Decades of software experience trained us on a specific economic model: build once, deploy everywhere, marginal costs near zero. That model drove every assumption about software adoption, from infrastructure planning to pricing to competitive strategy. It's the reason SaaS became the dominant business model of the last 20 years.
Frontier AI breaks this model at the foundation. The marginal cost of serving an additional user with a frontier model is not approaching zero - it's high and, for the most capable systems, stubbornly resistant to the efficiency improvements that made software so scalable. More users means more compute, and more compute means higher costs at every tier.
The implication for product managers: the AI features you're planning to ship are not governed by software economics. They're governed by something closer to manufacturing capacity. When a vendor tells you pricing will stay stable, they're making an assumption about compute costs that may not hold.
Most AI roadmaps treat capability access as a given. The more honest version acknowledges that frontier capabilities are rationed, not distributed. Your competitors who understand this are already planning around it. The ones who don't will find out the hard way when access tiers tighten and their product economics stop working.
Thoughts? Find me on Bluesky.