Looming energy crunch makes future uncertain for datacenters

Datacenter operators are facing a paradoxical crisis – demand for their services is greater than ever before, just as access to power, environmental concerns, rising costs and skill shortages have begun to threaten necessary infrastructure expansion.

Over the past year The Reg has tracked how widespread interest in new types of large language models, notably generative AI, has driven internet juggernauts to make them more widely accessible, expanding their server farms in the process to keep pace.

Just a few months ago, data from analyst Omdia indicated that capital spending on infrastructure for cloud datacenters was on track for a greater than 30 percent year-on-year hike by the end of this year.

"We haven't seen anything as disruptive that has happened before," Omdia's head of Cloud and Datacenter Research Vlad Galabov told The Register. "We've now added about $10 billion to our previous spending forecast [for this year]."

This large-scale investment is now coming from a wider group than just the big three cloud providers – AWS, Microsoft Azure, and Google Cloud. It not only includes the top 10 service providers, but also a growing number of specialized AI cloud businesses such as CoreWeave that are getting in on the act.

On their own the hyperscalers are still forecast to account for more than 60 percent of datacenter capacity by 2029, unless something drastic happens to shake up the continuing drift of enterprise workloads being drawn into their bit barns.

The total capacity of hyperscale facilities is set to grow almost threefold over the next six years on the back of AI demand, with spending estimated to hit $143 billion by 2027 alone.

The result is a wave of investment in new datacenters and expansion of existing facilities, as the hosting providers and the real estate companies aim to stay ahead of the capacity curve.

Many of these dollars are headed for places like Northern Virginia, already the hyperscale datacenter capital of the world, while other regions with less available free space to play with are experiencing a shortage of real estate to site new bit barn campuses and buildings.

An overview from early 2024 highlighted that in Europe, demand for data dormitory space was already outstripping supply in the previous year, with hyperscalers snapping up much of the available capacity.

Rental rates for leased capacity in this region have jumped considerably over the past two years, with build costs being the primary culprit. These were affected by difficulty in securing sufficient power for new sites and appropriate land for building. Things are exacerbated by the hyperscalers using tactics such as pre-letting, where they are willing to sign up for additional capacity before construction on the site even starts.

In the UK, speculators such as private equity investor Blackstone are now ready to pounce on suitable locations if they become available. The biz scooped up a slot in northern England formerly owned by battery startup Britishvolt earlier this year for a planned £10 billion ($12.45 billion) datacenter facility.

Securing access to power is becoming the biggest problem, according to at least one commercial property developer which vowed pump more cash into building bit barns, but only if it can just secure the energy supply needed for them.

This trend was identified in many markets around the world, according to a report from global commercial real estate organization CBRE which was published in July. It found that local governments are being forced to address power constraints by simplifying permitting processes and integrating more renewable energy into the grid.

Worries are growing that energy demands from AI-driven datacenters may be expanding faster than new capacity can be added to the grid, leading to potential brownouts and blackouts.

In November, Washington, D.C.-based think tank the Jack Kemp Foundation (JKF) warned that Americans could face a 70 percent hike in their electricity bills by 2030, unless urgent action is taken to boost generation and transmission capacity to ensure the grid continues to operate in the face of demand overreaching supply.

Analyst Gartner also forecast that the energy demands from AI infrastructure could expand 160 percent over the next two years, leaving 40 percent of bit barn facilities operationally constrained by the availability of power.

"The explosive growth of new hyperscale datacenters to implement GenAI is creating an insatiable demand for power that will exceed the ability of utility providers to expand their capacity fast enough," claimed VP Analyst Bob Johnson.

Power shortages are likely to continue for some time as new power transmission, distribution, and generation capacity will take years to come online, Gartner said. It advises enterprises to anticipate higher power costs when evaluating plans, and negotiate long-term contracts for hosting services at reasonable rates for power.

