NEA’s Tiffany Luck says enterprises are still figuring out their AI ROI

Companies that aggressively pushed AI usage are now facing cost overruns, with Uber reportedly exhausting its annual AI budget in just a few months. The article says some firms have cut Claude licenses for parts of their organizations and Meta has killed its internal leaderboard, signaling a shift from AI exuberance to tighter spending controls. The piece is broadly cautionary for enterprise AI adoption and budget discipline, but it does not describe a single company-specific financial shock.

Analysis

The first-order read is that AI usage is moving from a free-growth KPI to a controllable cost center, which changes bargaining power inside the ecosystem. The largest beneficiaries are the model vendors with enterprise-grade governance and usage controls, while the losers are companies that used “AI adoption” as an undisciplined proxy for productivity and are now discovering margin leakage in engineering, support, and ops. This should pressure smaller app-layer vendors that monetize on seat expansion without clear ROI, while advantaging infrastructure players that can prove unit economics and enforce spend caps.

For UBER specifically, the risk is not the absolute dollar amount of AI spend, but the possibility that management tightens experimentation just as automation payoffs start compounding. If AI was accelerating dispatch optimization, fraud detection, or support deflection, a pullback can create a 1–2 quarter lag in visible margin improvement, especially if the budgeting reset occurs during a period of heavier product investment. META is less exposed operationally because it can internalize tooling, but the governance signal matters: once internal scoreboards disappear, the market tends to infer that the marginal productivity gains are less defensible, which can compress confidence in future opex efficiency.

The second-order winners are governance software, cloud cost-management tools, and model-routing layers that help firms substitute cheaper models for premium ones without sacrificing output. The market is likely underpricing a near-term “AI spend rationalization” phase over the next 1–3 quarters, followed by a second wave of demand focused on observability, workflow control, and compliance. If board-level scrutiny turns from “use more AI” to “show payback,” procurement budgets should rotate away from raw model tokens toward tools that measure and optimize them.

The contrarian view is that the selloff risk is probably overdone for high-quality operators, because spend discipline usually improves ROI and extends the life of AI adoption rather than killing it. In other words, the headline is not a demand destruction story; it is a margin-discipline story. For the best-capitalized platforms, tighter controls can actually increase long-run adoption by converting novelty usage into repeatable workflow automation.

Key Decisions for Investors

• Short a basket of high-valuation AI application names with weak ROI disclosure for 1-3 months; pair against META on the view that internal discipline will be a relative positive for scale players while the basket faces budget resets and multiple compression.
• For UBER, wait for the next quarterly print or guidance update and look for a dip-buy only if management explicitly links AI spend to measurable margin gains; otherwise consider a small tactical short dated 4-8 weeks around risk of opex scrutiny and slower efficiency realization.
• Long governance/cost-optimization beneficiaries via a basket trade in cloud spend management and workflow observability names for 3-6 months; upside is a rerating as enterprises shift from experimentation to control, with downside limited if AI budgets simply pause rather than contract.
• Use META as a relative long versus UBER on a 2-4 quarter horizon: META’s scale and internalization should let it absorb AI governance tighter than consumer-platform operators, making it a cleaner beneficiary of disciplined AI deployment.