The Conundrum of Investing in AI Today

The AI stack is moving from a single-trade “buy picks-and-shovels” regime into a three-way squeeze: model vendors need pricing power, hyperscalers need utilization, and enterprise users need productivity gains. That combination is inherently unstable because margin expansion cannot simultaneously accrue to suppliers and customers forever; the first sign of strain will likely be usage caps, seat reductions, or slower incremental spend, not a collapse in headline adoption. The market is still pricing AI as a perpetual demand expansion story, but the more likely path is a normalization phase where only the most differentiated workloads retain premium economics.

Second-order winners are shifting toward businesses with either unavoidable workflow integration or pricing leverage. Coding and infrastructure-heavy use cases should keep compounding because they are objective, measurable, and easier to monetize; broad consumer-facing assistants are much more exposed to commoditization and subsidy withdrawal. That argues for staying selective on platform exposure: large incumbents can absorb the capex, but they may not earn a commensurate ROIC unless they can convert AI into higher ad yield, higher cloud attach, or workflow lock-in.

On the public names discussed, the best risk/reward appears in the “boring but essential” pockets where execution can improve without requiring a perfect AI outcome. Logistics and aerospace/defense suppliers can benefit from normalization in travel and procurement cycles, while healthcare logistics has a path to outsized upside if adoption remains durable. By contrast, the consumer-platform AI push looks strategically noisy: if management misallocates capital chasing a low-willingness-to-pay market, the returns could lag even if the underlying product is impressive.