The AI backlash is real. The adoption curve is stronger.

Why it matters

The backlash is forcing AI companies to pay for externalities, but adoption data shows founders and SMBs are already turning AI into everyday operating leverage.

The AI backlash is real, but the data points to diffusion, not retreat.

Casey Newton's July 1 Platformer column makes the strongest case for the opposition: communities are fighting data centers, workers are anxious about job loss, and consumers are beginning to feel the cost of the compute buildout in electricity bills, device prices and shortages of memory and storage. That account is not wrong. It is incomplete.

The missing half is the adoption curve. Technologies do not become ubiquitous because the public feels uniformly good about them. They become ubiquitous when founders, workers and small businesses keep finding ways to turn them into leverage despite the friction. On that measure, AI is behaving less like a speculative feature cycle and more like the early commercial internet: uneven, distrusted, expensive to scale, politically contested and already too useful to stop.

Stanford's 2026 AI Index says generative AI reached 53% population adoption within three years, faster than the personal computer or the internet. The same report estimates U.S. consumer surplus from generative AI at $172 billion annually by early 2026, up from $112 billion a year earlier. That is the number the backlash story has to confront: people may distrust AI in the abstract while still using it often enough to assign it large economic value.

Stanford HAI's economy chapter also shows why the internet analogy is not rhetorical. AI is spreading through consumer behavior, enterprise workflows and founder formation at the same time. The web first entered companies sideways through email, search, personal websites and bottom-up employee use. AI is following the same path, but faster, because it is riding on infrastructure the internet already built.

OpenAI's own research is self-interested, but it is still useful as a window into usage at scale. In its January 2026 report on ChatGPT usage and adoption patterns at work, OpenAI said ChatGPT had more than 700 million weekly active users and that more than a quarter of U.S. workers reported using ChatGPT for work. In a December 2025 enterprise report, OpenAI said ChatGPT served more than 800 million weekly users and that weekly messages in ChatGPT Enterprise had increased roughly 8x over the prior year. Those are company-provided figures, but they are consistent with third-party adoption data from Gallup and Stanford.

Gallup's February 2026 survey of 23,717 U.S. employees found that half of employed American adults used AI at work at least a few times a year, up from 21% in Q2 2023. Frequent usage is still much smaller: 13% used AI daily and 28% used it a few times a week or more, according to Gallup's AI indicator. That gap matters. AI is already mainstream by occasional use, but not yet by daily workflow ownership. The next few years will be fought over that conversion.

That is where founders enter the story. AI's first consumer wave made chatbots visible. The current startup wave is making AI operational.

Stripe's data shows the speed change. In Indexing the AI economy, Stripe said the top 100 AI companies on its platform reached $1 million in annualized revenue in a median 11.5 months, four months faster than the fastest-growing SaaS companies it compared them with. They reached $5 million in 24 months, compared with 37 months for the SaaS cohort. For AI companies founded from 2020 to 2023, the median time to $1 million annualized revenue was five months.

Those numbers do not prove that every AI company is durable. Stripe sees the companies that process payments through Stripe, not the whole market, and revenue velocity can hide high churn, high inference cost and weak defensibility. But the data does prove that founders are converting AI demand into paid products faster than prior software cohorts converted cloud demand into SaaS revenue.

AWS is telling a similar story from the infrastructure side. On June 30, AWS released an independent study of more than 3,400 founders and startup leaders across 20 countries. AWS said AI-native startups in the study reached billion-dollar valuations in 3.5 years, half the time it took comparable predecessors before generative AI, and reported 156% average annual revenue growth versus 65% for startups overall. Because AWS sells the cloud capacity these companies use, its framing should be read with that incentive in mind. The underlying point still holds: AI lowers the cost of product creation, customer support, coding, analytics and go-to-market, so smaller teams can attempt company-building patterns that previously required larger headcount.

The more important shift is not among venture-backed AI labs. It is on Main Street.

