How Microsoft is Tackling an AI Power Problem

Microsoft has disclosed that thousands of AI chips are sitting unused in inventory because the company cannot build data centres fast enough to deploy them.

The constraint proves a shift in the semiconductor industry, which has spent recent years managing chip shortages. 

Now, access to electrical power and the speed of data centre construction have emerged as the limiting factors for AI infrastructure expansion.

Satya Nadella, CEO of Microsoft, says the company’s deployment challenges stem from power availability rather than chip supply.

Speaking on the Bg2 Pod, Satya says: “The biggest issue we are now having is not a compute glut, but it’s power – it’s sort of the ability to get the builds done fast enough close to power.

“So, if you can’t do that, you may actually have a bunch of chips sitting in inventory that I can’t plug in. 

“In fact, that is my problem today. It’s not a supply issue of chips; it’s actually the fact that I don’t have warm shells to plug into.”

Warm shells refer to data centre buildings that have basic infrastructure in place but lack the electrical connections and cooling systems required to operate high-powered AI processors, specifically graphics processing units (GPUs) designed for machine learning (ML) workloads.

The admission recognises the lack of power marks a dramatic U-turn for the industry which has previously managed semiconductor shortages.

How Microsoft’s infrastructure expansion strategy shifts

In September 2025, Microsoft announced plans to invest US$80bn in AI data centres for fiscal year 2025. 

The company’s AI platform transformation is supported by its growing network of over 400 facilities in 70 regions.

However, despite ongoing investments, Microsoft experienced a 1.5GW self-build slowdown earlier in the year. 

A GW represents enough electrical capacity to power approximately 750,000 homes. 

The company shifted strategy from constructing its own facilities to spending US$11.1bn in Q1 2026 leasing ready-available capacity to grow flexibility.

Today, Microsoft’s deployment schedule is being impacted by factors outside its direct control, such as power procurement timelines, grid connection queues and construction delays. 

These factors depend on utility companies, regulatory approvals and local infrastructure capacity.

Microsoft also announced a US$30bn investment in AI infrastructure and operations across the UK between 2025 and 2028 in September 2025. 

The investment formed one of the biggest financial commitments the company has made in the UK and includes US$15bn in capital expenditure to develop the country’s cloud and AI infrastructure.

As a result, the milestone investment will allow Microsoft, in partnership with Nscale, a data centre operator, to build the UK’s largest supercomputer, utilising more than 23,000 Nvidia GPUs. 

Keir Starmer, Prime Minister of the UK, said at the time: “Microsoft’s landmark investment is a powerful vote of confidence in the UK’s leadership in AI and cutting-edge technology.

“This commitment will not only strengthen our digital infrastructure and support thousands of highly skilled jobs, but also ensure Britain remains at the forefront of global innovation as we deliver on our Plan for Change. 

“We are proud to partner with world-leading companies like Microsoft to build a future powered by British ingenuity and ambition.”

Why electric utility delays create deployment bottlenecks

Research from Bain & Company, a management consulting firm, highlights that electric utility connection delays of up to five years are causing the most significant obstacle for data centre growth, signifying an explanation behind Microsoft’s deployment schedules.

The research goes on to claim that by 2030, the world will require twice as much data centre power as is required in 2025. 

It forecasts that the world will see a demand increase of 163GW, driven mostly by AI.

Bain & Company also states that the US data centre electricity demand could double to 409TWh.

Meanwhile, TD Cowen, a financial services firm, published a report last month stating that US data centre capacity leased by hyperscalers in Q3 2025 had reached the total capacity leased in 2024. 

Microsoft was noted to have had a significant increase in activity.

So, with data centres demanding more energy than companies like Microsoft can generate, where will the energy come from?

Bob Johnson, VP Analyst at Gartner, a research and advisory firm, says in response to a report suggesting power shortages will restrict 40% of AI data centres by 2027: “Significant power users are working with major producers to secure long-term guaranteed sources of power independent of other grid demands.

“In the meantime, the cost of power to operate data centres will increase significantly as operators use economic leverage to secure needed power. These costs will be passed on to AI/Gen AI product and service providers as well.”