What smart people are saying about AI ROI crunch: 'Where's the revenue?'

The article highlights growing concern that AI spending is outpacing measurable productivity gains, with executives like Uber's Andrew Macdonald and Google’s Sundar Pichai saying CIO budgets are getting strained. Several industry voices warn that token-heavy AI usage may be inefficient or unsustainable, while others argue the lack of clear ROI is a temporary implementation issue. The main takeaway is a shift toward tighter AI budget scrutiny and a more skeptical corporate spending environment.

Analysis

The near-term read-through is not “AI demand is collapsing,” but that procurement is moving from novelty to budget scrutiny. That typically compresses spend growth at the low-ROI edge first: broad internal copilots, junior-dev assistance, and agentic workflows with weak attribution. The second-order effect is a winner-take-most dynamic among model vendors that can prove cost-per-task, not just benchmark quality; inference efficiency and admin tooling become the real monetization layer, while undifferentiated usage gets commoditized.

For software vendors, this is a gating issue for expansion revenue. If CIOs start enforcing token budgets and usage auditability, the losers are platforms selling open-ended seat expansion without measurable workflow displacement; the winners are those that can show hard savings in support, sales ops, and code generation. That matters for CRM more than the market is pricing: not because CRM is directly exposed to token bills, but because enterprise buyers under budget pressure delay add-ons and AI bundles unless payback is explicit.

The signal is mixed for NVDA. A spending backlash can slow the pace of incremental inference demand, but it also accelerates model optimization and on-prem/private deployment, which tends to favor efficient accelerators and software-defined stacks rather than pure capex exuberance. The bigger risk to NVDA is not immediate demand destruction; it is a narrative shift from “more tokens, more spend” to “same outcomes at lower cost,” which can cap multiple expansion over the next 1-2 quarters even if unit volumes keep rising.

GOOGL is relatively well positioned because cost discipline is becoming a product feature, not just a corporate headache. If enterprises are forced to rationalize spend, the vendors that can credibly market lower-cost, fast-response models should gain share and defend usage intensity. The contrarian view is that this backlash could actually extend the AI cycle: once waste is stripped out, usage per dollar rises and the surviving workloads become stickier, meaning the market may be underestimating how quickly spend migrates from experimental tools to durable production workloads.

Key Decisions for Investors

• Go long GOOGL vs. NVDA into the next 4-8 weeks: the market should reward the cheapest credible model stack while penalizing any narrative of AI demand fatigue; use a 1:1 pair and target 8-12% relative outperformance if enterprise buyers rotate toward cost-efficient inference.
• Trim or hedge UBER over the next 1-2 months: the stock is vulnerable to margin skepticism if management cannot connect AI spend to measurable unit economics; buy downside with 90-day puts or collar existing exposure around the next commentary cycle.
• Fade CRM AI-led re-rating until budget visibility improves: sell covered calls or reduce exposure for a 1-3 month window, since enterprise budget scrutiny can slow AI attach rates and defer incremental monetization.
• Hold NVDA but reduce gross exposure via call spreads rather than outright longs for 1-2 quarters: the secular thesis remains intact, but multiple expansion is at risk if the market shifts to ROI-focused procurement and prices in slower inference growth.
• Add selectively to private-market enablers of AI governance/observability rather than raw application names: the near-term beneficiaries are the tools that help CFOs measure token ROI, with a 6-12 month adoption window as enterprises impose spend controls.