AI broke the economics of software. Vayu wants to be the first to tell a CFO which customers are losing them money

The article highlights an emerging AI-software profitability problem: heavy customer usage can raise vendors' token, compute and third-party model costs beyond the fixed contract revenue collected. The dynamic creates margin pressure for AI product vendors and may force changes to pricing, usage limits or contract structures.

Analysis

The key investable distinction is between AI vendors that retain inference economics and those that have effectively sold uncapped compute through legacy SaaS pricing. Near-term, the risk is greatest for application software companies promoting rapid AI adoption while disclosing neither usage caps nor AI-specific gross-margin impact; revenue can appear to accelerate while incremental gross profit deteriorates. This should pressure EV/revenue multiples over the next 1-3 quarters if management teams begin separating AI revenue from AI infrastructure expense or guide to lower gross margin.

Microsoft (MSFT), Alphabet (GOOGL), Amazon (AMZN), and Oracle (ORCL) are relatively advantaged because AI workload growth can be monetized both at the application layer and through captive cloud infrastructure. Application vendors dependent on external model providers face a structurally weaker negotiating position: model-price declines help, but lower costs are likely competed away unless contracts include consumption-based pricing, rate limits, or premium tiers. The second-order beneficiary is cloud infrastructure, since customer usage growth is still revenue-positive for hyperscalers even where it erodes an independent software vendor's margin.

Consensus may be too focused on whether AI drives bookings rather than whether those bookings convert into gross profit and free cash flow. The decisive 6-18 month catalyst is contract repricing: vendors that move from bundled AI features to credits, overages, and model-routing can recover margins; those unable to do so risk an adverse-selection problem in which their heaviest users are their least profitable customers. Falsification for the bearish application-software view would be broad evidence of stable or expanding non-GAAP gross margin alongside accelerating AI usage and unchanged sales incentives.

Key Decisions for Investors

• Maintain a 1-3 month quality tilt toward MSFT and GOOGL versus high-multiple application software: both can capture AI demand through cloud and software distribution, whereas externally sourced inference costs create greater margin uncertainty for the latter.
• Screen upcoming software earnings for explicit disclosures of AI revenue, inference/model expense, gross-margin impact, usage caps, and consumption overages. Treat absent disclosure alongside heavy AI-adoption marketing as a short-watch signal rather than an immediate short recommendation.
• Consider a 6-12 month pair of long IGV constituents with demonstrated consumption pricing and stable gross margins versus short an equal-weight basket of AI-feature-led SaaS names after earnings if guidance implies gross-margin dilution without offsetting price increases; size modestly because model-cost deflation could reverse the spread.
• Use cloud capex commentary from MSFT, AMZN, GOOGL, and ORCL as the near-term read-through. Rising AI infrastructure demand with weak software gross-margin commentary favors the infrastructure/application relative trade; evidence that application vendors are successfully charging overages would invalidate it.