single technology cycles. He stated during the call:

Capital expenditure for data centers begins two years before servers are installed and begin generating revenue. Once a data center opens and connects servers, we immediately start generating substantial revenue and can continue to monetize it over a cycle of more than 30 years without reinvesting the initial startup capital.

He further elaborated on the economics of multi-generation servers:

For data centers with a lifespan exceeding 30 years, we can capture the economic benefits of at least five to six generations of servers. The economics of subsequent generations after the first are even better because there is no need to repeat the upfront data center construction investment.

Barclays believes this framework helps explain why hyperscale cloud providers are willing to bear large-scale upfront capital expenditures. Once built, data centers will continue to create returns across multiple server generations, with the marginal benefit of the initial investment increasing over time.