After the euphoria surrounding artificial intelligence (AI) and terms such as “AI vibe coding” marking 2025, the year 2026 is shaping up as a period of significant change in public and corporate perception of technology. This “climate change” in AI signals growing skepticism and a broader backlash beginning to manifest.
Despite this transformation, big technology companies, especially in Silicon Valley, continue to issue optimistic statements. Events like Microsoft’s Build and Google’s I/O conferences in May were filled with experts discussing “tokens,” the unit of measurement for interactions with AI, where each token corresponds to about three-quarters of a word.
However, these same conferences made ambitious but dubious statements about the future of AI. Demis Hassabis, CEO of DeepMind, stated at Google I/O that “Artificial General Intelligence is just a few years away,” while Mustafa Suleyman, CEO of Microsoft AI, spoke about building a “Humanistic Superintelligence.”
On Wall Street, although the market still shows interest, investor confidence is wavering. Shares of Nvidia, considered a bellwether for the AI sector, fell for a few days, recovered briefly after CEO Jensen Huang promised that AI agents would dominate the future, but fell again on Friday, June 6, 2026.
Even in the face of uncertainty, companies such as Anthropic, OpenAI and SpaceX continue to seek initial public offerings (IPOs) that value them in trillions of dollars. SpaceX’s plan, in particular, is largely based on the still untested idea of AI data centers operating in space.
Outside the technology sector’s cycle of enthusiasm, rejection of artificial intelligence has gained momentum. This movement is not limited to students booing pro-AI speakers at graduations, indicating a broader dissatisfaction.
Recent research corroborates this sentiment: a March Pew survey found that only 10% of Americans are optimistic about the future of AI. In the same month, an NBC survey showed that around 80% of registered voters in the US consider that both Democrats and Republicans are failing to manage issues linked to artificial intelligence.
This proportion of discontent is also reflected in the corporate environment. An April survey of white-collar workers showed that 80% refuse to use AI tools, even when required. In the last 30 days, 54% of professionals chose to ignore the company’s AI resources, carrying out tasks manually.
Full coverage: Latest News (EN)
These numbers suggest a level of dissatisfaction that borders on a widespread strike with AI across industries, extending beyond Silicon Valley and Wall Street. Protests against data centers, motivated by the opposition of 70% of Americans to these facilities close to their homes, are likely to grow, especially as they are already generating concrete results.
By 2025, at least 48 data center projects have been blocked or delayed, according to Data Center Watch, and resistance is intensifying. One example is the Stratos data center, planned for Utah, where local mobilization forced investor Kevin O’Leary, known from the Shark Tank program, to reduce the use of his land by 75%.
On Friday, O’Leary told local media, “We messed up. We made a lot of people angry,” acknowledging the impact of community opposition.
Political actions and the “let them eat tokens” metaphor in the AI landscape
The threat of electoral consequences may explain why politicians are starting to propose more serious measures. In the week in question, Senator Bernie Sanders called for the American public to own 50% of the shares of AI companies, former presidential candidate Andrew Yang suggested a tax on AI, and President Trump signed an executive order to regulate AI, a move that his AI czar, tycoon David Sacks, had long opposed.
On Friday, New York state lawmakers sent the governor a one-year moratorium on data centers. Additionally, Trump appeared to align with Sanders’ idea of the government acquiring a stake in OpenAI, which some critics of OpenAI’s current valuation have interpreted as a possible bailout.
The announcement of the White House executive order on AI came as Microsoft CEO Satya Nadella was making optimistic statements about the technology at the Build conference. This contrast creates a sense that we are witnessing two parallel worlds: opponents of AI and an alien technological regime that seems to be saying, essentially, “let them eat tokens.”
However, “revolution” is not the only path: beneath the surface of the optimistic rhetoric, the AI ecosystem itself shows signs of internal fragility, and the root of the problem lies in the costs associated with tokens.
The first signs of a setback in the artificial intelligence race in Silicon Valley
Uber is a notable example of a company that is firmly committed to artificial intelligence. The ride-hailing giant claims that 90% of its engineers use AI tools, most notably Anthropic’s Claude Code, and that up to 10% of its codebase is generated by AI agents.
Learn more: Record investments in AI redefine strategies of technology giants in the sector
Uber also encouraged “tokenmaxxing,” the maximizing use of AI tokens, a popular practice in Silicon Valley in 2025. However, the financial consequences of this approach began to emerge. “The budget that I thought I needed [for 2026] is already overrun”, declared CTO Neppalli Naga to the portal The Information on April 14, before the first four months of the year had even ended.
At the time, this information did not generate much impact on AI news. Only when Andrew MacDonald, COO of Uber, confirmed the meaning of this data at the end of May, in an interview with the Rapid Response podcast, did the issue gain relevance. MacDonald described Naga’s budget blowout as a “mind-blowing moment,” noting that such spending “becomes harder to justify because AI isn’t free… we’re going to have to start talking about token consumption.”
From that point on, the debate about token consumption intensified. Axios reported that an unidentified company spent half a billion dollars on tokens in a single month by not imposing usage limits on Claude licenses. It then emerged that Amazon and Meta had disabled their own internal AI booster panels, and other companies, such as Walmart and Starbucks, had scaled back their plans for AI agents.
In a leaked internal email, an Amazon senior vice president instructed employees to “stop using AI for the sake of using AI.” For many, this directive suggests a significant blow to OpenAI and Anthropic’s business model.
Both companies have dedicated years to developing models that, for the most part, consume more tokens. Currently, they promote agents capable of consuming tokens in even greater proportions, reaching 24 times more than a standard model.
