AI Vibe Shift: Understanding the Growing Backlash

AI vibe coding has emerged as a prominent topic, having been named one of the phrases of the year for 2025. As we approach 2026, a notable transformation in AI perceptions is taking shape.

This shift may not be evident from the latest declarations coming from the tech industry, which often maintains a sunny disposition. At Microsoft’s Build conference, similar to Google’s I/O event in May, numerous tech enthusiasts discussed tokens, the units that quantify AI prompts and responses (a token is roughly equivalent to three-quarters of a word).

Both events featured claims about advanced AI that lack credibility. DeepMind’s CEO, Demis Hassabis, asserted at Google I/O that “Artificial General Intelligence is just a few years away… we are standing in the foothills of the Singularity.” Meanwhile, Microsoft AI CEO Mustafa Suleyman claimed, “scaling laws are holding… we are building towards what we call Humanist Superintelligence.”

While Wall Street seemed to accept these declarations, investor confidence was beginning to waver. Nvidia stock, often seen as a barometer for AI, experienced volatility; it fell briefly before rallying after CEO Jensen Huang proclaimed that AI agents would dominate all sectors in the future, only to decline again later.

Currently, companies like Anthropic, OpenAI, and SpaceX are in pursuit of trillion-dollar IPOs, particularly based on the unproven idea of AI data centers in space.

Nonetheless, outside the confines of the AI bubble, a growing backlash has been evident, not just from students expressing dissatisfaction with pro-AI commencement speakers.

According to a Pew poll conducted in March, only 10 percent of Americans are enthusiastic about the future of AI. Around the same time, an NBC poll indicated that 80 percent of registered U.S. voters felt that neither Democrats nor Republicans were effectively addressing AI-related issues.

This sentiment is mirrored in an April survey of white-collar workers, where 80 percent reported refusing to use AI tools even when required to do so. In the past month alone, 54 percent of employees admitted to bypassing company AI systems to complete tasks independently.

These statistics reflect a significant level of discontent with AI across various industries, indicating a potential for widespread strike-like resistance beyond the borders of Silicon Valley and Wall Street.

Protests against data centers are likely to intensify, especially as 70 percent of Americans express opposition to having data centers located near them. Recent developments have shown that these protests can yield tangible outcomes.

In 2025, at least 48 data center projects faced delays or cancellations, as reported by Data Center Watch, with opposition growing fiercer. Take the Stratos data center project in Utah, where local dissent forced investor Kevin O’Leary to reduce his land usage by 75 percent.

O’Leary candidly admitted, “We screwed up,” during an interview with local news, acknowledging the backlash.

Are Politicians Responding to AI Concerns?

The growing threat of electoral backlash may explain why politicians are beginning to propose meaningful regulations.

This week, Senator Bernie Sanders advocated for U.S. public ownership of a 50 percent stake in AI firms, former presidential candidate Andrew Yang suggested implementing an AI tax, and President Trump signed an executive order on AI regulation, which his AI advisor, Silicon Valley leader David Sacks, had previously opposed.

New York State legislators have forwarded a one-year moratorium on data centers to the governor. Trump appears to be aligning with Sanders’ vision for government investment in OpenAI, with some skeptics interpreting it as a potential bailout.

The announcement of the White House’s AI executive order coincided with Microsoft CEO Satya Nadella’s optimistic remarks on AI at the Build conference, highlighting a stark contrast between the pro-AI rhetoric and public sentiment, which suggests an attitude of “let them eat tokens.”

Beneath the surface, the AI landscape is showing signs of instability—primarily driven by the realities of token consumption.

How Is Silicon Valley Reacting to AI Setbacks?

Among companies that fervently believe in AI, Uber stands out. The rideshare company reports that 90 percent of its engineers utilize AI tools, primarily Anthropic’s Claude Code, with AI agents contributing to around 10 percent of its codebase.

Uber also implemented leaderboards to encourage extensive use of AI tokens, a practice referred to as tokenmaxxing, which gained popularity in 2025.

The consequences of tokenmaxxing soon became apparent. “The budget I initially anticipated [for 2026] has already been exceeded,” CTO Neppalli Naga shared with The Information on April 14, just months into the year.

The impact of this revelation did not resonate significantly within the AI news cycle until Uber’s COO confirmed its implications at the end of May. Naga’s budgetary issues were described as a “head-exploding moment” by Andrew MacDonald on the Rapid Response podcast. Such expenditures “become increasingly difficult to justify as AI is not free… we need to initiate discussions about token consumption.”

Suddenly, discussions about token consumption became prevalent. Axios reported that an unnamed company had consumed half a billion dollars’ worth of tokens within a month due to a lack of usage limits on Claude licenses.

Following this, Amazon and Meta decided to dismantle their internal AI leaderboards, while companies like Walmart and Starbucks scaled back their AI implementation plans. In a leaked communication, an Amazon senior vice president instructed employees to “stop using AI just for the sake of using AI.” This development raises questions about the sustainability of OpenAI and Anthropic’s business models.

Both organizations have invested years in developing models that, for the most part, consume more tokens, while now promoting agents that can consume tokens at rates up to 24 times higher than standard models.

Despite their lofty ambitions, both companies are ultimately driven by the need to sell tokens.

What Caused the Downfall of Tokenmaxxing?

Some AI leaders, recognizing the changing landscape, are beginning to express these sentiments publicly. Ravi Kumar S., CEO of AI IT company Cognizant, labeled tokenmaxxing as “a vanity metric” during a Fortune conference. Kumar directed his criticism at OpenAI’s Sam Altman and Anthropic’s Dario Amodei, accusing them of “fearmongering.”

