The AI climate shift is real: why the backlash against Artificial Intelligence is growing in 2026
In March 2026, a Pew Research survey revealed that only 10% of Americans say they are excited about the future of Artificial Intelligence. That number alone would be enough to make headlines. But it is just the tip of the iceberg of something much bigger happening right now, far from the bright stages of Silicon Valley.
While CEOs of major tech companies take the stage at conferences to talk about superintelligence and AGI as if they were inevitabilities, the world out there is sending a very different message. Workers refusing to use AI even when required. Communities blocking the construction of data centers. Politicians proposing taxation and even public ownership stakes in AI companies. And inside the tech companies themselves, the numbers are starting to not add up.
If 2025 was the year of vibe coding — when folks embraced the idea of letting AI write code while you just gave it direction — 2026 is quickly becoming the year of the vibe shift. The turn is not just cultural. It is financial. It is political. And it is coming from multiple directions at once, including from inside the very tech bubble that bet so heavily on AI as the solution to everything.
Tokenmaxxing blew up budgets, protests are growing, and the backlash that many dismissed as overblown is increasingly looking like the start of a real cycle shift. 👇
Meanwhile, at the conferences: always sunny in Silicon Valley
Looking only at the big industry events, you might think everything is going swimmingly. Microsoft’s Build 2026 and Google I/O in May were packed with upbeat presentations where executives talked nonstop about tokens — the basic unit by which AI prompts and responses are measured, with one token equaling roughly three-quarters of a word on average.
The statements made at these events bordered on the grandiose. DeepMind CEO Demis Hassabis declared during Google I/O that Artificial General Intelligence was just a few years away and that we were standing at the foothills of the Singularity. Microsoft AI CEO Mustafa Suleyman assured everyone that the scaling laws still hold and that the company is building something called Humanistic Superintelligence.
Pretty words. But while those speeches were drawing applause in corporate auditoriums, Wall Street was already showing signs of hesitation. Nvidia shares, considered the main barometer of the AI market, fluctuated significantly throughout the week. They rose after CEO Jensen Huang insisted that AI agents will run everything everywhere in the future — presumably after they stop deleting databases and causing 30-hour service outages — and dropped again on Friday.
On the other side, companies like Anthropic, OpenAI, and even SpaceX continue chasing IPOs in the trillion-dollar range. In SpaceX’s case, part of that valuation rests on the still-untested concept of AI data centers in space. It is a scenario that mixes legitimate ambition with generous doses of speculation.
The people have spoken: the numbers don’t lie
Outside the big tech optimism bubble, the sentiment is radically different. That same Pew Research survey from March painted a discouraging picture for anyone selling the future of AI as both inevitable and desirable. The data is clear and consistent: the majority of Americans expressed concern about AI’s impact on the job market, the privacy of personal data, and the concentration of power in the hands of a few companies.
An NBC poll published in the same month showed that roughly 80% of registered voters in the United States believe that neither Democrats nor Republicans are doing a good job when it comes to AI. That is a level of bipartisan dissatisfaction rarely seen.
And in the corporate world, resistance is equally striking. An April survey of office workers revealed that 80% of them are refusing to use AI tools, even when use is mandatory. In the 30 days before the survey, 54% of workers reported that they simply ignored their company’s AI tools and completed their tasks themselves. We are talking about levels of corporate civil disobedience that echo general strikes — something that would have been unthinkable in relation to a technology that, at least in official messaging, was supposed to make everyone’s life easier.
Receive the best innovation content in your email.
All the news, tips, trends, and resources you're looking for, delivered to your inbox.
By subscribing to the newsletter, you agree to receive communications from Método Viral. We are committed to always protecting and respecting your privacy.
The skepticism is not uniform. It varies by age group, education level, and professional field. This suggests that the adoption problem is also one of communication and trust, going well beyond a simple learning curve. Reports from specialized consulting firms published throughout the first quarter of 2026 reinforce this trend, noting that the return on investment reported by companies that implemented generative AI consistently fell below expectations in sectors like legal, healthcare, and financial services.
What is tokenmaxxing and why it is at the center of the problem
To understand the scale of the financial hole that Artificial Intelligence is creating inside companies, you need to know the concept of tokenmaxxing. Put simply: tokenmaxxing is the practice of consuming as many tokens as possible — the billing unit of AI models — in every interaction, whether by sending more context, maintaining long conversation histories, or feeding models entire documents before each response.
In 2025, this was cool. Companies created internal leaderboards encouraging their engineers to use more and more tokens. The logic was straightforward: the more context, the better the output. The problem is that nobody properly calculated the budget impact of this when multiplied by thousands of simultaneous users over months on end.
