AI Tracker (J.P. Morgan, November 2026): $2.4 trillion of hyperscaler capex and R&D since 2022
Michael Cembalest's new AI Tracker gathers 67 pages of charts on the AI buildout and refreshes them each month. The headline numbers: hyperscalers have spent about $2.4 trillion on capex and R&D since 2022, the AI buildout is running at roughly 3.6% of GDP a year against 2.2% for the railroads at their peak, and a further $4 trillion sits off balance sheet in leases not yet started and purchase commitments. Free cash flow margins net of capex are falling, buybacks are shrinking and new investment-grade debt plus SPVs now fund a growing share of spending.
The tracker also follows model capabilities against task cost, token prices against consumption, GPU rental and memory prices, median AI spend per employee by sector, data center power costs and new cyber risks.
For founders it is the clearest single reference on whether the compute you build on keeps getting cheaper, and on how much inference revenue the industry must produce for the spending to pay off.