Google is giving its AI researchers priority over cloud customers. CEO Sundar Pichai has confirmed that the company is forgoing short-term Google Cloud revenue by directing some of its limited computing capacity toward developing advanced AI models. The decision comes as demand for Google’s cloud services continues to rise and some customers struggle to secure additional capacity.“Our cloud business is at any given moment in time forgoing short-term revenues because we are making compute trade-offs to support frontier AI development,” Pichai told Alex Kantrowitz during an interview on the Big Technology PodcastFrontier AI refers to the development of models that aim to reach the highest level of current capabilities. These systems require massive computing power to train and run, creating competition between Google’s internal research and paying cloud customers.Google Cloud generated $24.8 billion in second-quarter revenue, up 82% from a year earlier, while its contracted backlog reached about $514 billion. However, Pichai said the company must maintain enough computing power for Google DeepMind’s research, and its other businesses must plan around that need.
Why Google is prioritising AI research
Pichai described access to computing power for Google DeepMind as “non-negotiable.” He argued that developing advanced models gives Google an early view of new AI abilities and helps the company prepare products for future uses.He also expects frontier models to help researchers build their successors through recursive self-improvement. This means an AI system assists with the research, coding, or testing required to create a more capable model.Pichai rejected suggestions that Google should focus only on smaller, less expensive models such as Gemini Flash.“We won’t do it any other way,” he said about competing at the AI frontier. “We’ve been thinking about the AI frontier for well over a decade now.”
Google Cloud customers face limited computing capacity
Google’s decision comes amid a shortage of the chips, servers, and data-center space needed to run AI services. Pichai said accommodating a new customer seeking computing capacity in 2027 would be “incredibly difficult to do.”“Everyone wishes you had 2x of what you had planned for,” he said, describing the shortage across the technology industry.Building additional capacity requires decisions years before customers can use it. “The compute we have today is a function of decisions I made in 2024,” Pichai said.Alphabet expects to spend $195 billion to $205 billion on capital projects in 2026, with much of that investment supporting AI and cloud infrastructure.Its regulatory filing showed that Google Cloud had $513.9 billion in contracted revenue waiting to be recognised at the end of June. Alphabet’s second-quarter filing said slightly more than half of that backlog is expected to become revenue within two years.Demand is also coming from Google’s own products. Pichai said Gemini 4 Argon processed almost 25 trillion tokens in its first 72 hours, about three times what Gemini 3 handled over the same period. He added that Google was already capacity-constrained while serving the model.
Leave a Reply