Artificial Intelligence & Machine Learning
,
Governance & Risk Management
,
IT Risk Management
Power Capacity Is Growing But 26% of Projects Faced Delays, Experts Warn
For years, enterprise cloud computing has felt like a “sky’s the limit” endeavor, but that aspiration is coming crashing to the ground as the industry faces a hard new reality.
See Also: AI or Data Governance? Gartner Says You Need Both
Growth is being constrained by the very real physical limitations of the world’s power grids. This crunch has implications for CIOs struggling to achieve plans for artificial intelligence, cloud migration and digital transformation.
The data center industry has outlined an ambitious expansion road map, but the numbers tell a different story. Last year was a record year for data center capacity growth, with nearly 6 gigawatts of capacity deployed, but 26% of projects were delayed, according to a new report from Sightline Climate. This year’s outlook paints a similar picture. While 16 gigawatts of capacity is slated to come online, so far only 5 gigawatts is actually under construction. Industry projections anticipate that as many as 30% to 50% of those projects will face delays caused by power availability, permitting issues and construction constraints. The electricity needed for a 1 gigawatt data center could power 700,000 to 900,000 homes.
The shortage is the result of a perfect storm of government regulation, construction realities and hyperscalers vying for more reliable capacity to fuel their AI projects on their own terms.
The ‘Bring Your Own Power’ Mandate
Grid bottlenecks have driven heavyweight players including Oracle and Google to “give up on the grid” for AI training capacity. Instead, the companies are developing power strategies that bypass the current system to ensure reliable capacity, timelines and emissions, the report said.
Oracle is creating on-site and hybrid capacity through its involvement in the Stargate infrastructure project, and creating a 1 megawatt SMR-powered project in Nashville. Google recently spent $4.75 billion on clean energy company Intersect, giving it access to 10.8 gigawatts of solar, battery storage and gas power. Microsoft, meanwhile, is creating a smaller data center network using grid power, including nuclear power.
Companies are also rolling out Battery Energy Storage Systems, previously used for backup power, to expand capacity.
As the competition for access to power resources heats up, regulators are also putting the brakes on data center expansion plans. Proposed FERC regulations could push 100% of power network upgrade costs directly onto data center projects, a potential blow to the profit margins of data center builders.
States are also vying for data center dollars. Texas is the current leader in capacity slated to come online, at 22 gigawatts, because of fast interconnection timelines, abundant power generation and a developer-friendly market, but six projects are already delayed. Extreme weather and price volatility are potential red flags for hyperscalers, which will likely look to diversify to mitigate those risks.
Virginia and Pennsylvania are both data center hot spots, but they’re growing congested, so states in the Southwest Power Pool – the transmission operator serving the Dakotas, Nebraska, Kansas, Oklahoma, and parts of Texas and Arkansas – passed a plan to connect to its grid faster and attract data center development.
And regulators aren’t just putting a finger on the scales in the United States. In Ireland, new data centers will be required to source 80% of their power from new renewable energy sources and maintain 100% backup capacity.
Strategic Planning Implications for CIOs
For CIOs, this data center squeeze requires a shift in strategic planning, and energy strategy is now a required part of the digital transformation road map. AI projects require access to resources, and those resources face growing constraints. Technology teams need to evaluate your cloud and large language model providers’ power strategies, and ask if they’re connected to the grid, on-site or use a hybrid model.
Regulations could also drive up price tags, so look for costs passed along by data center developers who need to pay for grid upgrades and renewables sourcing. Consider contracting with data centers in specific regions, like the SPP belt, which is actively pitching faster connection timelines to woo large loads.
In the era of the power-constrained cloud, CIOs who know their power destiny may be the only ones who can guarantee their AI road map.