AI data centre demand to exceed supply by 500% by 2030
Mon, 6th Jul 2026 (Yesterday)
Iron Mountain and Structure Research forecast that global data centre demand will exceed supply by more than 500% by 2030, driven by growth in artificial intelligence infrastructure.
Annual global demand for data centre capacity is expected to reach nearly 90GW by the end of the decade as spending on servers, graphics processing units and facility buildouts rises.
Hyperscaler capital expenditure is projected to hit USD $375 billion this year, up 36% from 2024. About half of that spending is expected to go on servers and GPUs, with the rest directed to data centre capacity.
The outlook points to a marked shift in the infrastructure needed for artificial intelligence workloads. After an initial phase dominated by model training, demand is expected to shift towards inference as companies deploy real-time services at greater scale.
Inference capacity is expected to overtake training capacity in 2026. By 2030, inference is projected to account for 80% of AI critical IT load, implying roughly four times more inference infrastructure than training infrastructure.
That shift has implications for where facilities are built. Inference workloads often need to be closer to end users, likely increasing demand for data centres in and around densely populated urban areas rather than only in established remote campuses.
Regional buildout
The forecast suggests data hubs with more than 2GW of capacity will emerge in every major global region. In North America, Northern Virginia is projected to reach 8.5GW by 2030, while Dallas is expected to scale to 2.8GW and Phoenix to 2.7GW.
In Europe, London is forecast to reach 2.7GW, Frankfurt 2.68GW and Paris 2GW. Growth is also expected to accelerate in Madrid, Barcelona, Berlin, Dusseldorf and Lisbon.
Across Asia-Pacific, Tokyo is projected to reach 2.8GW, Sydney 2.4GW and Johor 2.2GW. Mumbai is expected to hit 2.15GW, adding to a broader regional expansion driven by demand for AI-related computing and storage.
Cost pressure
The report also identifies falling model costs as a major factor shaping adoption. The price of the cheapest large language models has been declining tenfold each year, lowering barriers to use across organisations.
Rather than reducing overall consumption, lower pricing is expected to broaden deployment. That could in turn create new cost management issues for companies using usage-based pricing models for AI services.
Businesses may need stricter internal controls over employee use of AI tools as token-based charging becomes more common. The forecast warns that unchecked use for low-value tasks could add materially to operating costs.
The findings come as demand for power, land and grid connections has become a defining issue for the data centre industry. Developers and operators in major markets are already contending with planning constraints and electricity shortages, even before the latest wave of AI infrastructure requirements.
A supply gap on the projected scale would intensify competition for suitable sites and utility access. It would also increase pressure on operators, cloud companies and enterprise customers to secure capacity well ahead of deployment needs.
The expectation that inference will dominate future AI workloads may also reshape investment priorities across the sector. Facilities designed for low-latency service delivery near population centres could become more important, even as very large campuses continue to handle heavy training workloads.
In Europe, that could sharpen focus on established hubs such as London and Frankfurt while boosting interest in newer markets with room to expand. In Asia-Pacific, the projections suggest established centres and rising secondary markets will both play a role in absorbing AI-led demand.