The U.S. Chamber of Commerce said in August 2025 that 58% of small businesses used generative AI, up from 40% in 2024 and more than double the rate in 2023. It also said 82% of small businesses using AI had increased their workforce over the prior year, a finding that cuts against the simplest version of the AI-equals-layoffs argument. The Chamber is a business lobby and has a deregulatory interest in the framing, so the figure should not be treated as neutral labor-market proof. It is still evidence that small businesses are adopting AI as operating leverage, not only as a headcount reduction tool.

A newer U.S. Chamber of Commerce Foundation and Ipsos survey, released June 17, 2026, is sharper because it asks workers at small businesses what they are actually doing. The Main Street AI Monitor found that half of small-business workers already use AI at work. Among those users, 64% said their primary use was personal productivity, such as drafting, summarizing and brainstorming. Only 6% said they used AI to automate workflows with minimal human involvement. When AI saved time or improved quality, 59% said they used the time to do more work or produce higher-quality output.

That is the mechanism by which AI becomes as common as the internet: not through a single killer app, but through millions of small workflow substitutions. A coffee shop owner writes product descriptions faster. A contractor drafts estimates. A dentist's office summarizes intake notes. A local law firm searches documents. A solo founder ships a working prototype over a weekend. None of those use cases looks like the future in isolation. Together, they look like the software layer of small business being rewritten.

The productivity evidence is not uniformly flattering to AI, and it should not be oversold. Some studies show modest gains, some show gains concentrated among less experienced workers, and some show negative effects when tools are bolted onto complex work without process redesign. The best early result remains narrow but important: in the NBER paper Generative AI at Work, Erik Brynjolfsson, Danielle Li and Lindsey Raymond found that customer support agents using a generative AI assistant saw a nearly 14% productivity increase, with larger gains for novice and lower-skilled workers. That is not AGI. It is an economically meaningful tool in a measurable workflow.

This is also why the job debate is harder than both sides want it to be. Platformer is right to highlight Stanford-linked payroll data suggesting pressure on junior workers in AI-exposed roles. Entry-level work is the part of the labor market most likely to be compressed first because AI is good at drafting, summarizing, coding boilerplate and answering first-pass questions. But compression is not the same as disappearance. The internet killed some intermediaries and created others; AI is likely to do the same at a faster tempo. The open question is whether companies use the gains to create more output and new roles, or simply to flatten career ladders.

The data center backlash is the hardest constraint because it is physical. A chatbot can go viral globally; a substation cannot. The International Energy Agency's Energy and AI work projects strong data-center electricity growth and says data centers will account for less than 10% of global electricity demand growth between 2024 and 2030 in its base case. In the United States, the pressure is more concentrated. The IEA says U.S. data centers are on course to account for almost half of electricity demand growth between now and 2030. That means the backlash will bite hardest in local permitting, utility rate design and grid interconnection.

But that is a constraint on where and how AI gets built, not evidence that AI usage will recede. The internet required fiber, server farms, routers, spectrum and smartphones. Those buildouts produced local fights, capital cycles and monopoly concerns. They did not stop adoption because the demand kept compounding. AI has the same structure, with one difference: the value is showing up simultaneously for consumers, founders, enterprises and small businesses.

The industry still has a credibility problem. AI companies have overpromised autonomy, understated hallucination risk, leaned on other people's copyrighted work, shifted infrastructure costs onto communities and used vague productivity claims to justify layoffs. The backlash is not irrational. It is a bill coming due.

But the answer to whether AI becomes ubiquitous is already being written outside the backlash discourse. It is in the small-business worker who uses AI to draft 20 customer emails before lunch, the founder who gets to revenue before raising a seed round, and the startup that sells into 50 countries in its first year because distribution is digital from day one.

The internet did not win because everyone loved it. It won because not using it became professionally and commercially expensive. AI is moving toward that same point. The backlash will shape the rules, the cost structure and the winners. It is unlikely to stop the adoption curve.