The Tokenmaxxing Hangover: Why Enterprises Are Rethinking AI Spending
Not long ago, a buzzword was dominating conversations in Silicon Valley boardrooms and Slack channels alike: tokenmaxxing. The idea was simple, almost intoxicating in its optimism — push AI usage as far as it will go, encourage every employee to use it constantly, and let productivity gains take care of themselves. CEOs became evangelists. Teams were urged to embed AI into every conceivable workflow. For a brief, heady moment, it felt like the future had arrived ahead of schedule.
Then the bills started coming in.
Reports emerged that Uber had burned through its entire annual AI budget in just a matter of months. Some companies began quietly cutting AI software licenses for certain departments. Meta reportedly shut down an internal AI usage leaderboard that had been encouraging competitive consumption of AI tools. Suddenly, the conversation shifted from "how much AI can we use?" to a far more uncomfortable question: "What are we actually getting for all of this?"
It is against this backdrop that Tiffany Luck, a partner at the prominent venture capital firm NEA, has been speaking candidly about where enterprises truly stand when it comes to measuring and understanding their return on AI investment.
The ROI Question No One Has Fully Answered Yet
For all the enthusiasm that has surrounded enterprise AI adoption over the past two years, Luck's perspective reflects a sobering but necessary reality check. Enterprises are still, in large part, figuring out how to define, track, and demonstrate AI ROI in a way that satisfies finance teams, justifies ongoing spend, and guides smarter future investment.
This is not a failure of AI as a technology. It is a failure of measurement frameworks to keep pace with rapid adoption. When companies moved fast to deploy AI tools across their organizations, many did so without establishing clear baselines, defining success metrics, or building the internal infrastructure needed to evaluate outcomes honestly. The result is that even companies spending millions on AI licenses and infrastructure often struggle to point to concrete, quantifiable value.
The tension here is real and growing. On one side, you have vendors, investors, and internal champions who believe — often correctly — that AI is transforming how work gets done. On the other side, you have CFOs and board members who want to see the numbers, and who are increasingly skeptical when those numbers are slow to materialize.
What Tokenmaxxing Got Wrong
The tokenmaxxing trend was, at its core, a volume-first strategy. The assumption was that more AI usage would naturally translate into more value. But volume and value are not the same thing, and enterprises are learning this distinction the hard way.
Encouraging employees to use AI as much as possible without a corresponding emphasis on use-case quality, workflow integration, and outcome tracking is a recipe for runaway costs. Tokens cost money. API calls cost money. And when usage is driven by novelty or executive mandates rather than genuine productivity need, those costs accumulate without producing proportional returns.
The companies that ran into budget overruns were not necessarily doing anything malicious or even particularly careless. They were operating in an environment where the cultural pressure to adopt AI was enormous and where the tools for measuring AI-generated value were still immature. That combination is dangerous for any budget.
How Smarter Enterprises Are Approaching AI Investment Now
The good news, according to investors like Luck who watch enterprise technology adoption closely, is that the market is maturing. The naive phase of tokenmaxxing is giving way to something more considered: a genuine effort to identify which AI use cases deliver measurable value and to concentrate investment there.
Several patterns are emerging among enterprises that are finding their footing with AI ROI.
- Use-case prioritization: Instead of deploying AI broadly and hoping for the best, leading organizations are identifying two or three high-impact workflows where AI assistance is proven to save time, reduce errors, or accelerate revenue, and doubling down on those before expanding.
- Baseline measurement: Companies are going back to basics, establishing clear before-and-after metrics for AI-assisted processes. This means measuring task completion times, error rates, output quality, and employee satisfaction with and without AI tools.
- License right-sizing: The move by some companies to cut AI licenses for certain teams is not necessarily a retreat from AI — it is a recognition that not every role benefits equally from AI assistance, and that spending on low-value seats is wasteful.
- Build vs. buy recalibration: Some enterprises are re-evaluating whether off-the-shelf AI tools serve their needs better than custom-built solutions, or vice versa, with ROI as the primary decision criterion.
What This Means for the AI Vendor Ecosystem
The enterprise recalibration has significant implications for AI vendors and the broader technology ecosystem. Vendors who built their growth projections on the tokenmaxxing wave — betting that usage would continue to surge indefinitely — are now facing harder conversations with customers who want to see value, not just usage statistics.
This is actually healthy for the industry. It creates pressure for AI companies to demonstrate genuine, measurable impact rather than simply capturing market share through aggressive pricing and promotional offers. The vendors that survive and thrive in this next phase will be those who invest in customer success, build robust analytics into their platforms, and help enterprise clients build the measurement frameworks they desperately need.
The Road Ahead for Enterprise AI
Tiffany Luck's observations reflect a broader inflection point in the enterprise AI story. The technology is real, the potential is enormous, and the long-term trajectory remains strongly positive. But the path from potential to proven ROI requires patience, discipline, and a willingness to make hard choices about where AI investment is actually warranted.
Enterprises that treat this moment as a reset rather than a retreat — using it to build smarter measurement practices, identify genuine use cases, and right-size their AI spend — will be far better positioned for the next wave of AI capability. Those that simply cut costs reactively without developing a more rigorous framework risk falling behind when the next generation of AI tools makes the value proposition even more compelling.
The AI revolution is not over. It is simply growing up. And for investors like Luck and the enterprises she works with, that maturation process, however uncomfortable, is exactly what the market needs.