Sundar Pichai: Google Cloud is giving up revenue to feed AI research

Sundar Pichai, chief executive of Google and its parent company Alphabet, has said that Google Cloud is giving up some short-term revenue so that the company's own frontier AI research gets first call on scarce computing power. He made the remarks in a podcast interview recorded on 20 October 2026, and they were widely reported over the following days. In a separate podcast conversation released the same week, Pichai described how software "agents" now summarise the company's daily activity for him. Alphabet shares fell about 2% on 21 October as investors weighed the cloud comments.

Cloud computing means renting out servers, storage and software over the internet instead of companies buying their own machines. For Google, Amazon and Microsoft, cloud has become one of the fastest-growing businesses, and much of the new demand comes from companies that want to train or run AI models. Those workloads need specialised chips, such as Google's own tensor processing units (TPUs) and graphics processors (GPUs), which are in short supply worldwide.

Pichai's comments are unusual because a chief executive rarely says openly that his company is turning away paying customers. His point is that when chips and data-centre space run short, Google gives its own AI research, including its Gemini models, first priority over extra rent from cloud clients.

The remarks also explain why the big technology companies keep raising their spending plans. Building a large data centre, connecting it to power and filling it with chips takes years. Pichai's statement that today's capacity was decided in 2024 shows how long the planning cycle is, and why a sudden jump in AI demand cannot be met quickly.

In the second interview, Pichai said agents running in Google's cloud now summarise the previous 24 hours of developments for him, a task he said could otherwise take about an hour a day. According to the reported remarks, the agents follow AI model progress, track how much computing each product area uses and monitor internal discussion groups. He said people, including himself, still make the important decisions about how computing resources are shared out. An AI agent here means software that can carry out a series of tasks on its own, such as reading, sorting and summarising information, rather than only answering one question at a time.

Short-term: Businesses hoping to rent large amounts of AI computing from Google for 2027 may find capacity harder to get and may spread their orders across other cloud providers. Investors will look for more detail when Alphabet next reports its quarterly results, since the size of the trade-off has not been given.

Long-term: If the bet pays off, Google's AI models and products could become stronger, which would in turn draw more customers to its cloud. If demand for AI computing slows, the heavy spending could weigh on profits.

Who is affected: Cloud customers, including start-ups and large companies in India that build AI products on rented computing; Google's cloud sales teams and partners; rival cloud providers, who may win business Google cannot serve; and chipmakers and data-centre builders, who stand to gain from the continued spending.

Questions and Answers