Venture Capitalists Say the AI Spending Boom Is Hitting a Wall. The ROI Question Is Getting Louder.

The Bill Came Due Earlier this year, 'tokenmaxxing' was the hottest trend in Silicon Valley—CEOs pushing employees to use AI as aggressively as possible. According to TechCrunch's June 17 reporting on NEA partner Tiffany Luck, the experiment hit a wall fast. Uber reportedly burned through its entire annual AI budget in a matter of months. Some companies cut Anthropic's Claude licenses for parts of their organizations. Meta shut down an internal AI usage leaderboard. Luck now deals with this pattern constantly. Her frame: enterprises are still figuring out how to measure return on AI spend, and startups that help companies track that ROI are stepping into a genuine gap.

The Model Layer Is Already Commoditizing Chi-Hua

Chien, co-founder of Goodwater Capital and the Accel associate who identified Facebook as a six-person Harvard startup in the early 2000s, made a blunter point in a separate June 17 interview with TechCrunch. The commoditization of the model layer is already underway. The biggest winners of the AI era won't be companies selling AI. They'll be companies that use AI to deliver something people actually want. Chien's specific data point: the capability gap between frontier AI models and what you can run on a smartphone—once roughly a two-year lag—will shrink to three months within the next year. When the most powerful AI is essentially in your pocket, the companies charging premium prices for model access face a serious structural problem. The analysis here draws entirely from the two TechCrunch interviews.

VC Mechanics Are Getting Weird

Chien also addressed the fast-follow funding round phenomenon, where a firm invests a large amount at one valuation, then a smaller amount weeks later at a dramatically higher valuation, inflating the headline number. His assessment: this has been going on for a while, but it's accelerating. Rounds that used to be separated by 12 to 18 months now close three to six months apart. Valuations are being 'marketed very aggressively,' in Chien's words, as tools for attracting talent and blocking competition rather than as honest signals of company value. He also explained why venture capitalists are more willing to air grievances publicly now. As large VC firms have vertically integrated—meaning they have enough capital to lead and follow rounds themselves—the old incentive to preserve relationships with co-investors has weakened. Less interdependence means less decorum.

The Strongest Counter-Argument Fair pushback exists here

Enterprise AI adoption following a bumpy spending curve is not evidence that AI lacks value. It may just mean the tooling and workflows needed to capture that value are still being built. Every major technology shift, from cloud computing to mobile, went through a phase where spending outpaced measurable returns before the productivity gains materialized. Luck herself, per TechCrunch, remains optimistic specifically about 'magic moments' in consumer AI applications and personal agents. The argument isn't that AI is a bubble. It's that the current measurement infrastructure isn't good enough to prove the value enterprises are paying for. That said, 'we haven't figured out how to measure it yet' is a harder sell when annual AI budgets are being exhausted in four months.

What This Means for the Market

The pattern Luck and Chien are describing has a concrete structural implication. If the model layer commoditizes as Chien expects, the valuation logic for companies like OpenAI and Anthropic—which are priced on premium model access—faces real pressure. Chien's view, stated plainly, is that the companies that matter in ten years will be the ones that embedded AI deeply enough into a product that people couldn't imagine using anything else. Not the ones that built the underlying model. Luck's focus on startups helping enterprises track AI ROI points to a specific near-term opportunity: companies that make the value of AI spending legible are solving a problem that is genuinely unsolved as of June 17, 2026. The open question neither investor fully answered: if measuring AI ROI is this difficult for sophisticated enterprises with dedicated engineering teams, what happens to the smaller businesses that bought into the tooling wave without the infrastructure to audit what they're getting? No named source in these interviews addressed that segment directly.