AlixPartners Co-CEO Rob Hornby on AI Bull Case

Rob Hornby says AI costs are falling faster than costs for comparable technologies historically, but remain substantial. He characterizes the sector as being in a transition period, with investment underway and returns uneven; moving from flat fees to per-token charges has made costs more visible.

Analysis

The investable question is not whether inference gets cheaper, but who captures the savings. Lower unit costs can expand usage, yet token-metered billing also makes customers more likely to cap, route, or shift workloads when value per token is unclear. That favors providers with distribution, workload data, and the ability to route between models; it pressures application vendors that absorb AI usage inside flat subscriptions. Cloud platforms and infrastructure suppliers may still benefit if cheaper inference drives enough volume to offset lower revenue per unit, so falling costs alone do not imply weaker infrastructure demand.

Over the next 1–3 months, earnings disclosures on AI-related revenue, inference cost per workload, and customer usage limits matter more than broad adoption claims. Over 6–18 months, the key risk is commoditization: model capability may diffuse faster than customers’ willingness to pay, leaving applications with incremental compute expense but limited pricing power. Conversely, lower costs could broaden use cases and lift aggregate workloads. The article offers no company-level economics or valuation evidence, so it is not a standalone directional signal.

Key Decisions for Investors

• No immediate broad AI-sector trade: treat this as a dispersion catalyst, not proof that AI investment returns are improving or deteriorating.
• Watch relative exposure between cloud/platform providers and seat-based software vendors. Prefer the former only where reported workload growth and monetization can be verified; avoid assuming cheaper tokens automatically translate into higher margins.
• Use upcoming earnings as the entry trigger: look for disclosed AI revenue or usage growth alongside stable/improving gross margins. Reassess if usage rises but margins weaken, or if customers report tighter AI budgets and capped workloads.
• Falsifiers for the cost-compression thesis include sustained growth in paid AI usage despite falling unit prices, or evidence that application vendors can pass usage costs through without slowing adoption. Track token pricing, usage limits, AI-related margin commentary, and cloud workload trends.