On Thursday, Amazon announced second-quarter earnings that surpassed expectations, earning significant praise from investors. Net sales increased by 20%, with cloud revenue standing out as a particularly strong area. This blend of favorable results was sufficient to drive Amazon’s stock up nearly 10% in after-hours trading.

Importantly, Amazon is not reducing its data center investment, contradicting the common belief that investors prefer companies to cut back.

A specific line item highlights Amazon’s commitment to infrastructure investment. For the fiscal year ending June 30, Amazon allocated $173 billion to property and equipment — which includes GPUs, natural gas turbines, and land acquisitions — up from $107.65 billion the previous year.

Additionally, it updated its 2026 capital expenditure forecast from $200 billion to $220 billion, even as it has started tapping into its cash reserves to help finance these costs. The company closed the quarter with $7.6 billion less cash compared to 12 months prior, marking its first instance of negative free cash flow this year.

Typically, rising costs would be a difficult situation for investors to accept. However, Amazon possesses a revenue engine that provides justification for this spending. AWS revenue increased by 37% year on year, reaching $42 billion for the quarter. While this may not be enough to directly offset capex spending in simple arithmetic, it indicates that demand is increasing in tandem with supply. Given the lengthy time lag between initiating a data center project and making its capacity available, this is a positive sign for investors.

Importantly, Amazon’s AI initiatives extend beyond just constructing large data centers. The company is also making significant long-term investments in chips such as the Trainium TPU and the Arm-based Graviton processor. These projects, while not reflected in capex figures, have the potential to significantly enhance margins for its cloud business.

“We expect the AI sector to follow a similar margin trajectory to what we experienced in the core business previously,” stated Jassy during the Q2 earnings call. “AWS and Amazon Bedrock can operate a tremendously successful business without having their own frontier model, primarily because there won’t be a singular model that dominates.”

This situation is not exclusive to Amazon. We observed analogous trends at Microsoft and Google, whose stocks also rose following the announcement of robust cloud revenue. Conversely, companies like Meta that invest heavily in capex without a clear revenue stream are still facing considerable skepticism from investors. Meta’s stock dropped by 8% after disclosing quarterly earnings this week, as investors concentrated on its cash flow issues and ongoing expenditures.

Of course, investors favor revenue while disdaining expenses — that’s the nature of markets. However, it’s crucial to understand the broader implications of the AI economy. Currently, investors view cloud service providers as the most dependable aspect of the AI ecosystem, while maintaining doubts regarding the economic viability of AI labs and startups.

But Amazon’s hosting revenue represents someone else’s AI expenses. In the case of Anthropic, it’s literally the same funds.

If this spending proves unsustainable for the large labs and their clients, the revenue stream will not remain stable for Amazon and other cloud service providers. There is real competition and differentiation at all levels of the stack, but if demand for AI wanes, it will adversely impact all parties involved.

Ultimately, it all ties back to David Cahn’s $3 trillion inquiry. There is either sufficient demand to justify this expansion or there is not. While cloud-hosting services like AWS might be somewhat distanced from this demand issue, they are not insulated from its effects.