Global AI revenue outside China hit $25 billion in Q1 2026, exceeding the industry's estimated $21 billion in depreciation costs tied to data center and chip investments for the second consecutive quarter, per research from Exponential View cited by Bloomberg. The milestone matters: it is the first systematic industry-level evidence that AI infrastructure spending may be self-sustaining rather than purely speculative. Azeem Azhar, Exponential View founder, told Bloomberg: 'For now, the economics are holding. But the margin for error is narrow.'
The caveat is sharp: depreciation consumes more than two-thirds of that $25B revenue, leaving a thin buffer for power, labor, and financing costs. The analysis assumes a six-year depreciation life for GPUs and servers; some investors argue this is optimistic given chip obsolescence risk, though H100 and older chips remain in strong demand four-plus years post-launch. The math remains tight: J.P. Morgan estimates the industry needs $650B in annual AI revenue just to hit 10% return on infrastructure being built. Current revenue is $50-150B annually (generous assumptions), widening the gap.
The flip side: token pricing dynamics are tight. Every 10% price drop drives 12-18% more usage, meaning total spending rises as per-unit costs fall. Demand still exceeds supply. Generative AI revenue over 12 months reached $110B and is scaling 3x faster than prior tech waves (internet, mobile, cloud). Depreciation assumptions are being stress-tested by board-level spending plans: hyperscalers commit ~$725B capex in 2026, up 77% YoY.
For teams: clearing depreciation is a necessary, not sufficient, condition. It signals machines are no longer pure promise but doesn't confirm the business throws off cash. The next test: whether revenue grows fast enough to match infrastructure already ordered and financed. Margin of error is narrow; watch Q2/Q3 results for acceleration or deceleration in token demand. Open-source and Chinese models are commoditizing frontier tier faster than expected.