Altman and Amodei have since moderated their previous doomsday forecasts regarding AI job loss, likely influenced by their impending IPOs—a shift in sentiment that may be driving the current vibe change. their recent strategies have also capitalized on user misunderstandings surrounding AI costs.

Earlier this year, Anthropic discreetly adjusted the pricing of Claude for many clients, implementing charges based on token usage. OpenAI is contemplating the discontinuation of its “unlimited” ChatGPT plans, a significant departure from a year ago when Altman claimed, “intelligence too cheap to meter” was on the horizon.

This transformation is not limited to these two AI leaders. Microsoft has begun lowering its own token expenses while simultaneously increasing prices for external users, even prior to the optimistic statements made at Build.

Microsoft began revoking developer access to Claude Code, redirecting them to Microsoft Copilot in May. By June 1, GitHub Copilot users transitioned from a fixed subscription model to a per-token payment structure.

Frustrated users flooded Reddit, voicing their discontent over the sudden rise in costs associated with AI prompts. In one notable instance, a user of Claude reported using 50 percent of his monthly credits on a single prompt.

During a recent OpenAI livestream, Altman acknowledged the shift in user sentiment, stating, “At the beginning of the year, people were completely satisfied with their spending… but now, all of a sudden, it’s a significant issue.” In a CNBC interview, Altman recognized the “ton of waste” in AI expenditures and noted that companies were questioning, “how long do I have to wait for [AI benefits] to translate into revenue?”

Altman admitted this was a “fair issue,” and when asked about the resolution timeline, he suggested, “The industry will figure that out pretty quickly… in another year or two.”

Could the Vibe Shift Signal the End of the AI Bubble?

The timeline for OpenAI and Anthropic to resolve these challenges largely hinges on the outcomes of their IPOs.

As Gary Marcus, a professor and noted critic of generative AI, remarked, “Nobody knows when this will all collapse, but 2026 will likely be remembered as the year when retail investors were left holding the bag.”

While Marcus has accurately predicted AI-related issues since 2022, he may still be mistaken. he has a strong intuition, influenced by comments from Anthropic co-founder Daniela Amodei, that both companies have incurred massive expenditures, leaving them “months from bankruptcy” and with “no options” but to pursue trillion-dollar IPOs.

OpenAI, in particular, has been operating at a loss exceeding one billion dollars per month due to the cost of providing free access to ChatGPT for millions of users.

This Tweet is currently unavailable. It might be loading or has been removed.

Financial bubbles associated with emerging technologies often conclude with an Emperor’s New Clothes scenario, where enough observers can no longer ignore the glaring inconsistencies in the hype.

This phenomenon was evident at the end of the dotcom bubble in 2000, when a business transaction surfaced that seemed absurd on its face (the largest media conglomerate acquiring the company that distributed dial-up internet through CDs)—prompting widespread ridicule and a shift in market sentiment. Overhyped, unprofitable dotcom enterprises were suddenly exposed, leading to a stock market crash.

Are Human Workers Outpacing AI in Value?

Today, the AI bubble is more resilient than its dotcom counterpart, built on the success of NVIDIA, a company that has profited from selling essential technology to AI developers for years. even NVIDIA is grappling with the escalating costs associated with AI.

“The expenses of computing far exceed the costs associated with employees,” one NVIDIA executive told Axios in April. This reality underscores that even NVIDIA is susceptible to the ramifications of tokenmaxxing. the most promising trend in AI is the hiring of human workers, as they are becoming more cost-effective than AI and are essential for ensuring quality in AI outputs.

Cognizant’s Kumar proudly shared that his company recruited 20,000 graduates last year and plans to hire more this year—a notable shift in the industry.

The narrative surrounding job loss, token consumption, and data center construction has evolved significantly. Public and environmental opposition has intensified, and the anticipated number of new data centers is not materializing as initially projected. Journalist Ed Zitron has conducted extensive investigations, analyzing satellite imagery for signs of construction at proposed data center sites.

What remains consistent is that awareness of AI hallucinations persists; users often underestimate how frequently many AI models generate inaccurate outputs. For instance, Google has not disclosed the frequency of hallucinations in its Gemini 3.5 Flash model, although a December study indicated that Gemini’s accuracy may only range from 68.8 to 83.8 percent.

Hallucinations are increasingly prevalent. The notion that OpenAI, Anthropic, and SpaceX are legitimate trillion-dollar AI enterprises worthy of inclusion in top index funds, despite their lack of profitability, is one such hallucination (notably, as I write this, the S&P 500 has officially distanced itself from this illusion).

Another misconception is the belief that NVIDIA will always maintain its dominance, even as many of its clients develop their own AI chips—a situation highlighted by Michael Burry, famously known for his shorting of stocks.

The assumption that customers desire AI integration in all products contradicts numerous surveys indicating the opposite sentiment. the expectation that AI-generated content will dominate the future is met with skepticism from the generation poised to influence that future, who often ridicule AI outputs.

If these misconceptions dissipate from the minds of Silicon Valley and Wall Street, the significant AI vibe shift of 2026 will be realized.

This article reflects the opinion of the author.

Disclosure: Ziff Davis, Mashable’s parent company, in April 2025 filed a lawsuit against OpenAI, claiming it infringed Ziff Davis copyrights in training and operating its AI systems.