The AI Overselling Problem Has Finally Hit the Enterprise Wall
Enterprise AI backlash describes the growing frustration among large organizations that invested heavily in AI systems, only to find that oversold promises, weak returns on investment, and fears about data and competitive advantage have created real AI adoption fatigue instead of the productivity revolution they were told to expect. Tech vendors didn’t misjudge the technology’s potential; they misread their customers’ patience. That is the core of today’s AI hype cycle: models are powerful, but the sales pitch treated them as magic. Palantir CEO Alex Karp openly says enterprise leaders feel they are “paying for tokens that create no value” and that these models have been “completely, irresponsibly, oversold”. When the people writing the cheques start talking like this on television, the market is not just cooling—it is correcting.
Karp’s Broadside: Tokens, Lost Alpha, and AI Adoption Fatigue
Alex Karp’s televised outburst is less a rant and more a signal flare for the enterprise AI backlash. In his words, something has “gone completely wrong” in how AI labs sell their products to large companies. Executives tell him they are wasting time with tokens while handing over their data and “alpha” — the hard-earned edge that defines their business. They fear that AI labs gain more from their inputs than the enterprises gain from the outputs. That mistrust is now visible in spending behavior: the recent “token backlash” has seen firms shift from tokenmaxxing to reining in usage and scrutinizing efficiency instead of blindly chasing AI-powered everything. Karp insists he is channeling “the voice of American business,” arguing that enterprises are “twice as livid” in private as he sounds on air. This is AI adoption fatigue in its purest form.
A Hyperscale Bet Colliding With Public Skepticism
While enterprises cool on AI overselling, infrastructure spending is exploding, creating a stark mismatch between capital and confidence. The four largest hyperscalers — Amazon, Google, Microsoft, and Meta — have guided their 2026 capital expenditures to roughly 715 billion, up more than 70% from the already-record 410 billion they spent in 2025. Datacenter construction spending has now eclipsed public transportation spending in the US, and datacenter construction jobs are outpacing office and home construction jobs combined. At the same time, tech leaders are noticing anti-AI sentiment — in some places, it shows up as anti-datacenter yard signs — and warning that it must be actively countered. When public opinion sours while infrastructure races ahead, you get an AI hype cycle where the physical bet on the future is huge but the social licence to operate is fragile.
The Nuclear Power Warning: Oversell Now, Pay for It for Decades
John Carmack’s analogy to nuclear power is a blunt warning: let fear and frustration set the narrative, and a transformative technology can be strangled “based on vibes” for generations. He notes how anti-nuclear efforts turned public opinion against a technology with enormous potential, leaving the US far behind on an energy source that could have reshaped policy and industry. He sees AI at a similar fork in the road. Today’s anti-AI sentiment may start with yard signs and local opposition to datacenters, but it is fed by something deeper: a credibility gap created by years of AI overselling to enterprises. Carmack argues that public opinion “shouldn’t be ceded unchallenged” and calls this moment “a transition more vibrant than the industrial revolution” with millions already seeing returns. If the industry keeps talking in miracles while delivering meh, it risks a nuclear-style reputational winter.
From Hype Cycle to Trust Cycle: What Vendors Must Fix
The lesson in all of this is uncomfortable for AI vendors: you cannot talk your way out of an AI overselling enterprise problem you talked yourself into. Enterprises now expect proof of value, clear data boundaries, and transparent economics instead of vague “transformational” claims. The token backlash shows they will cut spending when the math doesn’t work. At the same time, tech leaders warn that anti-AI sentiment is real and growing, and that public opinion around datacenters and AI infrastructure must be engaged directly, not dismissed. The path forward is not less ambition, but more honesty. Stop promising industrial-revolution-level change from a handful of prompts; start treating AI as a powerful but limited tool that earns trust through measurable outcomes. If the industry pivots from hype cycle to trust cycle now, it might avoid becoming the next nuclear power story.
