Alphabet's Q2 2026 earnings crystallized a question that's been building all year: how much more expensive can the AI infrastructure arms race get before public markets stop rewarding it? Alphabet raised its full-year 2026 capex guidance to $195-205 billion -- a bigger annual number than Microsoft, Amazon or Meta have individually guided to -- even as Google Cloud posted 82% revenue growth and a $514 billion backlog that should, on paper, justify the spend.
The market's answer was to sell the stock anyway. GOOGL fell despite beating on both revenue and profit, precisely because the capex raise signals the spending curve keeps steepening rather than flattening -- and investors are increasingly split on whether that's rational infrastructure-building ahead of durable demand, or a self-reinforcing spending war where no single hyperscaler can afford to blink first.
The same week reinforced that this isn't just a hyperscaler phenomenon: AMD committed up to $5 billion directly to Anthropic to lock in a multi-year chip-purchasing relationship, while Anthropic separately maintains a reported $1.25 billion-a-month arrangement leasing SpaceX satellite-linked compute capacity. Capital is now flowing directly between labs and chipmakers, not just from hyperscalers into their own data centers -- widening the set of companies with direct financial exposure to the capex cycle continuing.
Total AI infrastructure spending industry-wide is projected near $1.5 trillion for 2026, and by widely cited industry estimates, the AI industry as a whole needs to generate roughly $3 trillion in economic value to justify the chips and data centers currently being built -- a gap that's easy to state and genuinely hard to verify is closing on schedule. For VCs and LPs, Alphabet's backlog-versus-capex tension is the cleanest public proxy available for the entire private AI infrastructure thesis: real revenue growth is showing up, but so is real uncertainty about whether the spending curve can keep steepening without a correction.