Google is raising its 2026 capital expenditure target to $205 billion, a $15 billion increase from prior guidance, because demand for AI compute still outruns the massive infrastructure the company has already built. The new figure, disclosed on Wednesday’s Q2 earnings call, also comes with a warning from CFO Anat Ashkenazi that spending will “continue increasing significantly into 2027.” The message is unambiguous: the hyperscaler arms race is accelerating, not plateauing.
The supply constraint is the story
“We’re still in a supply-constrained environment,” Ashkenazi said, adding that capacity has grown significantly over three years yet demand still outpaces investment. Cloud revenue jumped 82 percent in the quarter, driven by Google Cloud Platform’s enterprise AI products and infrastructure. The company also booked its first revenue from tensor processing unit systems deployed directly into customer data centers. Meanwhile, token consumption, the meter by which AI usage is priced, is scaling fast. More than 2,000 enterprises consumed over 100 billion tokens in the last year, and nearly 500 Google Cloud customers processed over 1 trillion tokens. Chief executive Sundar Pichai described the momentum as spanning data analysis, customer management, agent building, automation and cybersecurity.
Power, permits and a billion-euro fine
A Synergy Research Group report published Thursday projects U.S. data center capacity will double in three years, with hyperscaler-owned operational capacity doubling in two. The same report notes that constrained power availability and rising local opposition are crimping new plans, though developers “will continue to find ways around those issues.” On the same day Google outlined its expanded buildout, the European Commission fined the company 890 million euros ($1.01 billion) for self-preferencing its services on Search and restricting businesses from steering consumers to cheaper alternatives. The penalty is a reminder that regulatory risk scales alongside compute budgets.
Tokens become the new cost center
The token metrics Pichai cited are not just vanity numbers; they are becoming a direct cost concern for enterprises running agentic workloads at scale. As agents consume more compute per task, legacy IT systems face strain that existing budgets did not anticipate. Google’s capex trajectory suggests the company expects that strain to keep growing faster than efficiency gains can offset it. The next quarter will show whether the $205 billion plan narrows the supply gap or simply defines a new, higher floor for the industry’s capital intensity.