Share with your CFO
Google’s latest earnings report revealed a capex (capital expenditure, the money companies spend on physical infrastructure like data centers and chips) revision so large it’s rattling the entire AI investment thesis. The company raised its 2026 spending estimate to as much as $205 billion, a figure where even the new floor exceeds last quarter’s ceiling. As Elizabeth Lopatto reports at The Verge, the broader AI buildout is showing simultaneous signs of cost escalation, pricing pressure, and circular financing, with Nvidia guaranteeing OpenAI’s debt while Oracle’s credit risk hits an 18-year high.
What this means for your business
The structural problem here isn’t the dollar figure. It’s that Google, with more financial forecasting horsepower than almost any organization on earth, couldn’t predict its own infrastructure costs within a $15 billion band. That forecasting failure is the tell. Every enterprise finance leader who has approved AI infrastructure investment based on vendor-supplied cost projections should treat this as a data point about the reliability of those projections, not just as background noise from a hyperscaler’s earnings call.
The circular financing dynamic deserves direct attention. Nvidia is reportedly engaged in deal talks totaling roughly $750 billion, including guaranteeing OpenAI’s debt. When the dominant chip supplier starts backstopping the debt of its own customers to keep demand flowing, that’s not a bullish demand signal, it’s a supplier propping up a market that can’t fully sustain itself. The practical implication: AI infrastructure pricing may currently reflect vendor-supported demand rather than organic enterprise willingness to pay, which means quoted costs and vendor roadmaps carry more uncertainty than they appear to.
The China angle compounds this. Chinese AI labs are producing competitive models despite constrained access to high-end GPUs, which suggests the compute intensity assumptions baked into US hyperscaler spending plans may be structurally wrong. If capable models can be built with fewer chips, then the economic case for current data center buildout scales rests on a premise that is actively being falsified. That’s worth watching on your next infrastructure renewal: the vendor telling you compute costs are stable is working from a model that may already be obsolete.
Concept deep-dive: Circular financing
Circular financing occurs when a supplier funds the customers who buy its products, creating the appearance of demand that the market hasn’t independently generated. Think of a car manufacturer offering zero-percent loans to move inventory: sales look strong until the financing dries up. In the AI context, Nvidia guaranteeing the debt of companies that use its chips to run AI workloads means the “demand” signal embedded in Nvidia’s order book is partially self-created, which makes it an unreliable indicator of actual enterprise consumption.
Based on reporting from AI’s finally expensive enough to make Wall Street nervous, originally published 2026-07-28 15:33:00.
