Google has more money to spend on AI infrastructure than most countries spend on roads. It still can’t conjure data centers out of thin air.
Speaking on the Big Technology Podcast, Alphabet CEO Sundar Pichai said lining up compute capacity for customers planning around 2027 is a serious challenge. He called the situation “incredibly difficult,” which is a striking admission from a company whose guidance calls for $195–205 billion in capital spending in 2026.
A 2024 plan meeting 2027 demand
Pichai’s core point is about timing. The compute Google can offer today was locked in by decisions the company made back in 2024.
At that point, the full wave of agentic AI demand had not yet arrived. Agentic AI refers to systems that carry out multi-step tasks on their own, rather than answering a single prompt and stopping.
Data centers come with extended lead times, so a capacity decision made today pays off years later. Demand, meanwhile, moves on a much shorter clock.
Pichai said Google has leaned on short-term leases to help customers in the meantime. According to the research findings, the company started using that approach as early as 2025.
Research first, revenue second
Pichai also indicated that Google is putting its AI research and development ahead of immediate revenue when deciding where compute goes.
Google is also not alone in feeling the squeeze. Amazon Web Services is reportedly sold out of capacity through 2028, which suggests the crunch is industry-wide rather than a single company’s planning miss.
The scale of the build-out
The industry is not standing still. Tens of gigawatts of data center capacity are under construction or in advanced planning for 2027, according to the research findings.
On top of that, more than $1 trillion in non-cancellable lease contracts have been signed so far this year. Non-cancellable is the key word: these are commitments companies cannot easily walk away from if conditions change.
Alphabet’s own spending plan tells a similar story. Capital expenditure guidance of $195–205 billion for 2026 points squarely at AI infrastructure.
Pichai framed the current constraints as part of a structural adjustment for the industry rather than a temporary hiccup. He said he keeps a long-term view on how this infrastructure will eventually be used as AI demand keeps climbing.
What this means
For businesses building on AI, the immediate takeaway is practical. Compute for 2027 is scarce, and customers who wait to secure it may find the shelves bare. Short-term leases can help, but they are a workaround rather than a guarantee.
The research findings suggest companies tied to data center construction and management could benefit, since demand for capacity looks set to stay strong for the next several years.
The risk sits in the commitments themselves. Over $1 trillion in leases that cannot be cancelled assumes demand keeps rising. If AI usage grows more slowly than expected, those obligations stay on the books regardless.
The other detail worth watching is the gap between planning and delivery. Google’s current shortfall traces back to choices made in 2024. That means the decisions being made right now will shape who has capacity, and who is stuck waiting, well beyond 2027.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.