Google Cloud has a $462 billion backlog it cannot serve because AI demand is outstripping its own infrastructure, and its own engineers are part of the problem

Google Cloud has a $462 billion backlog it cannot serve. That figure nearly doubled in the latest quarter and represents more than seven times its 2025 cloud revenue 1. Alphabet CEO Sundar Pichai confirmed that revenue fell short of its potential because capacity constraints prevented Google from serving all waiting customers 1.

But the backlog figure only tells half the story. The more revealing question is who is consuming the GPU capacity that paying customers are waiting in line for.

Google mandates that its own engineers use AI to generate code, a directive that creates internal demand for the same GPU infrastructure external cloud customers are queuing for 1. The more successful that internal mandate is, the more it competes with the customers whose contracts make up that backlog. Google has become, in effect, one of its own largest infrastructure customers.

Amin Vahdat, a VP who leads Google's global AI and infrastructure team, told employees at a November all-hands meeting that the company needs to double its AI serving capacity every six months and reach 1000x scale within four to five years 2. Serving capacity is the ability to run AI products for users when they query them. It is distinct from the compute used to train models, and it determines whether a cloud customer waiting for Gemini API access gets served this quarter or next.

The bottleneck is physical, not financial. Shay Boloor, chief market strategist at Futurum Equities, described the constraint as "the power, the cooling, the networking bandwidth and the time needed to build these energized data center capacities" 2. Data centers take years to build. Power purchase agreements take years to negotiate. Chips take months to manufacture and ship. None of those timelines compress because an earnings call demanded more output.

Alphabet is spending at record levels and the trajectory points higher. Reuters reported in February that Alphabet's 2026 capital expenditure could double from already record levels to fund the buildout 3. A Google spokesperson told Fortune that demand for AI services means the company is "being asked to provide significantly more computing capacity" 2.

But doubling capex does not close the gap if internal demand scales faster than physical infrastructure can grow. Every engineer running AI tools to generate code draws from the same finite pool of GPU capacity that a paying cloud customer is waiting for. When that pool is already oversubscribed by a factor of seven, internal usage is not a rounding error. It is capacity drawn from the same line.

For cloud customers, the implications are direct. The provider they depend on for AI infrastructure is capacity-constrained, and the constraint is worsening even as spending soars. A backlog that nearly doubled in a single quarter signals that supply is falling further behind demand, not catching up 1. And the company's own engineering organization is standing in the same line.

The real question for investors: will self-generated internal demand keep crowding out the revenue-generating kind? If every new data center has to serve both internal engineering workloads and external customer contracts, the marginal economics of each buildout are split before construction begins.

Google's plan to double serving capacity every six months would be aggressive under any conditions. With internal AI demand compounding alongside customer demand, it may be the minimum required just to keep the backlog from growing further. Right now, the backlog is still growing.