Moody’s AI Spending Surge Will Keep Driving Hyperscaler Race

As the AI race continues to push the data centre industry, such innovations are creating new demands for hyperscalers.

These companies are now experiencing the challenge of data centre capacity – as breakthrough AI models like the latest large language models (LLMs) require vastly more computing resources than previous generations. 

Technology giant Nvidia has estimated that responses from reasoning models require more than 100 times the computing resources of prior generations. 

With this in mind, Moody’s latest report assesses the race to expand data centre capacity and finds that “growth in reasoning models creates a challenge for AI service providers to balance high volumes of throughput with rapid response times for user queries, substantially raising AI data centre capacity demand”.

Additionally, open-source AI models are broadening access and encouraging new startup companies, which is ultimately driving capacity demand even further. 

Moody’s notes that “all major AI labs serving popular models are short of capacity as demand for inferencing tokens has exploded, requiring them to cap the usage”.

Tracking the hyperscaler spending spree

Capital spending by the top five US hyperscalers – Amazon Web Services (AWS), Microsoft, Alphabet, Meta and Oracle – rose 66% to US$211bn in 2024, according to Moody’s, with their growth mainly driven by AI infrastructure investments. 

Moody’s says these figures exclude finance leases, meaning the full scale of investment is larger.

It notes that Microsoft’s combined future operating and finance lease commitments, primarily for data centres that had not yet commenced, increased to US$105bn at year-end 2024, from US$26bn two years ago. The company’s annualised run-rate revenue related to AI increased to over US$14bn in the quarter ended March 2025.

Likewise, Amazon estimates that its AI business has a multibillion-dollar annual revenue run rate, growing at triple-digit year-over-year percentages. The company operates through its AWS division, though Amazon does not report separate capital expenditure figures for AWS.

AWS reported a revenue of US$108bn for 2024, representing growth of 19% year-over-year, as the cloud provider continues expansion of its AI capabilities.

According to Moody’s, US technology companies represented 44% of total global installed data centre capacity in 2023, before the AI buildout exploded. The research states that “capital spending understates the full-scale of investments, however” as companies increasingly resort to leasing capacity with options to renew, sublease, take on more capacity or exit commitments.

The rapid growth in AI has spawned “neo cloud” startup companies specialising in AI services, such as Coreweave, Crusoe and Lambda, the report says. As a notable data centre partner, Coreweave’s capital spending increased nearly three-fold in 2024 to US$8.7bn and the company had more than 1.3 gigawatts of contracted power relative to 360 megawatts of active power at year-end.

The sovereign AI boom

Numerous nations have committed substantial investments to develop national AI infrastructure based on native datasets for training, languages, cultures and practices. 

The report notes that The People’s Republic of China has allocated US$138bn for emerging technologies including AI and semiconductors, while the European Union committed €200 billion (US$235.7bn) for InvestAI initiatives, including €20 billion (US$23.57bn) specifically for AI data centres.

Moody’s highlights countries like Canada, South Korea, India and Japan as key nations that have invested heavily into AI infrastructure and sovereign AI platforms. 

While some sovereign AI projects involve partnerships with US hyperscalers, Moody’s expects “the majority of the spending to benefit domestic data centre developers and cloud services providers”. 

These investments encourage development of domestic foundational AI models and ecosystems to reduce reliance on US-based model developers.

Examples include HyperCLOVA X, Korea’s first major home-grown generative AI (Gen AI) model and Colosseum, an AI supercomputer in Italy. 

Microsoft has also announced commitments to invest more than US$35bn across 14 countries within three years to expand AI and cloud data centre infrastructure. The company recently partnered with the UAE’s technology holding group G42 to develop AI infrastructure in Kenya.

Data centre risks to evolve, Moody’s says

Data centres represent long-lived assets with useful lives of 15 years or more, Moody’s says. With the unpredictable rate of AI adoption, rapid pace of innovation and intensifying competition, there is plenty of uncertainty about return levels and timeframes for investments.

The report highlights that all hyperscalers remain short of data centre capacity, a situation likely to persist through 2025. This has hindered their ability to convert customers’ contractual commitments for data centre use into revenues. 

Their reported contracted revenue backlogs have steadily increased, driven by customers' commitments for future capacity.

Moody’s anticipates that Nvidia’s accelerated computing processors, which power a substantial majority of AI workloads, will remain in short supply in 2025. Hyperscalers may shift spending plans between data centre capacity and IT equipment, but combined capital expenditure is expected to increase provided tariffs on semiconductor and IT hardware imports do not increase materially from current levels.

Additionally, the potential for higher US tariffs, retaliatory responses by trading partners and heightened macroeconomic uncertainties from trade policy actions increase risks related to AI infrastructure spending. 

If tariffs are imposed, the report says that costs of IT products imported into the US will likely increase.

Capital intensity of US hyperscalers has increased significantly due to AI investments. The report finds that capital expenditure relative to revenues has risen sharply as companies make long-term investments to acquire capacity while AI technologies evolve rapidly.

Economic models will change with increasing competition between proprietary AI models such as OpenAI’s ChatGPT and Alphabet’s Gemini and numerous open-source offerings from Meta, DeepSeek, Alibaba and Mistral.

Moody’s suggests that hyperscalers face a wide gap between data centre buildout and first revenues. Return on long-term investments will depend upon the pace of rapid technological changes, making timing and scale of future computing needs uncertain. Returns will also depend on the competitive environment several years into the future.

“Although AI-related services have added sizeable revenues for hyperscalers in a short period of time, the vast capital spending needed to build out AI infrastructure has increased risk,” the Moody’s report says. “The hyperscalers have high revenue growth rates but their capital expenditures relative to revenues have increased sharply. 

“The hyperscalers need to make long-term investments to acquire capacity while AI technologies are still evolving rapidly.”

Explore the latest edition of Data Centre Magazine and be part of the conversation at our global conference series, Tech & AI LIVE. 

Discover all our upcoming events and secure your tickets today.

Subscribe to our Weekly Newsletter for the latest insights, trends and expert analysis.

Data Centre Magazine is a BizClik brand