• by Laurie Sullivan , 86 minutes ago

During a recent all-hands meeting, Amin Vahdat, Google vice president and general manager of AI infrastructure, told employees that to meet demand for computing power, the company must quickly build out its incremental capacity.

These incremental increases in computing and processing power will affect not only how ads are created — in speed and timing — but how they are served to consumers, especially as agentic consumer purchasing becomes an everyday occurrence.   

At a presentation to employees, Vahdat said Google must double computing power every six months, CNBC reported. A slide in the presentation showed capacity of about 1000x in within four to five years.

Doubling computing power every six months and the projected 1000x capacity increase are crucial to advertisers because this immense computational power is required for today's and future ad systems.

Complex machine-learning algorithms and agentic buying systems require significant computing power to analyze data and adjust bidding strategies and ad placements in milliseconds, to a much greater extent than in the past.

Real-time ad bidding, analytics, and optimization are intended to improve return on ad spend (ROAS) and lower cost per acquisition (CPA) by focusing on bringing data into the platform to adjust budgets on the impressions most likely to convert to sales.

“The competition in AI infrastructure is the most critical and also the most expensive part of the AI race,” Vahdat said at the meeting, CNBC reported.

The presentation occurred about a week after Alphabet reported better-than-expected third-quarter results.

Vahdat’s compute-capacity prediction sounds similar to Moore's Law -- the observation that the number of transistors in an integrated circuit (IC) doubles about every two years. This has been replaced with Google’s prediction for compute capacity.

Gordon Moore, an American businessman, engineer, and the co-founder and emeritus chairman of Intel Corporation, wrote an article in 1965 that forever changed how engineers think about technology. In the article, he predicted the future of the semiconductor industry.

The goal for Google is to provide an infrastructure that is far “more reliable, more performant and more scalable than what’s available anywhere else,” Vahdat said. It also will need more efficient models, achieved through the company’s custom silicon technology.

Last week in a blog post, Google announced the public launch of Ironwood, its seventh-generation Tensor Processing Unit which the company engineers -- saying that it is nearly 30 times more power-efficient than its first Cloud TPU from 2018, with the capacity to support AI models such as Gemini 3 and other Google models.