NEA investor: Uber burned through its annual AI budget in months as the market demands returns

Tokenmaxxing, the race by companies to use AI as much as possible with no regard for cost, has produced its first serious hangover. Uber spent its annual AI budget in a few months, Meta* shut down an internal leaderboard, and some companies cut back their Claude licenses. NEA venture investor Tiffany Luck calls it an ROI "hour of reckoning" and believes the market is simply growing up.

What is tokenmaxxing

In early 2025, a fashion spread among Silicon Valley CEOs: demand that employees use AI tools as widely as possible. The idea seemed intuitive: the more experiments, the higher the chance of finding a breakthrough use. Companies piled on subscriptions to GPT, Claude and Gemini without setting clear KPIs. Employees were encouraged to ask AI for everything, from code to strategy presentations. Meta* introduced an internal leaderboard tracking who "uses" AI more than anyone else. Corporate token budgets grew exponentially, while questions about return on investment were put off until later.

But "later" arrived sooner than expected.

When the bill came

Uber, according to sources, spent its entire annual AI budget in just the first few months. It is not an isolated case: a number of large companies have quietly cut corporate Claude licenses, first of all for departments that could not show measurable value. Teams that "used AI" but could not tie it to specific results lost access.

Meta* went even further: it shut down the very internal leaderboard it had launched to motivate employees. The metric of "who generates more queries" proved meaningless once the cost of each query became noticeable.

Tiffany Luck of NEA calls this a natural development: the market is entering a phase in which, for every dollar spent on AI, a company must be able to account with concrete figures: lower costs, faster processes or revenue growth.

IPOs and personal agents

Luck sees the current rethink not as a crisis but as normal maturing of the market. Companies that made it through the experimentation phase now understand where AI really delivers and where it was a showcase for shareholders. Among the themes she singles out as key:

• AI IPOs: the first public offerings of AI companies will take place amid heightened skepticism, with investors looking not for growth rate but for unit economics
• Personal agents: the next wave of value will come not from corporations but from individual users with agents configured for themselves
• Metric shift: "usage" as a KPI is on its way out, replaced by cost reduction, faster processes and revenue growth

For investors like NEA, this means a new filter: they increasingly look at how well an AI company can prove its product's ROI to the customer, rather than just show impressive reach numbers.

What it means

The tokenmaxxing era did not kill AI in corporations. It killed thoughtless AI use. Companies that came through this cycle and can show real results will come out stronger. The market has grown up, and that is good news for those building real products rather than chasing pretty dashboards.

*Meta has been designated an extremist organization and is banned in Russia.