Google Needs to Double AI Serving Capacity Every 6 Months to Keep Up with Demand

Google’s AI infrastructure chief has signaled an urgent need to drastically scale up computing resources, revealing to employees that the company must double its serving capacity every six months to keep pace with the escalating demand for artificial intelligence services.

At a company-wide meeting on November 6th, Amin Vahdat, a Google Cloud vice president, delivered a presentation titled “AI Infrastructure,” a copy of which was obtained by CNBC. A key slide highlighted “AI compute demand,” stating the ambitious objective: “Now we must double every 6 months…. the next 1000x in 4-5 years.”

“The competition in AI infrastructure is the most critical and also the most expensive part of the AI race,” Vahdat emphasized during the meeting. Alphabet CEO Sundar Pichai and CFO Anat Ashkenazi were also present, addressing questions from employees.

The presentation followed Alphabet’s better-than-anticipated third-quarter earnings report, where the company revised its capital expenditure forecast upward for the second time this year, projecting a range of $91 billion to $93 billion. Executives also foreshadowed a “significant increase” in capital expenditure for 2026. This investment reflects the intense competition with hyperscaler peers Microsoft, Amazon, and Meta, all of whom have also increased their capex guidance. The collective spending of these four tech giants is projected to exceed $380 billion this year, a testament to the fierce battle for AI dominance.

Vahdat clarified Google’s strategy: “Google’s job is of course to build this infrastructure but it’s not to outspend the competition, necessarily,” adding, “We’re going to spend a lot,” while prioritizing the development of infrastructure that is “more reliable, more performant and more scalable than what’s available anywhere else.” This highlights a focus on strategic efficiency alongside raw capital expenditure.

Beyond infrastructure expansion, Vahdat pointed to Google’s efforts to enhance capacity through more efficient AI models and its custom-designed silicon. The recent public launch of the seventh-generation Tensor Processing Unit (TPU), dubbed Ironwood, underscores this approach. Google claims Ironwood is nearly 30 times more power-efficient than its initial Cloud TPU from 2018, signifying significant strides in the energy efficiency of its AI compute.

Vahdat also noted DeepMind’s research capabilities as a critical advantage, providing insights into future AI model architectures and resource requirements, allowing for proactive infrastructure planning.

Addressing the core challenge, Vahdat stated that Google needs to “be able to deliver 1,000 times more capability, compute, storage networking for essentially the same cost and increasingly, the same power, the same energy level.” He acknowledged the difficulty of this task, emphasizing that “through collaboration and co-design, we’re going to get there.” This call to action signals a company-wide effort to optimize resource utilization and drive innovation in AI infrastructure.

Pichai echoed the urgency of the situation, telling employees that 2026 will be “intense” due to AI competition and the escalating pressure to meet cloud and compute demand. The CEO also addressed concerns regarding a potential AI bubble, a topic gaining traction among investors and in Silicon Valley circles.

An employee question read aloud: “Amid significant Al investments and market talk of a potential Al bubble burst, how are we thinking about ensuring long-term sustainability and profitability if the Al market doesn’t mature as expected?”

Pichai acknowledged the validity of these concerns, stating, “It’s a great question. It’s been definitely in the zeitgeist, people are talking about it.”

He reiterated the risks of insufficient investment and highlighted the robust performance of Google’s cloud business, which reported 34% annual revenue growth to over $15 billion in the quarter, with a backlog reaching $155 billion.

“I think it’s always difficult during these moments because the risk of underinvesting is pretty high,” Pichai stated. “I actually think for how extraordinary the cloud numbers were, those numbers would have been much better if we had more compute.” This statement highlights the direct link between infrastructure capacity and revenue potential in the current AI-driven market.

Pichai assured employees that the company maintains a disciplined approach, supported by the strength of its underlying businesses and balance sheet. “We are better positioned to withstand, you know, misses, than other companies,” he said, projecting confidence in Google’s ability to navigate potential market volatility.

Looking ahead to the coming year, Pichai cautioned employees that “there will be no doubt ups and downs.”

“It’s a very competitive moment so, you can’t rest on your laurels,” he said. “We have a lot of hard work ahead but again, I think we are well positioned through this moment.”

Google declined to comment.

The discussion around a potential AI bubble intensified ahead of Nvidia’s quarterly earnings report. Shares of AI beneficiaries such as CoreWeave and Oracle experienced declines, extending a month-long downturn. Earlier in the week, Pichai acknowledged “elements of irrationality” in the market and suggested that no company would be immune from the consequences of a potential bubble burst.

However, Nvidia CEO Jensen Huang challenged the AI bubble narrative during the chipmaker’s earnings call: “We see something very different.” Nvidia, a key supplier to Google, reported robust 62% revenue growth, exceeding expectations, and provided strong guidance for the fourth quarter.

Despite Nvidia’s positive results, market sentiment remained cautious, with Nvidia shares falling 3.2% on Thursday, contributing to a 2.2% decline in the Nasdaq. Alphabet’s stock also experienced a 1.2% decrease.

Google recently unveiled its latest AI model, Gemini 3, designed to deliver improved performance on complex tasks compared to its predecessors. This launch reflects the ongoing competition with companies like OpenAI to deploy advanced AI tools as widely as possible.

Pichai emphasized that capacity constraints are currently a limiting factor. He cited the example of the video generation tool Veo, which experienced an upgrade last month.

“When Veo launched, how exciting it was,” Pichai said. “If we could’ve given it to more people in the Gemini app, I think we would have gotten more users but we just couldn’t because we are at a compute constraint.” This demonstrates precisely how capacity limitations translate to lost opportunities.

Another frequently asked employee question inquired about the discrepancy between the accelerating rate of capital expenditure and operating income growth. The question requested the company’s strategy for maintaining “healthy free cash flow” over the next 18 to 24 months.

Ashkenazi responded, highlighting several potential avenues for growth, including migrating more customers from physical data centers to the cloud.

More broadly, she asserted, “The opportunity in front of us is significant and we can’t miss that momentum,” conveying a sense of urgency and strategic focus.

Clarification: This story has been updated to precisely articulate Amin Vahdat’s remarks on meeting demand through both capacity expansion and efficiency maximization.