The Uber case is a textbook example. The ride-hailing company proudly stated that 90% of its engineers were using AI tools, primarily Anthropic’s Claude Code. About 10% of Uber’s codebase was written by AI agents. The company even had leaderboards to incentivize maximum token consumption.
Then the bill arrived. In April 2026, Uber CTO Neppalli Naga revealed to The Information that the budget he had projected for the entire year had already been blown — and the year had barely passed the four-month mark. By late May, COO Andrew MacDonald confirmed the impact on a podcast, calling the blown budget a head-exploding moment. The conclusion was unavoidable: that level of spending becomes hard to justify because AI is not free.
Uber’s case was not isolated. Axios reported that an unidentified company had burned through half a billion dollars in tokens in a single month after failing to put usage limits on Claude licenses. Amazon and Meta shut down their internal AI leaderboards. Walmart and Starbucks scaled back their AI agent plans. In a leaked email, a senior vice president at Amazon asked employees to stop using AI just for the sake of using it.
Let them eat tokens: protests in the streets and in politics
Beyond the numbers that don’t add up inside corporations, the backlash against Artificial Intelligence is also taking concrete shape in the streets and in legislatures. Protests against data center construction spread across the United States throughout 2025 and gained even more momentum in 2026. A Gallup poll showed that 70% of Americans say they don’t want data centers near where they live.
The complaints are varied but quite concrete: absurd electricity consumption, excessive water use for server cooling, local environmental impact, and the feeling that these facilities are imposed on communities without any consultation or direct benefit. This is not a fringe movement — it is organized resistance with tangible results.
According to Data Center Watch, at least 48 data center projects were blocked or delayed in 2025. And the fight is getting even more intense. The case of the Stratos data center, planned in Utah by Shark Tank investor Kevin OLeary, is a good illustration. Local opposition forced OLeary to reduce land use by 75%. On Friday, he admitted to local TV that they had messed up and made a lot of people angry.
On the political front, the week was particularly eventful. Senator Bernie Sanders argued that the American public should hold a 50% stake in AI companies. Former presidential candidate Andrew Yang proposed a specific tax on AI. And President Trump finally signed an executive order on AI regulation — something his own AI czar, David Sacks, had been publicly opposing.
In New York State, lawmakers sent a one-year data center moratorium to the governor’s desk. And Trump appeared to align with Sanders’ idea of the government taking an equity stake in OpenAI, a move that some critics saw as a bailout in disguise.
The White House executive order was announced right as Microsoft CEO Satya Nadella was making optimistic statements about AI at Build. The feeling of watching a tale of two worlds — the anti-AI crowd on one side and an AI regime disconnected from reality essentially saying let them eat tokens on the other — was hard to ignore.
The token bill is cracking Silicon Valley from the inside
But hold off on the revolution: even beneath the polished surface of the conferences, the AI regime is showing cracks on its own. And it all revolves around tokens.
Some AI leaders, sensing which way the wind is blowing, started saying out loud what a lot of people were already thinking. Ravi Kumar S., CEO of Cognizant, called tokenmaxxing a vanity metric during a Fortune conference. Kumar went further and took direct aim at Sam Altman of OpenAI and Dario Amodei of Anthropic, accusing both of fearmongering with apocalyptic predictions about jobs.
The irony is that both Altman and Amodei have walked back their own predictions of an AI-driven jobs apocalypse — conveniently now that their respective companies are preparing billion-dollar IPOs. But what is really hurting both CEOs is that they are profiting from user confusion about the complex costs of AI.
In early 2026, Anthropic quietly changed Claude’s pricing for many customers, switching to per-token billing. OpenAI is considering eliminating its unlimited ChatGPT plans — a drastic change compared to a year ago, when Altman was promising intelligence too cheap to meter. Microsoft started cutting token costs for itself while raising prices for everyone else. Developers had their Claude Code access revoked and were pushed toward Microsoft Copilot. On June 1, GitHub Copilot users were migrated from a flat subscription to a per-token billing model.
Reddit filled up with furious users documenting how their AI prompts had suddenly become expensive. In one extreme case, a Claude user burned through 50% of their monthly credits on a single prompt. Sam Altman himself admitted during an OpenAI livestream that at the beginning of the year people were totally happy with what they were spending, and that now, all of a sudden, it had become a huge problem. In an interview with CNBC, Altman acknowledged there is a ton of waste in AI spending and that companies were asking how long they would have to wait for the benefits to show up in revenue. His most honest answer? The industry will figure it out fast, in another year or two. 🤷
Is the bubble going to pop? The dot-com ghost haunts the market
How long OpenAI and Anthropic have to solve the return-on-investment question depends largely on what happens with their IPOs. AI generative critic and professor Gary Marcus made a bold prediction: nobody knows when this whole thing will come crashing down, but 2026 will be remembered in hindsight as the year retail investors were left holding the bag.
