A looming shortage of electrical power is set to constrain datacenter expansion, potentially leaving many industry growth forecasts looking overly optimistic.
In its latest report, "Five Predictions for 2026," Uptime Institute says that power will become the defining constraint on datacenter growth in 2026 and beyond.
This is because it simply isn't possible to add extra grid and generating capacity at the same rate as new server farms are popping up, so something is going to have to give.
The AI-driven infrastructure boom can be traced back to the release of ChatGPT at the end of 2022. But even before that, a shortage of available power saw moratoriums slapped on new datacenter builds in Amsterdam, Dublin, and Singapore, as well as delays to projects elsewhere.
Since then, the rate at which new capacity is added has effectively doubled, with Uptime estimating that the total global datacenter power load associated with AI will hit 10 GW by the end of 2026.
Part of the problem is that some of the planned AI facilities are enormous, but these can likely still be built in about three years, while a large solar farm takes about five years, a wind farm or gas turbine generator site around six, and nuclear power plants at least a decade.
When average datacenter sizes were smaller, pockets of spare grid capacity could be found to serve them, Uptime says. But today, projects are being measured in hundreds of megawatts, and even gigawatt facilities are planned. Sourcing this amount of power in the timescales needed is a challenge in almost any location.
The first wave of large campuses opted for one of two strategies to secure large amounts of power at short notice, the report states – on-site gas turbines or to repurpose power formerly used by cryptocurrency miners.
But high demand means that buyers are now facing shortages of gas turbine generator equipment, with a three to four-year waiting list, and prices have doubled in some markets.
Meanwhile, the supply of available sites formerly occupied by cryptocurrency mining operations in places where power is (or was) plentiful and cheap, is running out.
Uptime spells out the bottom line bluntly: "It is unclear how the industry will continue to deliver capacity at the rate that many projections forecast."
But even if datacenter growth slows, there still remains the problem of their greenhouse gas emissions, and operators will increasingly consider carbon capture and storage (CCS) schemes in future, the report claims.
A projected 75-125 GW growth in server farm power requirements through 2030 will likely have to be met through on-site and grid-based natural gas generation, putting the industry's aggressive net-zero emission commitments at risk.
These commitments can't simply be disregarded, as they are enforced by regulations in some regions, or baked into contracts with enterprise customers, which need to account for emissions in their own corporate sustainability reports.
However, Uptime notes that of six carbon capture technologies it identifies, only amine solvent systems are commercially available. These systems use absorption by chemical solvents to remove carbon dioxide from the exhaust gas stream.
There are five others, of which solid sorbent and chilled ammonia systems appear to be the most promising, but these are still in early development and not markedly cheaper or more efficient, the report states.
But storage of captured carbon is a major limiting factor in the adoption of such systems, depending on the availability of an accessible (and permitted) geological formation for this purpose. This could become a significant factor in where datacenters are sited, Uptime states, and will also likely rule out retrofitting CCS systems to most existing generation capacity.
According to the report, Google has already inked a power purchase agreement (PPA) with a gas-fired generator site using CCS in Illinois, while Microsoft has expressed its openness to procuring electricity from such facilities.
One of the main issues facing CCS systems to date has been cost, but Uptime says the economics look better compared with the cost of procuring carbon offsets or energy attribute certificates (EACs), which can range from $10 to $1,000 per metric ton. The cost of electricity from a generator site using CCS will be comparable to or less expensive than buying energy plus offsets, making it competitive with other low-carbon energy options.
Interestingly, Uptime claims that amid all the AI hype, most datacenter operators are "continuing to plan in the face of uncertainty," rather than executing on clearly AI-driven demand. Meanwhile, the need for general-purpose compute, storage, and connectivity remains strong and is likely to persist regardless of the trajectory of AI applications.
But a segmentation in the industry is emerging. The most demanding AI workloads are driving a distinct class of high-density infrastructure with greater power and cooling requirements. The majority of datacenter capacity, however, continues to support more traditional workloads. This separation may ultimately benefit the industry by clarifying risks, investment priorities, and operational strategies. ®
Source: The register