Trending Articles
Trending Topics
Trending
Last year, Microsoft said it would invest $80 billion in data centres to power AI, significantly more than previous years. It was even more than Alphabet’s $75 billion and Meta’s promised $65 billion.
Over the past few months, reports have trickled out that Microsoft is easing off its AI investments, pausing or slowing data centre projects, including a $1 billion project in Ohio.
What’s going on? Noelle Walsh, President of Microsoft’s Cloud Operations, nodded to the reports in a post in LinkedIn, but didn’t clarify so much as add to the confusion.
She wrote that MIcrosoft has doubled its data centre capacity over the past three years, adding more in 2024 than ever before, with 2025 set to be another record year.
“This year, Microsoft is on track to spend more than $80B to continue building out our datacenter infrastructure and these investments are informed by near-term and long-term demand signals,” she wrote.
But a few paragraphs down, Walsh admitted that plans had changed – with some projects paused or progress slowed.
“By nature, any significant new endeavor at this size and scale requires agility and refinement as we learn and grow with our customers,” she wrote.
“What this means is that we are slowing or pausing some early-stage projects. While we may strategically pace our plans, we will continue to grow strongly and allocate investments that stay aligned with business priorities and customer demand.”
In short: Microsoft invested in AI because it saw a big opportunity, it pledged $80 big ones this year, and now it’s just not really sure exactly how or where that money is best spent.
One analyst said the reason was simple: the withdrawal from new capacity leasing was down to a decision not to support additional training workloads from OpenAI, according to Reuters.
Now, Microsoft’s still going to spend all that cash chasing AI, it just isn’t sure how yet. Indeed, another analyst suggested the pullback was down to a chillier relationship with OpenAI.
That follows Microsoft in January publishing a blog post to explain that OpenAI was no longer entirely exclusive to Microsoft’s Azure, though Microsoft retained a right of first refusal on hosting OpenAI’s workloads. So OpenAI can seek relationships beyond Microsoft, if it says so or if Microsoft lacks the capacity.
That follows rumours that OpenAI was complaining that lack of compute was holding back products, as well as an exception made last year to allow OpenAI to use Oracle’s infrastructure. Indeed, Microsoft’s odd little exclusivity blog post landed on the same day OpenAI announced it was part of the $500 billion over four years infrastructure buildout dubbed the Stargate Project, which will see Oracle and Nvidia team up to build more data centres for OpenAI.
By cooling its relationship with OpenAI, Microsoft might be saving itself a bundle while others rush out their AI infrastructure, driving up costs, amid an already expensive time to build tech thanks to Trump’s trade war.
The shift doesn’t mean that Microsoft is necessarily pulling back from AI. Indeed, Microsoft remains one of OpenAI’s major backers and has crammed Copilot into every nook and cranny of its software. Plus, as Walsh noted, that $80 billion is still on the table.
Instead – and this may be mere optimistic supposition on my part – Microsoft may have come to some realisations about AI.
First, that there’s no immediate ROI looming, no matter how much you charge customers for Copilot features they’re not all that keen on.
Second, that AI is overhyped and we’ve hit – as Gartner calls it – the Peak of Inflated Expectations. As a result, Microsoft is wisely battening down the hatches to ride out the Trough of Disillusionment until it can ride the Slope of Enlightenment to the Plateau of Productivity.
There are other explanations for Microsoft delaying its data centre investments.
Maybe it’s taking a sensible pause to rethink the best places to build data centres. Perhaps it’s working out where they’ll have the least environmental impact while still being able to access the immense energy required to power these systems. A recent International Energy Agency study suggests building data centres near other data centres may lead to bottlenecks.
Perhaps Microsoft has figured out ahead of the rest of the crowd that the AI revolution is going to take decades, so there’s no need to rush. It makes better sense to think through such investment and get it right rather than rush through data centres that might suddenly be out of date, the wrong technology or unneeded. Consider how the whole DeepSeek disruption has suggested might be the case.
Microsoft keeps insisting it will spend that $80 billion, but it has plenty of reasons to do so carefully – with or without OpenAI. We’ll have to keep close track of its figures to see if every billion promised on AI infrastructure does get spent this year.