Marcus, who has been frequently right in his warnings about AI problems since 2022, could be wrong on this specific prediction. But his hunch is based on comments from Daniela Amodei, co-founder of Anthropic, suggesting that both companies have burned through so much cash that they could be months from bankruptcy with no option other than pursuing trillion-dollar IPOs. OpenAI, in particular, has been losing more than a billion dollars a month — the cost of offering ChatGPT for free to hundreds of millions of people.
Financial bubbles built around technologies inevitably end with an emperor-has-no-clothes moment — when enough people point and laugh, and the courtiers can no longer sustain the hype. That is exactly how the dot-com bubble burst in 2000, when a deal so absurd on its face — the world’s largest media empire being bought by the company handing out dial-up internet on CDs — made the entire market stop and question what was going on. The climate shifted, overvalued and unprofitable dot-com companies were exposed, and the stock collapse followed shortly after.
Humans are becoming cheaper than AI
Times have changed, and the AI bubble is more robust than its dot-com predecessor. It is built on top of at least one company that is actually making money from all of this — Nvidia, which sold the picks and shovels of the AI gold rush for years and seemed invulnerable for a good while.
But even Nvidia is learning lessons about the prohibitive and growing cost of AI. A company executive admitted to Axios in April that the cost of computation already far exceeds the cost of employees. That means even Nvidia is vulnerable to the tokenmaxxing effect.
And that is why the hottest thing in the AI world these days is, believe it or not, hiring humans. Because they are becoming cheaper than AI and are still needed for quality control of what AI produces. Cognizant CEO Ravi Kumar bragged that his AI company hired 20,000 recent graduates last year and plans to hire even more this year. If that is not a vibe shift, nothing is. 🧑💻
The hallucinations nobody is talking about
The climate shift has hit the job market, token budgets, and even data center construction — which has fallen below expectations, as satellite image analyses of planned sites reveal little to no actual building activity.
But there is one vibe that still has not changed: hallucinations. Most users still have no idea how often AI models make up information. Google, for example, refuses to disclose how frequently Gemini 3.5 Flash hallucinates, but an internal study published by the company itself in December found that Gemini may be accurate only between 68.8% and 83.8% of the time. That means in up to a third of interactions, the model could be producing incorrect information.
And the hallucinations go beyond chatbots. There is the collective hallucination that OpenAI, Anthropic, and SpaceX are genuine trillion-dollar giants that deserve to be in benchmark indexes like the S&P 500, despite being unprofitable companies. While this article was being written, the S&P 500 officially decided not to include SpaceX — a real-time reality check.
There is the hallucination that Nvidia will stay on top forever, even as companies that represent the majority of its revenue are developing their own AI chips — which is why Michael Burry, the investor who became famous for betting against the housing market before the 2008 crisis, continues to short Nvidia stock.
There is the hallucination that consumers want AI in absolutely everything, when survey after survey shows exactly the opposite. And the hallucination that AI-generated content will dominate the future, when the youngest generation — the very one that will shape that future — looks at so-called AI slop with contempt and laughter.
What this cycle reveals about the future of Artificial Intelligence
Hype and correction cycles are nothing new in the history of technology. What makes this moment different is the speed at which the backlash is organizing and the diversity of fronts on which it is happening simultaneously. It is not just the financial market revising valuations. It is not just a specific group of workers worried about jobs. It is a convergence of discontent involving local communities, lawmakers, professionals across different sectors, and even a significant portion of the very engineers and researchers who build these technologies.
When the questioning comes from inside and outside at the same time, it tends to produce more lasting changes than the previous resistance movements that the tech industry was able to absorb and neutralize with relative ease.
Tokenmaxxing as a metaphor goes beyond API costs. It represents a broader tendency in the AI industry to maximize the consumption of resources — computational, energy, data, human attention — without establishing clear limits or sustainability criteria. When that maximization logic meets the organized resistance of communities, workers, and regulators, the collision is inevitable.
What is at stake is not whether Artificial Intelligence will keep evolving — it will. What is being contested now is who defines the terms of that evolution, who benefits from it, and who pays the costs. Both OpenAI and Anthropic spent years building models that, for the most part, consume ever more tokens. Now they are promoting agents that can consume tokens at a scale up to 24 times greater than a standard model. However lofty their missions may be, both companies are in the business of selling tokens.
The vibe shift of 2026 may be the beginning of a more mature and honest relationship between society and AI — one where promises need to be verifiable, costs shared more fairly, and the data feeding these systems treated with more respect and transparency. If these collective hallucinations clear from the feverish minds of Silicon Valley and Wall Street, the great AI climate shift of 2026 will be complete. 🧩
