Google says it must double AI capacity every six months as compute race intensifies
The pressure to expand capacity has become increasingly visible inside Google
Google executives have warned employees that the company must double its AI serving capacity every six months to keep up with soaring global demand for AI services.
The message, delivered by Amin Vahdat, Google Cloud vice president and head of AI infrastructure, during an all-hands meeting on 6th November, highlights the extraordinary scale of resources required to compete in an increasingly crowded AI arms race.
According to CNBC, Vahdat told employees that Google will need to deliver 1,000 times more compute, storage, and networking capacity within the next five years.
"Now we must double every 6 months… the next 1000x in 4-5 years" he said.
Achieving that, Vahdat added, will require aggressive investment and tight integration across research, silicon design, and datacentre engineering.
"It won't be easy but through collaboration and co-design, we're going to get there."
Vahdat said the company's long-term advantages include its DeepMind research arm and continued advancements in custom silicon.
Google recently announced Ironwood, its seventh-generation Tensor Processing Unit (TPU), which the company claims is nearly 30 times more power efficient than the first Cloud TPU released in 2018.
By leaning more heavily on its own chips, Google aims to avoid the supply crunch that has plagued the broader AI industry, particularly shortages of Nvidia GPUs, which remain in extremely limited supply.
The meeting occurred a week after Alphabet reported better-than-expected third-quarter earnings and raised its capital expenditures forecast to $91-93 billion for the second time this year, with an even more dramatic spike planned for 2026.
Nvidia bottlenecks slow AI rollouts
The pressure to expand capacity has become increasingly visible inside Google.
Sundar Pichai pointed to the launch of Veo, Google's video-generation model, as an example of a product limited by compute shortages. Despite early excitement, Veo could not be rolled out more broadly because the company lacked sufficient GPU capacity to support more users.
Industrywide constraints remain severe. Nvidia reported this week that its AI chips are still "sold out," even as the company posted a 62% jump in quarterly revenue.
Google is one of Nvidia's largest customers, and the semiconductor giant says it is racing to expand output to meet unprecedented demand.
The AI bubble question
During the meeting, employees raised concerns about whether Google's sky-high spending could backfire if the AI market cools.
The fear has rippled across Silicon Valley and Wall Street in recent months amid a downturn in several AI-related stocks and worries of a possible bubble.
Pichai acknowledged that questions about a bubble are "definitely in the zeitgeist" and reiterated comments he made in a recent BBC interview, where he warned that if a bubble were to burst, "no company is going to be immune, including us."
However, he added that the greater risk lies in moving too slowly.
"The risk of underinvesting is pretty high," Pichai said.
He noted that Google Cloud, which grew 34% year-over-year last quarter to more than $15 billion in revenue, could have performed even better if more compute capacity had been available.
Pichai said that 2026 will be an "intense" year marked by fierce competition, heavy infrastructure deployment, and ongoing pressure to meet customer demand for AI and cloud services.
Though he described Google's investment strategy as disciplined, both he and Vahdat made clear that the company sees aggressive scaling as essential to long-term success.