Google's Pichai warns AI computing shortages could persist into 2027

The Google CEO says demand for computing power is outstripping supply across the industry, pushing companies to explore new ways to secure chips, data centre capacity and electricity.

Alphabet CEO Sundar Pichai has warned that the artificial intelligence industry is struggling to secure enough computing capacity to keep pace with rising demand, with Google facing constraints that could make it difficult to fulfil customer requirements as far ahead as 2027.

"Today, when you sit and look at anything, everyone is short, not just us, and everyone wishes you had 2x of what you had planned for," Pichai said in an interview with Alex Kantrowitz on the Big Technology Podcast this week.

Pichai said Google's current computing capacity reflects investment decisions made in 2024, highlighting the difficulty of anticipating demand in a fast-moving industry. He added that securing computing capacity for customers seeking access in 2027 was already proving challenging.

"If today a customer is coming and asking us for compute in '27, that is incredibly difficult to do," he said.

The comments underscore the infrastructure challenge facing AI companies as they race to develop increasingly capable models and deploy them across consumer and enterprise products. The constraints extend beyond processors to the data centres, electricity supply and supporting infrastructure required to run AI workloads at scale.

Pichai also pushed back against comparisons between the current AI infrastructure boom and previous waves of overinvestment in technologies such as fibre-optic networks. He argued that computing capacity was already being used, with Google facing shortages on a daily basis.

"We are at early stages of the most consequential technology ever. This is not like fibre. It's not dark. We are using it. We are capacity constrained on a daily basis," he said.

He pointed to long-term customer contracts with committed revenue as evidence that Google's continued infrastructure investment was supported by tangible demand, rather than simply expectations of future AI adoption.

Google needs space to grow

The scale of future computing requirements is also prompting Google to explore alternatives to conventional data centres, including the possibility of placing computing infrastructure in space.

Pichai said Google had been working on Project Suncatcher for two to three years, exploring whether solar-powered satellites equipped with the company's AI chips could eventually support machine learning workloads.

"It is tough for us to exponentially extrapolate our compute needs 20 years out, and I don't think Earth would look nice if that's how you plan to do it," Pichai said, explaining the thinking behind the project.

Google's research explores the possibility of placing AI computing infrastructure in sun-synchronous orbits, where satellites could receive sunlight for much of their operational time. The company is also investigating the challenges of operating its Tensor Processing Units (TPUs) in space, including radiation exposure, thermal management and communications between satellites.

The project remains experimental and is not an immediate solution to Google's current capacity constraints. However, it reflects the scale of the long-term infrastructure challenge technology companies are beginning to contemplate as AI computing requirements expand.

The other constraint: Power

The shortage of computing capacity is also drawing attention to the electricity required to operate AI infrastructure. Google's latest move in the US illustrates how technology companies are looking beyond conventional power procurement to unlock additional capacity from existing electricity grids.

On October 22, Google announced an expanded agreement with energy technology company Voltus to develop a 1-gigawatt virtual power plant across the PJM electricity grid region in the United States.

The agreement builds on an earlier 100-megawatt arrangement and aggregates flexible electricity resources from homes and businesses, including batteries, smart thermostats and industrial assets. These resources can adjust or temporarily reduce consumption when electricity demand is high, helping support grid reliability.

The initiative does not represent 1 GW of new power generation. Instead, it seeks to co-ordinate existing resources to make electricity demand more flexible and improve the availability of capacity on the grid.

The agreement highlights another dimension of the AI infrastructure race: securing enough electricity and making more efficient use of available grid capacity could become increasingly important as data centres expand.

For Google, the immediate challenge is ensuring it has enough computing power to serve customers and develop frontier AI models. Its longer-term exploration of space-based computing and its efforts to unlock flexible electricity capacity point to different approaches to a problem Pichai says is already affecting the industry.

The common thread is the growing difficulty of scaling infrastructure at the pace AI demand requires -- and the possibility that the next phase of the race will be shaped as much by access to computing power and electricity as by advances in the models themselves.