If you told me two years ago that the demographic most hostile to AI would be Generation Z and younger, I would have thought you were out of your mind. Surely the younger generation, which has always been more technically adept, would want to use that to their advantage and embrace the technology, potentially making themselves more skilled, productive, and—above all—hirable.
But that’s not what’s happening. In May, students booed speakers at graduation ceremonies all over the country when they mentioned artificial intelligence. Google’s chairman, Eric Schmidt, got hit when he tried to suggest to graduates at the University of Arizona that AI has world-changing potential. Gloria Caulfield, VP of strategic alliances for the investment firm and real estate developer Tavistock, inspired boos in a commencement speech at the University of Central Florida when she compared the rise of AI with the Industrial Revolution. And students at Middle Tennessee State University shouted at Scott Borchetta, CEO of Big Machine Records, just for mentioning AI.
Data shows this goes beyond the anecdotal. A recent Gallup poll measuring AI adoption and attitudes among Gen Zers showed that those who are excited about AI have dropped significantly over the last year, from 36% to 22%. And anger toward AI is on the rise, from 22% to 31%. While other age groups are also skeptical of AI, the fact that the youngest generation of workers is undergoing such a sharp decline in attitude toward a new technology is without precedent.
Clearly, the AI industry has a massive PR problem, and it’s coming at a pivotal time. With anti-AI sentiment growing and midterm elections approaching, several politicians are now supporting efforts to halt or at least slow down the building of data centers—the facilities that fuel AI with the computing power it needs to function. If capacity slows to the point where it can’t keep up with demand, the rising cost of computing power will put serious limits on how far several industries, including the media, will be able to go with AI.
The infrastructure gap
In fact, that’s already starting to happen. Anyone who uses Claude regularly knows the pain of repeated outages; the service’s status dashboard shows an embarrassing amount of red over the past 90 days. The Claude Code revolution has meant demand for Anthropic’s AI has exploded in 2026. While the company is making moves to meet that demand (including signing a deal to buy computing power from Elon Musk’s SpaceX), it also stopped allowing builders to leverage their Claude subscriptions for third-party software. Those setups can sometimes use thousands of dollars of compute for the relatively cheap $200-a-month Claude Max subscription. They now have to use Anthropic’s platform or pay as they go.
That move angered a lot of people, but the greater point is that it’s forcing them to rethink what they’ve built. They may need to move to a cheaper model (perhaps one that’s open source), figure out a way to use the Claude platform, or shut it down altogether.
This kind of crossroads was probably inevitable—as demand for AI increases, free-compute loopholes will continue to close—but if computing power were plentiful and cheap, it wouldn’t be happening quite so fast. At least that’s the argument from the industry that’s been pushing for trillion-dollar infrastructure projects like OpenAI’s Stargate. For AI to deliver on its promises, compute needs to run like water from the heavens. That means more data centers, and more power plants to keep them running.
It’s no wonder, then, that environmental concerns are one of the chief reasons behind Zoomers’ anger over AI. I wrote months ago that AI’s massive energy use was becoming a serious PR problem, and it has only been exacerbated since then. And it’s not just an issue for politicians and podcasters to fight about: At the companies I advise about AI adoption, workers’ concerns over environmental impact consistently come up in internal surveys. It may even be affecting some decisions whether to use AI at all.
Making compute count
There is debate over whether the concerns about pollution and water use are overblown, but even if they are, the increasing cost of compute is real. The leaders at the forefront of AI transformation at their companies are past the point of simply giving every employee a ChatGPT account; they’re looking to help their workforce leap forward with agentic processes, automated workflows, and rapid prototyping through vibe coding. They may even be encouraging their engineers to get obsessed with “tokenmaxxing.”
We don’t know how the politics around data centers are going to play out, but one thing AI leaders can do now is get good at governance. This goes beyond just training their workforce on the tools and what the different models are good at (though that’s important). It means finding the right balance between experimentation and progress: Workers need the freedom to innovate their own workflows, but the organization also needs to ensure the compute it’s buying is put to good use. That doesn’t just mean “keeping costs down”—it’s knowing the cost might be high, but the outcome will be worth it.
Through my consulting work with media companies and PR agencies, I’ve seen how this can work in practice. One agency did a pilot with a vibe-coding tool: Usage exploded initially as workers experimented and pushed the limits of the platform. Inevitably, several different teams and individuals ended up building similar things, but because communication was high—they ran regular internal workshops and project reviews, learning from their people and providing direction along the way—they zeroed in on use cases that actually performed well, in this case speeding up and automating media intelligence. That agency ended up adopting an entirely different platform and sunsetting the initial tool based on what they found through the governance process.
There are a host of other examples, but the point is: To really make AI, and especially agents, work for your team, they’re going to need to be able to perform compute-heavy work. But the reality is computing power will likely stay expensive for a while. To avoid harsh limits and strict rules, which can quash innovation, it’s crucial that leaders define what success looks like, train their teams on how to use the tools and models, and build systems that encourage collaboration and catch waste.
That’s what good governance looks like. The political struggle over AI and data centers will rage on no matter what, but AI-forward companies can still chart a path ahead: one that works within the very real cost constraints, but ensures their workers feel them as little as possible.
