AI Is a Supply Chain Rather Than an Industry: Here’s Why
John O’Connell was recently published in Kiplinger, the second in his four-part series on AI concentration risk, arguing that AI isn’t an industry but a six-layer supply chain. His point is that investors talk about AI the way they talked about “tech” in 1999, as one industry with one business model, when it actually runs through distinct layers, each with different economics, competitors, and risks: chip design (Nvidia, AMD), chip fabrication (TSMC, ASML), data movement (memory and networking), data-center construction (power, cooling, real estate), the hyperscalers writing the checks, and the software layer that has to monetize it all. The trap is that owning Nvidia, AMD, Broadcom, and a hyperscaler across three different funds feels like diversification, but it’s really one bet, sized three times over, since every layer depends on the same four customers spending $700B-plus in 2026 at a pace none has sustained before. He also flags that the lower-profile middle layers often trade at lower multiples for the same demand, and that the monetization layer only gets paid if enterprise adoption arrives on schedule, unlike the layers below it that get paid regardless. His advice is to map the top 10 holdings of every growth fund you own to a layer, check whether your exposure sits entirely in the headline-grabbing layers one and five, and ask what happens to each holding if hyperscaler capex merely slows rather than reverses, since a deceleration from 70% to 20% growth is still a down year for every layer beneath it.
