US AI Expansion Faces Power Limits as Europe Advances Adoption

Expanding artificial intelligence infrastructure is becoming increasingly difficult as electricity availability limits data center construction in the United States, while European countries continue adopting AI across their economies. Goldman Sachs said on October 12, 2026, that demand for computing capacity is outpacing supply, prompting major technology companies to secure electricity contracts extending into 2028 and beyond. The infrastructure gap also highlights a transatlantic divide.

Apollo's analysis, written by Huw van Steenis, estimates that hyperscaler capital expenditure in the United States will approach $800 billion in 2026, while US data center capacity is approximately three times Europe's. The United States also holds a dominant share of global AI models and semiconductor value. Europe, however, has recorded stronger AI adoption than the United States in several countries, according to Apollo.

The findings distinguish the regions' roles: American companies dominate infrastructure development, while European businesses are increasingly using AI systems built on infrastructure they did not develop themselves.

Goldman Sachs identified electricity availability as a growing constraint on further expansion, based on discussions with technology company management teams. Grid capacity for 2026 and 2027 is essentially fully booked, the bank said, leading major cloud and technology providers to pursue power agreements for subsequent years.

The shift means access to high-voltage electricity infrastructure can determine whether companies can deploy additional computing equipment. Although capital expenditure is rising because of component inflation and demand for more chips, the availability of power is emerging as a separate limitation on new data center projects. Goldman Sachs also reported that revenue generated per unit of power has increased by around two-thirds over the past year. The bank said computing demand remains elevated despite expectations of slower AI model training, as companies increase spending on AI safety and pacing monitoring, which also require computing resources.

Enterprise software spending is undergoing a related shift, with data management and cybersecurity platforms positioned to benefit from AI adoption. Goldman Sachs said information technology leaders are directing budgets toward organized data systems and security tools to protect information and manage risks associated with unauthorized AI use within organizations. Application software is also increasingly being designed for AI agents rather than direct human interaction. Goldman Sachs said this transition allows vendors to incorporate more advanced AI capabilities into existing products, although changes in how software is used could eventually affect pricing structures and monetization.

Meta illustrates the importance of generating returns from existing computing infrastructure. Goldman Sachs said the company's core business produces two to three times the return on investment per megawatt of power compared with the benchmark discussed in its analysis, while Meta explores additional revenue opportunities. Google, meanwhile, is using an integrated strategy spanning custom silicon, cloud computing, security and distribution, according to Goldman Sachs.

Sales of its tensor processing unit chips form one component of this broader approach, which combines hardware and services to support its core business. The investment outlook also depends on how effectively companies turn computing capacity into revenue. Goldman Sachs said established platforms with demonstrated core-business monetization remain among the more dependable uses of computing resources as infrastructure capacity stays constrained.