The rapid advancement of artificial intelligence has ushered in a new challenge for global tech giants — securing enough electricity to power the next wave of AI models.

Despite leading the industry, OpenAI Chief Executive Officer Sam Altman and Microsoft CEO Satya Nadella admit that neither can yet determine the exact amount of power required to sustain future AI growth.

In a recent conversation on the BG2 podcast, Nadella highlighted a surprising shift in industry constraints.

While concerns in recent years focused largely on acquiring high-performance GPUs to train AI systems, companies are now discovering that hardware is outpacing the available power infrastructure needed to operate it.

“The biggest issue we are now having is not a compute glut,” Nadella said. “It’s power — and the ability to construct data centers quickly enough near reliable energy sources.”

He revealed that Microsoft currently has chips sitting unused because there is not yet sufficient power-ready data center capacity to operate them.

“I don’t have warm shells to plug into,” he added, referencing facilities prepared for equipment installation.

This marks a pivotal transition for companies historically grounded in fast-moving, highly scalable technologies such as software and silicon.

The energy sector, by contrast, requires long-term planning, multi-year infrastructure development, and regulatory coordination — a pace far slower than the rapid evolution of AI.

Electricity demand in the United States was relatively stable for more than a decade. However, surging growth in data center construction over the past five years has begun to outstrip utility expansion plans.

As a result, many developers are bypassing public grids by building direct power supply systems to feed energy straight into their server facilities.

Altman warned that this rush for energy could carry financial risks. If cheaper and more efficient energy sources emerge sooner than expected, companies locked into high-cost long-term power contracts could face significant losses.

He also emphasized the speed at which AI efficiency and capability are advancing. “If we can continue this unbelievable reduction in cost per unit of intelligence — averaging around 40 times improvement yearly — that’s a very steep infrastructure challenge,” Altman said.

To address this, Altman has invested in several energy ventures, including nuclear fission company Oklo, fusion firm Helion, and solar thermal startup Exowatt.

While these technologies hold promise for scalable, low-cost power, many are still years away from widespread deployment.

In the meantime, tech companies are increasingly turning to solar energy due to its rapid setup time, modularity, and declining cost.

The sector’s similarities to semiconductor manufacturing — both rely heavily on silicon and can scale through modular installation — make solar a more familiar and adaptable option for data center operators.

Still, even solar requires time to deploy, and AI demand is rising faster than energy infrastructure can be built.

While Altman acknowledges the possibility that some investments may outpace actual demand, he continues to project exponential growth in the need for compute power, aligning with the economic principle known as Jevons Paradox: greater efficiency often leads to greater consumption.

“If the price of compute fell by a factor of 100 tomorrow,” Altman said, “usage would increase by far more than 100.

There are countless forms of innovation that become feasible only when compute becomes dramatically cheaper.”

As AI accelerates, the global energy landscape is being reshaped in real time. What was once a race for better chips is now a race to power them.

The world’s leading tech companies may find themselves not just software innovators — but major players in the future of global energy development.

Source: Techcrunch