McKinsey & Company has published an analysis arguing that software companies will not survive the shift to AI by bolting new features onto old business models — the models themselves, from pricing to packaging to cost structure, need an upgrade.

The piece lands at a moment when the economics of selling software are under visible strain. AI features raise compute costs per user, and many vendors still charge under seat-based subscriptions designed when marginal cost per customer was near zero. For product managers, that mismatch shows up directly in gross margin, the revenue left after cost of goods sold, and in unit economics that no longer pencil once inference costs scale with usage.

What pressure does AI put on software pricing?

McKinsey's core argument is that AI changes both sides of the software P&L at once. On the cost side, running AI features consumes expensive compute that grows with usage rather than with headcount. On the value side, customers increasingly expect AI capability inside products they already pay for, which limits how much of the new cost vendors can pass through as separate line items.

The traditional per-seat license compounds the problem. If an AI feature makes each user more productive, customers need fewer seats, and revenue falls precisely when the product delivers more value. Pricing tied to usage, outcomes, or hybrid structures is the direction McKinsey points vendors toward, though each carries its own friction: usage pricing is harder for customers to budget for, and outcome-based pricing is hard to define and measure credibly.

Where does the piece stop short?

The analysis is a directional argument rather than a playbook. It makes the case that standing still is the riskiest option, and that business-model design is now a product-management concern, not just a finance exercise. It does not prescribe a single pricing model, and any PM applying it Monday should treat the specific mechanics — what to meter, what to gate, what to charge for — as open design problems requiring their own experimentation and data.

The failure modes deserve equal attention. Repricing an installed base triggers churn and procurement fights. Underpricing AI features eats margin quietly. Overpricing them stalls adoption and invites competitors who rebuild around AI-native cost structures from day one, without legacy revenue to protect.

The takeaway is plain: in the AI era, the business model is part of the product, and product teams that treat packaging, pricing, and cost-to-serve as design inputs — tested and iterated like any feature — will adapt faster than those that leave the question to annual planning cycles.