An unprecedented capital buildout
The scale of capital flowing into AI infrastructure since 2023 is, by any historical comparison, extraordinary — massive investment in data centers, specialized AI chips, and the power generation needed to run them, from both the frontier labs themselves and the cloud providers (Microsoft, Google, Amazon) and chipmakers (Nvidia, and increasingly others) that supply them. Commitments in this space are frequently large enough that precise figures are best treated as directional rather than fixed — they change too fast, and get restated too often, for any number written down today to stay accurate for long. What matters for your mental model isn't the exact dollar figure at any given moment, but the pattern: an enormous, sustained bet that AI capability will keep improving with more compute, and that the resulting products will generate enough revenue to justify the spend.
This spending flows through a genuinely long chain: capital funds chip purchases, chips need data centers, data centers need power (increasingly a real bottleneck in some regions), and all of it has to eventually be justified by revenue from AI products — subscriptions, API usage, enterprise licensing. Several of the largest deals in this space are circular in a way worth understanding: a chipmaker invests in a lab, which uses the capital to buy chips from that same chipmaker, which reports the sale as revenue. This isn't necessarily fraudulent or irrational, but it does mean some of the reported financial activity in AI is less independent evidence of real end-demand than it might first appear — a pattern worth watching rather than ignoring or automatically distrusting.