The market has spent the last year trying to price compute. I think the harder problem over the next few years will be pricing absorption, and those are very different problems.
I've spent time talking with funds, banks, and investors about their pain points and helping them integrate AI into their investment workflows. Compute is measured in GPUs, data centers, and capex, which is why a stock like CoreWeave moves on every whisper of softening demand. Absorption is measured by whether an organization actually changes how work gets done, and that shows up nowhere on an earnings call.
Microsoft just reported that Copilot crossed 30 million paid seats, an impressive milestone. But seat counts tell us how much AI has been purchased, not how deeply it's changed the way people work, and that's much harder to measure. Whether 2025 and 2026's infrastructure spend, and whatever gets committed for 2027 and 2028, was worth it won't be decided by hyperscalers alone. It will be decided by whether accountants, lawyers, traders, consultants, engineers, marketers, and small business owners actually change how they work every day. For anyone underwriting the spend, that's a risk. For whoever helps close the gap, it's the opportunity.
Markets know how to price GPUs. They're still learning how to price human behavior.