Regardless of their high-level missions, the fundamental goal of both companies is token sales.
The reasons behind the end of the practice of “tokenmaxxing” in the artificial intelligence sector
Some AI industry leaders, sensing the change in direction, are beginning to speak out openly. Ravi Kumar S., CEO of Cognizant, an IT company specializing in AI, called “tokenmaxxing” a “vanity metric” at a Fortune conference. Kumar criticized OpenAI’s Sam Altman and Anthropic’s Dario Amodei, accusing them of “scaremongering.”
Altman and Amodei, with IPOs on the horizon, backed away from previous predictions about an AI-driven jobs apocalypse, which is already a shift in perception. However, what really hurts them is the fact that they are profiting from users’ lack of clarity about the complex costs of artificial intelligence.
Earlier this year, Anthropic quietly adjusted the price of Claude for many customers, charging by token. OpenAI, for its part, is considering ending its “unlimited” ChatGPT plans, a notable shift from Altman’s promise a year earlier of “intelligence too cheap to measure.”
This change is not restricted to the two AI giants. Microsoft began reducing its own token costs and increasing prices for all other users, even before the optimistic statements at the Build conference.
On the same topic: Apple stops advertising for Vision Pro in 2025 due to a significant drop in sales
In May, Microsoft began the process of revoking developers’ access to Claude Code, directing them to Microsoft Copilot. On June 1, Github Copilot users were transitioned from a fixed subscription model to a token subscription model.
Forums like Reddit have been flooded with complaints from users furious at the sudden increase in the price of their AI prompts. In one extreme case, a Claude user spent 50% of their monthly credits on a single prompt.
“At the beginning of the year,” Altman said on an OpenAI livestream this week, “people were completely happy with the amount they were spending… now, all of a sudden, [it’s] a big problem.” In an interview with CNBC on Monday, Altman admitted to “a lot of waste” in AI spending and mentioned that companies were asking, “how long do I have to wait for [the benefits of AI] to show up in revenue?”
Altman called this a “fair” question, but the only close answer he offered was: “The industry will resolve this very quickly…in a year or two.”
Could uncertainty about AI lead to market bubble collapse?
Still, how long OpenAI and Anthropic will have to resolve this issue largely depends on the success of their IPOs. “No one knows when this will all collapse, but 2026 will be remembered in retrospect as the year retail investors were left high and dry,” predicted Gary Marcus, a professor and a prominent critic of generative AI, on Monday.
Marcus, whose predictions about AI problems since 2022 have proven increasingly accurate, may not be entirely wrong here. He suggests, based on comments from Anthropic co-founder Daniela Amodei, that both companies spent so much money that they were “months away from bankruptcy” and had “no other options” other than pursuing billion-dollar IPOs.
OpenAI, in particular, has been reporting losses in excess of a billion dollars per month, a cost associated with providing ChatGPT for free to hundreds of millions of users.
Financial bubbles built around technologies inevitably end with an “emperor’s new clothes” moment. Eventually enough people point and laugh that the defenders can no longer sustain the uproar. That’s what happened with the internet bubble in 2000. Such an absurd business emerged – the world’s largest media empire being acquired by the creators of dial-up internet distributed on CDs that the markets couldn’t help but be incredulous. The climate has changed. Overvalued and unprofitable internet companies began to look naked, and a stock collapse quickly followed.
Times have changed, and the AI bubble is more resilient than its .com-era predecessor. It is supported by a single company that currently profits enormously from this race: Nvidia. Nvidia has been providing the “picks and shovels” to seekers of “AI gold” for so many years that it seemed invulnerable. However, even Nvidia is learning about the prohibitive and rising cost of artificial intelligence.
Follow: all about AI crisis
“The cost of computing is far beyond employee costs,” an Nvidia executive told Axios in April. So even Nvidia is vulnerable to “tokenmaxxing”. And that’s why the hottest trend in AI right now is hiring humans, which are becoming cheaper than AI and are necessary for quality control of artificial intelligence production.
Cognizant’s Kumar proudly highlighted that his AI company hired 20,000 graduates last year and plans more this year in a clear change of scenery.
Therefore, the expectation of a “jobs apocalypse” has changed. The perception about tokens has changed. And enthusiasm for building AI data centers has also wavered, not just due to public and environmental opposition, but the fact that there are not as many data centers under construction as expected, as revealed by journalist Ed Zitron’s investigative work.
What’s left? Possibly the only aspect that has not changed is the issue of AI “hallucinations”, as users are still unaware of the frequency with which most artificial intelligence models produce false information. Google, for example, doesn’t reveal Gemini 3.5 Flash’s hallucination rate, but an internal study from December found that Gemini can be accurate between 68.8% and 83.8% of the time.
And hallucinations aren’t hard to find these days. The “hallucination” that OpenAI, Anthropic, and SpaceX are genuine trillion-dollar AI giants, deserving of listing in major index funds despite being loss-making (breaking news: at the time of writing, the S&P 500 has officially opted out of this “hallucination”).
There is also the “hallucination” that Nvidia will always maintain its leadership position, even as companies that represent the majority of its business develop their own AI chips (which is why Michael Burry, of “The Big Short” fame, continues to short the company’s shares).
Another “hallucination” is the belief that customers want AI in everything, when repeated surveys indicate the opposite. And the “hallucination” that content generated by artificial intelligence will dominate the future, in a scenario where the generation that will lead this transformation mocks “AI trash”.
If these false perceptions disappear from the fevered minds of Silicon Valley and Wall Street, the great AI paradigm shift of 2026 will be complete.