AI Has an Energy Problem. Who Pays for It?
Artificial intelligence was supposed to live in the cloud. Increasingly, it looks more like heavy industry
SIGNAL | OpenAI’s planned Ohio campus shows how quickly artificial intelligence is becoming an energy infrastructure question. The challenge is no longer simply whether enough electricity can be generated — but who builds the system around AI, who finances it and who ultimately carries the cost.
Nvidia has agreed to provide substantial financial backing for an enormous AI campus being developed by SB Energy in Pike County, Ohio, with OpenAI as its customer. The site could eventually provide 8 gigawatts of AI compute capacity under a 20-year lease. Nvidia will also invest $1.5 billion in SB Energy and supply the AI infrastructure used on the campus.
That makes Ohio more than another data-centre announcement. It shows what happens when the AI race moves beyond models and chips and starts colliding with the physical limits of the economy.
🟦 Why is a chipmaker suddenly backing energy infrastructure?
Selling AI chips increasingly depends on much more than producing the chips themselves. Large AI systems need land, buildings, cooling, transmission capacity and enormous quantities of reliable electricity. Without those layers, additional computing capacity cannot be deployed.
Nvidia is therefore moving further into the infrastructure surrounding its own market. At PORTS-Pike, it will be the exclusive AI compute infrastructure provider while also providing credit support connected to land, power and the shell buildout for the initial 4.25 GW phase. That creates an unusual industrial loop.
Nvidia helps make the infrastructure possible that allows OpenAI to buy more Nvidia compute. The AI supply chain is starting to finance itself. When infrastructure becomes a bottleneck to future chip sales, supporting that infrastructure becomes part of the commercial strategy. Nvidia is no longer only financing technology. It is helping de-risk the physical system required to deploy it.
🟦 Is electricity becoming a bigger constraint than chips?
It may become one of them. The International Energy Agency expects global data-centre electricity consumption to more than double by 2030, reaching around 945 TWh. AI workloads are the primary accelerator of that growth, with the United States accounting for the largest share of projected growth, followed by China.
But the problem is not simply the total amount of electricity. It is where and when that electricity is required. A new AI campus can concentrate several gigawatts of demand in one region. Electricity networks, substations and generating capacity cannot necessarily be expanded at the same speed as computing hardware.
The bottleneck therefore moves. First there were not enough advanced chips. Soon, the harder question may be whether there is enough infrastructure behind them.
🟦 So where will all that power come from?
There is no single answer. AI infrastructure is likely to rely on a mixture of natural gas, renewables, nuclear power, storage, grid expansion and more efficient computing. At the same time, data centres can become more flexible about when certain workloads run and where compute is located. Ohio illustrates the scale of what may be required.
SoftBank and SB Energy plan at least 10 GW of new generation connected to the wider project, while $4.2 billion is expected to go into regional power infrastructure through a partnership involving AEP Ohio. That is no longer an IT procurement decision. It is energy planning.
🟦 But who pays for the grid around AI?
This may become the more difficult political question. Building new AI capacity can require new substations, transmission lines, generation and grid reinforcement. Those assets have to be financed somehow.
Should the technology companies creating the demand pay? Should utilities finance them through their balance sheets? Should governments treat them as strategic infrastructure? Or should part of the cost eventually appear in electricity tariffs paid by businesses and households?
Ohio is already experimenting with that boundary. AEP Ohio has introduced a separate tariff framework for large data-centre loads, including stronger financial commitments from customers seeking large connections. That matters because AI infrastructure creates another mismatch.
Grid assets are built for decades. Technology cycles can change within years. If expected demand fails to materialise, utilities and regulators must determine who carries the risk of infrastructure built around it. That makes grid tariffs more than an electricity issue. They are becoming part of industrial policy.
The principle matters far beyond Ohio. The benefits of AI infrastructure may be widely distributed. Its infrastructure costs should not automatically be.
🟦 Does the rest of the world face the same problem?
Yes — but not under the same conditions. The United States has enormous capital markets, energy resources and land, allowing projects such as PORTS-Pike to combine computing, generation and finance at extraordinary scale.
China has another advantage: it can coordinate large infrastructure systems and increasingly connect the geography of computing with the geography of available energy.
Europe faces a different challenge. It wants more sovereign AI capacity while simultaneously modernising electricity networks, expanding clean generation and electrifying transport and industry.
Europe is therefore not adding AI demand to an empty system. In many regions, new industrial facilities, renewable generation, housing and electrification are already competing for limited grid connections. AI adds another large claimant to that queue. Europe’s AI ambitions will therefore compete not only for capital and talent, but for physical access to the electricity system itself.
The question is therefore larger than which region develops the best model. It is which region can build the physical system beneath it.
THE SIGNAL
AI is no longer simply a software layer. As compute requirements expand, artificial intelligence is beginning to pull semiconductors, electricity generation, grids, data centres and capital into one increasingly integrated infrastructure system.
Ohio offers a glimpse of what that system may look like. Nvidia does not merely sell the chips. It helps secure the infrastructure in which those chips will operate. Energy companies build around the compute. Financial commitments stretch decades into the future.
The next phase of the AI race may therefore be decided somewhere surprisingly old-fashioned: at the power plant, the substation and the electricity grid. And once that happens, one question becomes difficult to avoid.
Who pays for it?
Credit
Illustration: Altair Media / AI-generated
Caption
AI is becoming physical infrastructure. As compute demand grows, chipmakers, data centres, energy companies and electricity grids are becoming increasingly intertwined — raising a new question over who builds, finances and ultimately pays for the system beneath AI.