US energy companies need to revamp the way their businesses operate to accelerate resource planning and construction for additional or upgraded power generation, transmission, and distribution lines, Bain & Company recently concluded.

The reassurinly expensive consultants are lining up to dish out advice. However, if you're an industry bigwig such as Bill Gates or former Google chief Eric Schmidt, then your rather quaint faith in technology means you believe that AI is going to solve the very issues it is causing.

Datacenter operators are trying to take matters into their own hands with schemes to bring more generating capacity online, either by feeding this into the grid or providing onsite power for their own campuses.

Google, for example, has a newly forged agreement with solar energy firm Intersect Power to construct new bit barn capacity alongside additional power generation wherever possible, to create industrial parks powered by clean energy.

The energy sources being considered by industry cover the whole gamut from wind to nuclear power and even electricity generated from geothermal energy in a deal struck between internet giant Meta and Sage Geosystems in August.

According to Schneider Electric's datacenter chief, on-site deployments of gas-powered turbines are the best short-term option. Pipeline companies have confirmed that talks are underway with some operators to supply them with gas, and the opportunities must be tempting as fossil fuel giants including Chevron and ExxonMobil are now looking to get in on the game.

Others are itching to hit nuclear button. Amazon managed to acquire an existing data facility built next to a Susquehanna nuclear power plant in Pennsylvania in 2024, while Microsoft is seeking to restart the inactive Three Mile Island nuclear power plant.

Nuclear energy is likely to be a long-term gamble, however, especially the development of small modular reactors (SMRs) that could be sited on a datacenter campus. These aren't expected to be ready until the mid 2030s, but this hasn't stopped the likes of Amazon doling out half a billion dollars to companies working on this technology.

• Google thinks the grid can't support AI, so it's spending on solar for future datacenters
• Datacenters could blow up your electric bill thanks to AI
• Energy companies told to recharge for AI datacenter surge
• Energy exec punts datacenter power options out to long term

In the meantime, the AI-driven thirst for energy is also causing environmental concerns, not least because it is prolonging the life of coal-fired plants in the US. A power company in Omaha had to abandon plans to stop burning coal to produce electricity, it was reported recently, because of the demand from local bit barns.

We reported in September on estimates that the datacenter industry is set to emit 2.5 billion tonnes of greenhouse gas (GHG) emissions worldwide between now and the end of the decade, three times more than if generative AI had not been developed.

Industry giants Microsoft and Google revealed in their respective sustainability reports that catering to AI has knocked their plans to be net-zero by 2030 off course, with Redmond's CO₂ emissions up 29.1 percent from the 2020 baseline, while Google's greenhouse gas emissions have shot up 48 percent since 2019.

The amount of water being used up by datacenters for cooling AI accelerators is also of a hot topic, though disputes continue to run over the exact size of the problem. In Virginia, for example, it was claimed that facilities had upped their use of drinking water by more than 250 percent between 2019 and 2023, while others claimed that most sites use recycled sewage water and newer facilities use no water for cooling.

Issues like these are increasingly falling under the eye of regulators, with the European Commission's Energy Efficiency Directive requiring all but the smallest of datacenters to report on metrics such as their water, power consumption, and heat reuse.

This was cited as a concern by Uptime Institute, which claimed that the majority of facilities are only able to report on just two well-established metrics – power consumption and Power Usage Effectiveness (PUE).

Uptime's review also revealed that rising costs are seen as the biggest issue for datacenter operators, specifically high prices for energy and labor, plus costly infrastructure upgrades.

Staffing remains a worry, with datacenter expansion driving up demand and competition for already scarce skilled workers. The survey found that in North America and Europe there is difficulty finding skilled junior-level operators, while in China and the Middle East there is a lack of experienced operations managers.

However, while the datacenter industry faces a number of challenges, one thing it seems unlikely to run out of is money. Investors remain confident in their returns from the sector over the next several years, thanks to AI and cloud services. At least 70 percent expect to see funding for projects continue despite potential power supply issues. ®