Microsoft is reportedly preparing one of the biggest infrastructure expansions in technology history, with plans to grow its global data-centre capacity to around 38 gigawatts by 2032.
The figure, reported by Reuters via Bloomberg on September 10, would represent more than a threefold increase from Microsoft’s current footprint. Microsoft has not publicly confirmed the 38GW target itself, so the number should still be treated as a reported roadmap rather than a formal company forecast.
Even with that caveat, the direction is unmistakable. Microsoft has already said it expects to roughly double its overall data-centre footprint within two years as demand for Azure and AI computing keeps rising.
This is no longer just a cloud expansion
A few years ago, data-centre growth was mostly discussed in terms of cloud storage, enterprise software and streaming. The AI boom has changed the scale of the conversation.
Training and running large AI models requires enormous amounts of computing power, networking, cooling and electricity. Microsoft is building for that future across Azure, Copilot and its wider AI ecosystem, while rivals including Google, Amazon, Oracle and Meta are also pouring billions into new capacity.
We are already seeing the same pressure elsewhere. Oracle’s AI cloud backlog has climbed to $664 billion, while NVIDIA and its partners are targeting up to 2GW of additional AI infrastructure in Australia.
Put beside a reported 38GW Microsoft target, those numbers show how quickly AI is turning computing infrastructure into a strategic race of its own.
Microsoft was already signalling a huge buildout
The reported 2032 target is consistent with Microsoft’s recent public comments. In earnings materials, the company said it was adding capacity at an unprecedented pace and expected to increase total AI capacity by more than 80% in a year. It has also highlighted its Fairwater data-centre project in Wisconsin, designed to scale to roughly 2GW.
Microsoft has increasingly argued that raw capacity alone is not enough. In September, Azure infrastructure president Rani Borkar described a shift toward “useful yield” — getting more useful AI output from every unit of silicon, memory, networking and power.
That matters because simply building more data centres is expensive. Power supply, grid connections, cooling, chips and construction timelines are already becoming constraints across the industry.
The power question is getting harder to ignore
Thirty-eight gigawatts is an extraordinary amount of capacity. For readers, the important point is not the number alone but what comes with it: more pressure on electricity grids, more demand for advanced chips, and a much bigger infrastructure bill behind every AI assistant and cloud service.
That debate is already becoming political. Reuters reported this week that US lawmakers are considering measures aimed at stopping data-centre expansion from pushing more electricity costs onto households.
For Australia, the same conversation is likely to become increasingly relevant as new AI facilities are proposed around Sydney, Melbourne and other major infrastructure hubs.
What it means for Microsoft — and everyone else
If Microsoft reaches anything close to 38GW by 2032, the company will not simply be expanding Azure. It will be building the physical foundation needed to compete for the next generation of AI workloads.
That puts even more pressure on Amazon, Google, Oracle and emerging AI-cloud providers to keep spending. It also strengthens demand across the rest of the technology supply chain — from NVIDIA GPUs and networking hardware to memory, power equipment and cooling systems.
The AI race is increasingly being decided not only by who has the best model, but by who can actually power and run those models at enormous scale.
And at 38GW, “enormous” may be an understatement.
Sources: Reuters reporting on Microsoft’s reported 2032 capacity plan; Microsoft investor statements and the Official Microsoft Blog.
