AI spending spree continues as Microsoft commits $80B for 2025

The AI datacenter spending splurge looks likely to continue in 2025, with Microsoft alone saying it will invest $80 billion this year on building out infrastructure to train and deploy AI models.

In a blog post that reads almost like a call to arms, Microsoft president and vice chair Brad Smith described AI as a "golden opportunity" for American economic competitiveness, claiming this is set to become a world-changing general-purpose technology.

"AI offers not only new tools for people's work but also new ways to help people learn almost anything. We have the opportunity as a country to equip all Americans with the skills needed to use AI to pursue higher-paying jobs and more successful careers. This should be our national north star," he states in the post.

Yet progress requires new alliances and large-scale infrastructure investments, Smith said, confirming Microsoft is intending to invest approximately $80 billion in AI-enabled datacenters around the world. More than half of this will be invested in the US, naturally.

And Microsoft is not alone in this respect. Last year, Amazon reported that it was on track to spend $75 billion in capital expenditure during 2024, and even more in 2025. "The majority of it is for AWS and, specifically, the increased bumps here are really driven by generative AI," CEO Andy Jassy said in November.

According to figures from Taiwan-based market watcher TrendForce, AI servers are projected to make up more than 70 percent of the total value of the server industry in 2025, adding up to about $298 billion.

Shipments of servers targeting AI increased 46 percent during 2024, it claims, while for 2025 TrendForce forecasts that such hardware will see year-on-year growth of nearly 28 percent, while the proportion of these actual server shipments will increase to more than 15 percent of the total server market.

This trend has also been followed for some time by analysts at Omdia, which in its most recent report late last year estimated that datacenter capex was on track for upwards of 40 percent growth for the whole of 2024, with servers representing nearly half of that, and expected to hit $225-230 billion in value.

Looking ahead to this year, Omdia forecasts that AI systems will make up 20 percent of overall server shipments, but they will account for 73 percent of server capex spending, roughly aligning with TrendForce estimates. This is because servers procured for AI tend to be highly configured and therefore much more costly than those intended for other workloads, especially as Nvidia GPU accelerators can cost tens of thousands of dollars apiece.

Omdia put a figure on this as well; it estimates that Nvidia's kit represented a whopping 43 percent of server capex during 2024, although it says the GPU giant will struggle to grow this share much further in 2025.

Among other predictions, the analyst biz says that specialized AI cloud businesses will continue to spend in order to grow their share of the market, picking out CoreWeave in particular, which it reckons could outspend Oracle this year.

Growth can't last forever

How long this continues will depend on whether AI turns out to be yet another industry bubble waiting to burst. Qualms have already been raised about the sheer volume of investment dollars being swallowed by AI, with little to show for it so far.

• Sorry, but the ROI on enterprise AI is abysmal
• Microsoft accused of 'greenwashing' as AI used in fossil fuel exploration
• Microsoft accuses Google of creating a lobbying front called 'Open Cloud Coalition'
• Amazon stretches working life of its servers an extra year, for AWS and its own ops

Omdia says that clients in highly regulated industries had told it that 2024 would not be the year when they had the necessary regulatory frameworks in place to go big on AI in production. Many were stuck running AI pilots last year, but are scheduled to begin deployment of industry-specific AI models this year, it claims.

Some enterprise investment in 2024 was also limited by GPU availability, which will be significantly better in 2025.

So the AI datacenter frenzy appears likely to continue for the near future, despite the warnings about the ballooning energy requirements this is driving, and the knock-on effect on the environment. Financial services biz Morgan Stanley issued a report last year estimating that the industry is likely to emit three times more greenhouse gases between now and the end of the decade than if generative AI tech had not been developed.

Still, it is a golden opportunity, right? ®