Enterprises Hit the Brakes on AI Spending as ROI Remains Elusive

What happened

Earlier this year, Silicon Valley briefly elevated "tokenmaxxing" to a management priority. The idea was simple: chief executives urged employees to push generative AI usage as far as possible, consuming model tokens at maximum volume in hopes of discovering transformative productivity gains. For a few months, the strategy served as a proxy for digital transformation. Then the invoice arrived. NEA partner Tiffany Luck told TechCrunch that Uber reportedly exhausted its entire annual AI budget in just a few months, a speed that caught finance teams off guard. The sticker shock has produced visible retreats. Some companies have cut Claude licenses for segments of their workforce, and Meta eliminated its internal leaderboard that had gamified AI adoption among staff. Luck noted that despite the initial enthusiasm, enterprises are still trying to figure out whether their spending produces a real return on investment. The tension between exploratory adoption and fiscal discipline is now a central theme in enterprise AI strategy discussions.

Why it matters

The shift from uncapped experimentation to defensive budgeting suggests generative AI is entering a colder, more scrutinized phase inside large corporations. When a company of Uber's scale burns through a yearly technology allocation in a single quarter, it demonstrates how usage-based pricing for frontier models can scale costs faster than operational value. Trimming licenses and killing internal gamification tools like leaderboards are telltale signs that many deployments have been judged speculative rather than mission-critical. For vendors, the takeaway is that procurement departments are now insisting on clearer business cases before renewals or expansions. For employees, the freewheeling era of unlimited AI access is ending, replaced by finance-driven guardrails. The larger risk is that if early corporate adopters cannot show tangible efficiency gains or revenue contributions, the next cohort of enterprise buyers may postpone purchases until pricing models mature or product reliability improves. This dynamic could slow the near-term revenue curves of AI startups that have relied on seat-based expansion.

Public reaction

No strong public signal was available from Reddit or broader social discussion channels at the time of publication.

What to watch

Observers should track whether AI providers react to buyer pressure with pricing innovation. If runaway inference costs remain a pain point, vendors like Anthropic and OpenAI may face demands for consumption caps, flat-rate enterprise tiers, or even outcome-based contracts. Inside corporations, expect leaderboard gamification to give way to stricter governance: per-department inference budgets, mandatory use-case reviews, and ROI dashboards tied to workflow output rather than raw token volume. Upcoming quarterly earnings calls from technology-heavy employers will reveal whether AI spending is framed as a growth driver or a line item to optimize. CIO surveys and IT spending forecasts for the second half of the year will offer an early read on whether this caution is temporary or structural. Finally, monitor whether the venture narrative shifts from "adopt everywhere" to "adopt where it pays," as investors like Luck recalibrate what healthy AI traction looks like in their portfolios.

Sources

• TechCrunch: NEA's Tiffany Luck says enterprises are still figuring out their AI ROI