Data centre restrictions risk slowing the expansion of artificial intelligence infrastructure and driving investment overseas, according to technology investor Serhiy Tokarev, who argues that governments should prioritise new generation and grid capacity instead.
The warning comes as electricity networks struggle to keep pace with demand from large computing facilities. Ireland’s data centres already consume almost a quarter of the country’s electricity, while new grid connections around Dublin were effectively frozen for three years.
Under revised Irish rules, new projects must provide generation or storage capacity either on site or nearby. Amsterdam has previously imposed a moratorium on new facilities, while Frankfurt regulates where they can be built through zoning.
New York State introduced a moratorium of up to a year in July 2026 on certain environmental permits for large data centres. Officials said the pause would allow them to assess their effect on the power grid and the environment. By May, projects representing almost 12GW of potential demand were waiting in the state’s grid interconnection queue.
Tokarev, co-founder and general partner of Roosh Investment Group, said his experience investing in technology companies had made energy infrastructure a key consideration for businesses and investors. Reliable electricity, he argued, was increasingly decisive in determining whether companies could expand and attract capital.
Data centre restrictions and the global AI race
He said the ability to control computing infrastructure was becoming an important source of economic power. Domestic capacity can reduce the risk that a foreign provider changes its terms, raises prices or restricts access to services on which entire industries rely.
That issue is central to competition between the United States and China. American firms remain prominent across the AI industry, including advanced models, chips and cloud platforms, while Washington regards the energy supply supporting data centres as a national security concern.
China is directing computing development towards its western regions through the “Eastern Data, Western Computing” programme, where land and energy are more readily available. It is also expanding generation, adding almost 500GW of renewable capacity in 2025, according to the International Energy Agency.
Europe risks falling further behind, Tokarev said. Brussels plans up to seven AI gigafactories and aims to at least triple the European Union’s data centre capacity within five to seven years, but construction is being slowed in some countries and European companies continue to rely mainly on US cloud providers.
Restricting construction does not remove demand for AI, he argued, and could instead push investment abroad while increasing reliance on overseas infrastructure and technology.
Grid bottlenecks are concentrated in key locations
Data centres accounted for about 70 terawatt hours of the 800TWh increase in global electricity consumption in 2025, according to the International Energy Agency. Industry, air conditioning, heat pumps and electric vehicles contributed considerably more.
The immediate difficulty is the rapid concentration of demand in a limited number of locations, where power must be delivered to new computing facilities quickly.
In the United States, lead times for large transformers have risen beyond two years, compared with roughly 50 weeks before the pandemic. Morgan Stanley estimates that American data centres could face a power shortfall of about 49GW by 2028.
Satya Nadella, Microsoft’s chief executive, has acknowledged that electricity has become the company’s “binding constraint”. Some of the firm’s graphics processing units are reportedly in inventory because there is nowhere to connect them.
New gas-fired plants cannot provide an immediate solution. GE Vernova, Siemens Energy and Mitsubishi Power account for about two-thirds of global gas turbine production, while their factories are booked years in advance. GE Vernova reported 116GW in gas power equipment orders and reserved production slots, and was targeting annual turbine output of 20GW in the third quarter of 2026.
The rising power density of servers is creating an additional design challenge. A rack once typically required between 7kW and 12kW, while systems based on Nvidia’s GB200 require between 120kW and 132kW. Conventional air cooling is no longer sufficient at that level, meaning new sites must be designed for liquid cooling from the beginning.
Technology companies are increasingly trying to secure supplies themselves. Microsoft has agreed to buy electricity from the 835MW Three Mile Island reactor, which is being restarted to serve its needs, while Google has signed an agreement with Kairos Power for small modular reactors, with the first expected to come online by 2030.
Ukraine’s distributed power response
Ukraine’s experience during the war illustrates how pressure on the centralised grid can accelerate the development of local generation. Repeated attacks have forced the country to deploy power sources close to businesses and other users where the wider network could no longer guarantee reliable supply.
Ukraine has about 1.8GW of distributed gas-fired generation, with more than 500MW under construction, and is streamlining procedures for grid connections and permits.
Tokarev said the Ukrainian model could not be directly transferred to large AI data centres, but showed how treating energy resilience as a priority could encourage investment in generation and speed up connection processes.
He called for governments, grid operators and businesses to plan energy and digital infrastructure together. Measures could include reserving grid capacity for new facilities, expanding transformer production, encouraging long-term power purchase agreements and supporting generation and storage near data centres.
Businesses could also invest in their own plants, storage and cooling systems, while reducing electricity use at peak times in return for faster grid connections. A combination of grid power, gas, nuclear generation and renewables may be needed because no single source can meet the demand alone.
The ability to build and power data centres quickly is likely to play a growing role in investment decisions. Tokarev said success in the AI race would depend not only on models and chips, but also on whether countries could expand computing capacity, electricity generation and grids at the same time.
