Executives at the biggest technology companies paint a picture of universal embrace. Their models improve daily. Usage metrics climb. Yet fresh surveys and usage data tell a different story. Large segments of the population engage with artificial intelligence only now and then. Others avoid it completely.
Gabriel Weinberg, founder and CEO of DuckDuckGo, laid out the numbers in a recent analysis. He challenged the narrative pushed by a New York Times article from June 2025 that suggested once people try AI they apply it to every task. “Once you’ve tried AI, you use it ‘for everything.’ No, in fact most people who’ve tried it are just occasional AI users,” Weinberg wrote on his site (gabrielweinberg.com, June 13, 2026). The piece draws on data from Microsoft, Gallup, Searchlight Institute and others to map actual behavior.
Triangulation of those sources points to a split. Roughly one third of Americans actively use AI. Another third turn to it occasionally. The final third never touch it. Microsoft’s telemetry from the first quarter of 2026 shows just over 30 percent of U.S. working-age adults using AI tools. That figure rose three percentage points from late 2025. Still, 70 percent remain on the sidelines. And even among users, sessions often stay short. Many log less than 90 minutes a month on major platforms.
Searchlight Institute’s 2026 survey found 58 percent of Americans have tried AI. Of those, 30 percent count as regular users while 29 percent stay infrequent. Desktop traffic data from Datos in mid-2025 reinforces the pattern. Only 21 percent of users visited AI sites ten or more times per month. Sixty-two percent recorded zero visits. The rest fell somewhere in between. The Argument, another 2026 report, noted most people reach for AI once a week or less.
Younger generations show mixed signals. Gallup polls from 2025 and early 2026 revealed 79 to 81 percent of Gen Z respondents had used AI at least rarely. Yet 21 to 19 percent reported never trying it. Anxiety ran high at 41 to 42 percent. Anger registered with 22 to 31 percent. These numbers suggest familiarity does not always equal enthusiasm.
Concerns drive much of the hesitation. Top worries include job displacement, cited by 42 percent. Privacy violations follow at 35 percent. Misinformation and lies worry 33 percent. A solid majority told pollsters the government should set safety and privacy rules for AI even if that slows American progress relative to China. The sentiment appears across age groups and political lines.
Public perception of AI’s overall value remains tepid. Weinberg highlighted a net positive rating of just 8 percent. That sits next to social media’s 7 percent. By comparison, earlier technologies such as the internet or solar power scored far higher when they matured. People simply do not buy the optimistic forecasts coming from corporate leaders. Skepticism runs deep.
Weinberg offered an analogy to meat consumption. Ninety-five percent of Americans eat meat. Yet 70 percent say they have cut back on red meat. Thirty percent eat it rarely or occasionally. Twelve percent avoid red meat altogether. Four percent identify as vegetarian and 1 percent as vegan. The pattern shows a continuum of behavior rather than a binary choice. AI follows similar lines. Some dive in. Others sample. Many abstain for ethical, practical or ideological reasons.
But there’s more to the picture. Recent research highlights the hidden costs behind growing AI use even among the minority who engage heavily. A study published in Patterns examined data center demands (cell.com, 2026). Artificial intelligence now acts as the main driver of electricity growth in those facilities. The International Energy Agency estimated AI accounted for 15 percent of data center power in 2024, possibly reaching 20 percent by year’s end. Projections suggest AI systems could demand 23 gigawatts by the close of 2025, pushing their share close to half of total data center consumption.
Those figures carry environmental weight. Data centers produced roughly 182 million tons of carbon dioxide emissions in 2024 from electricity generation. They consumed 560 billion liters of water in 2023 for cooling. A separate analysis from the Center for Sustainable Systems at the University of Michigan noted AI and machine learning represented only 0.01 percent of global greenhouse gas emissions in 2024 yet account for over half of recent data center load growth (css.umich.edu, September 2025, with 2026 updates).
Individual queries add up. One ChatGPT request uses about ten times the electricity of a standard Google search, according to IEA figures cited across multiple reports. Image generation can consume energy equivalent to charging a smartphone. As enterprises scale deployments the token counts climb fast. A Wall Street Journal report on token economics noted AI has become one of the fastest-growing line items in corporate technology budgets, sometimes reaching half of IT spend (deloitte.wsj.com, 2026).
Businesses face pressure on both sides. Some companies report AI automation costing more than the full-time employees it replaces. Others push for shorter contracts as vendor expenses mount. The Information documented shifts at firms like Atlassian and HubSpot away from flat per-user fees toward consumption-based models (theinformation.com, April 2026). Token usage patterns vary wildly. Light users nibble. Power users burn through millions of tokens monthly on complex agents and workflows.
Discussions on X in recent days reflect the tension. One user noted companies blowing through tokens while seeking staff reductions. Another highlighted that most people still prompt AI one query at a time rather than building persistent systems. The gap between casual consumption and sophisticated application grows wider.
DuckDuckGo itself takes a different path. The company makes every AI feature optional. Its duck.ai chatbot emphasizes privacy. That stance aligns with the regulatory preferences expressed in surveys. Users who want AI can have it. Those who don’t face no pressure.
The picture that emerges is no longer one of inevitable takeover. Early adopters and technology insiders operate inside a bubble. They test the newest models, integrate agents into workflows, and debate token efficiency. Outside that circle behavior looks far more measured. Some workers use AI to draft emails or summarize reports. Others ignore the tools entirely. Students experiment then retreat when results feel unreliable. Executives tout productivity gains while their rank-and-file report minimal change.
Future growth depends on addressing the barriers. Job fears will not vanish quickly. Privacy rules could slow innovation yet build trust. Better data on real usage, beyond self-reported surveys, will help. Microsoft’s telemetry offers one window. Traffic analytics from Datos and Searchlight add others. Combined they paint a more sober portrait than the hype suggests.
AI will keep improving. Models will grow more capable. Yet adoption may follow the uneven path of previous technologies. Electricity reached nearly every home but took decades. The internet transformed commerce and communication yet left millions offline or lightly engaged for years. Artificial intelligence appears headed for a similar trajectory. Heavy use by a minority. Occasional assistance for many. And a persistent group that opts out for reasons both practical and principled.
That reality carries implications for strategists, policy makers and investors. Billions pour into infrastructure to meet projected demand. If actual consumption stays patchy those investments could face scrutiny. Environmental costs already draw attention. So do questions about concentration of power among a few model providers. The gap between rhetoric and reality deserves close attention in boardrooms and Washington alike.
Weinberg closed his piece on a measured note. People occupy a spectrum of views and habits. Ignoring that spectrum risks misreading the market and the society AI aims to serve. The data supports his view. So do the fresh studies on energy footprints and corporate spending patterns. The age of universal AI use has not arrived. And it may never look quite like the boosters predict.
WebProNews is an iEntry Publication