After the AI binge, companies balk at soaring bills

AI usage costs are rising sharply as agent-based workflows consume far more tokens and compute, prompting companies to rethink adoption and spending. OpenAI and Anthropic are reportedly preparing to go public later this year, while some firms are shifting to cheaper open-source models to control expenses. The article highlights mounting compute shortages and signs that AI monetization is becoming more difficult as investor subsidies fade.

Analysis

The market is shifting from a land-grab to a unit-economics regime, and that is structurally bearish for the middleware layer that grew up assuming ever-cheaper inference. Once AI becomes a line item with a measurable ROI hurdle, usage will concentrate in workflows with hard payback, while “nice-to-have” copilots get quietly sunsetted or moved to lower-cost models. That transition should compress gross margins for pure-play AI vendors first, then expose which software incumbents were cross-subsidizing AI features to defend share.

The second-order winner is the open-source stack and the infrastructure that helps enterprises arbitrage model cost: orchestration, routing, caching, and prompt optimization. If customers move from premium frontier models to acceptable open-weight alternatives, the value pool shifts away from model branding toward deployment efficiency. That dynamic is also negative for the high-growth narrative around enterprise AI seats, because revenue may still grow while consumption per workflow falls, capping upside to top-line surprise.

META and UBER matter because they are early evidence that management teams are becoming more disciplined about AI ROI. For META, the key risk is not near-term expense blowout but future budget prioritization: if internal productivity gains are hard to prove, AI spend becomes a candidate for reallocation, which can slow the pace of experimental tools and vendor purchases. For UBER, the market should focus on whether AI is improving dispatch, support, and pricing enough to offset spend; if not, AI becomes a margin drag rather than a multiple-expanding narrative, and that can pressure the stock over the next 1-2 quarters.

The contrarian point is that this is not necessarily an AI demand collapse; it is a pricing normalization. If frontier model providers are forced to raise prices, the demand destruction may be smaller than feared because enterprises are already designing around cheaper models and more selective use cases. The bigger bear case is for valuations built on linear token growth, not for AI adoption itself.

Key Decisions for Investors

• Short META and UBER on rallies over the next 2-6 weeks; the near-term setup is weaker narrative support for AI spend discipline and limited evidence that incremental AI outlays are translating into visible productivity. Risk/reward favors a tactical short with a tight stop if management reiterates aggressive AI-driven margin expansion.
• Pair trade: long open-source enablers / infra efficiency names against short frontier-model exposure over 1-3 months. The trade is that cost compression will be captured by orchestration, routing, and deployment tooling rather than by premium API vendors; target 1.5-2.0x downside/upside skew if enterprise buyers shift mix faster than expected.
• Buy downside on high-multiple AI software names with obvious token-heavy usage models via 3-6 month put spreads. This is a good expression if revenue is still growing but margins are at risk from pricing pressure and customer optimization behavior.
• Favor vendors with direct cost savings use cases over general-purpose copilots in new capital deployment for the next 1-2 quarters. Use a quality screen: if the AI workflow cannot prove sub-12-month payback, avoid adding exposure or trim existing positions.
• Set a watchlist trigger on any model price increases from frontier providers over the next 90 days; that is the catalyst for a broader re-rating of AI spend assumptions and could benefit cost-optimized software names while pressuring usage-based revenue leaders.