Moody’s: AI’s Data Centre Impact, Challenges & Innovation
The intensified demand for data processing capabilities poses substantial challenges for hyper-scale companies.
The latest report from Moody’s highlights that breakthrough AI models, including advanced large language models, necessitate vastly greater computational resources compared to their forerunners.
One prominent industry player, Nvidia, has quantified this burgeoning demand, estimating that reasoning models require over 100 times the computational power compared to earlier iterations.
- AI demand surges, with data centre energy use set to double by 2028 (AI data centres will make up 20% of this)
- Hyperscalers face increasing competition from cost-efficient rivals like DeepSeek and open-source AI models
- Misjudging AI demand – overbuilding or underbuilding – could harm industries
Moody’s analysis shows how the evolution in reasoning models compels AI service providers to meet high data throughput needs swiftly, which elevates the demand for data centre capacity significantly.
Furthermore, the proliferation of open-source AI models is fuelling further capacity demands as they invite new startups into the fold.
“All major AI labs serving popular models are short of capacity as demand for inferencing tokens has exploded, requiring them to cap the usage,” Moody’s observes.
These insights point to a clear tension between data innovation and infrastructure capabilities in the AI sector.
Hyper-scaled investments in AI infrastructure
With the burgeoning AI field, leading US hyperscalers like Amazon Web Services (AWS), Microsoft, Alphabet, Meta and Oracle have markedly increased capital investments in AI infrastructure.
According to Moody’s data, these leaders boosted their capital expenditure by 66%, reaching US$211bn in 2024, primarily directed at bolstering AI infrastructures.
The report highlights that Microsoft’s data centre commitments, including finance leases not yet commenced, swelled to US$105bn by the end of 2024 from US$26bn two years prior.
Concurrently, Amazon projects its AI revenue is accruing at a triple-digit annual rate through AWS, despite not disclosing specific capital expenditures for the division.
The report further unveils that US tech giants account for 44% of global installed data centre capacity pre-AI explosion, a figure that fails to capture the entirety of investments due to increasing lease practices.
This significant capital outlay pivots around the need to accommodate the 'neo cloud' startups like Coreweave, Crusoe and Lambda, which specialise in AI services and are expanding rapidly.
Worldwide sovereign AI engagement
Nations are also heavily investing in AI infrastructure, as highlighted by Moody’s report.
Countries such as China and members of the EU are channelling funds into developing AI capacities aligned with local data, languages and practices.
China’s allocation reaches US$138bn for emerging technologies, while the EU has earmarked €200bn (US$220bn) for InvestAI, including substantial investments specifically into AI data centres.
Further investments from Canada, South Korea, India and Japan target reducing dependency on US-based AI model developers by nurturing local foundational AI models and ecosystems.
Key examples include Korea’s HyperCLOVA X and Italy’s Colosseum supercomputer.
Additionally, Microsoft’s pledge to invest over US$35bn across various countries underscores public-private partnerships shaping this sector.
“AI revenues are scaling rapidly, but risks are increasing with long-term investment in AI data centres and uncertain returns.”
Evolving risk dynamics in data centres
Moody’s outlines the inherent risks tied to these hefty investments as data centres, with life spans stretching over 15 years, adapt to the volatile pace of AI innovation and adoption.
The scarcity in data centre capacity hampers hyperscalers’ abilities to convert user commitments into actual revenue and sustains high revenue backlogs driven by customer demand for capacity.
Nvidia's accelerated processors, crucial for AI workloads, will continue in short supply into 2025, lengthening the time for data centre buildouts to translate into revenues.
Coupled with the possibility of increased tariffs on IT imports, these factors pose uncertainty over costs and ROI for hyperscalers facing rapidly evolving AI technologies and heightened global competition.
The Moody's findings conclude with an acknowledgement of the considerable revenue potential AI services possess for hyperscalers.
Yet, they emphasise the heightened financial exposures due to the capital-intensive demands of AI infrastructure setup, as companies strive for strategic, long-term positions within this dynamically advancing realm.
