The public conversation about artificial intelligence has reached a strange contradiction. Workers are told that AI is coming for their jobs, managers are told that companies must “adopt or die,” investors are told that AI will transform margins, and governments are told that economic leadership depends on winning the compute race. Yet the evidence still does not show that AI has delivered broad, global, cost-effective productivity gains at the scale promised. In fact, one of the most revealing comments came from inside the industry itself: Bryan Catanzaro, Nvidia’s vice president of applied deep learning, said that for his team, “the cost of compute is far beyond the costs of the employees.” Axios reported the quote on April 26, 2026, and it was quickly amplified because it captured what many companies are discovering: AI is not automatically cheaper than people.

That statement matters because Nvidia is not an AI skeptic. It is one of the central companies selling the hardware behind the AI boom. If an Nvidia executive says compute costs can exceed payroll, the issue is not anti-technology bias. It is the basic economics of turning AI from a demo into a working business system. AI has real uses. It can help programmers, analysts, customer-service agents, writers, designers, researchers, and managers complete some tasks faster. But a task-level speedup is not the same thing as a company-wide profit improvement, and it is definitely not the same thing as a global productivity revolution.

The gap between AI hype and AI economics is now large enough that the fear around AI deserves a closer look. Companies do not need AI to be universally cost-effective in order to benefit from people believing it is. Fear itself has value. It can discipline workers, justify layoffs, support stock prices, pressure employees to work faster, rationalize surveillance, push customers toward automation, and make executives look decisive in front of investors. In other words, even before AI has proven its broad cost-effectiveness, the story that AI will replace you can already be useful to employers.

The point is not that every executive is lying, or that AI is useless. The point is that the fear narrative is doing economic work right now.

The evidence for broad AI payoffs is still weak

The most important distinction in the AI debate is between localized benefits and economy-wide benefits. A company can save time in one department while losing money overall. A worker can draft an email faster while other workers spend more time checking AI-generated errors. A software developer can produce code more quickly while the organization spends more on review, integration, cloud infrastructure, cybersecurity, compliance, and maintenance.

Several studies show that AI can improve performance in narrow settings. A well-known NBER study of customer support agents found that access to a generative AI assistant increased productivity by about 14% on average, with the largest gains for less-experienced workers. Other experiments found significant time savings in professional writing and coding tasks, including a study where GitHub Copilot users completed a programming task 55.8% faster. These findings are real, but they are task-level findings, not proof that AI lowers total enterprise costs across the economy.

The broader evidence is far less dramatic. In February 2026, an NBER working paper based on nearly 6,000 senior executives in the United States, United Kingdom, Germany, and Australia found that executives reported little own-firm impact from AI over the previous three years; nine in ten reported no impact on employment or productivity. That is a remarkable result because it comes after years of intense AI adoption pressure and billions of dollars in spending.

The MIT “GenAI Divide” research, widely reported in 2025, reached a similarly sobering conclusion: only about 5% of enterprise generative AI pilots achieved rapid revenue acceleration, while the vast majority stalled with little or no measurable profit-and-loss impact. The problem was not that workers had never tried AI. The problem was that enterprise value depends on workflow integration, data quality, governance, training, and measurement—not just access to a chatbot.

This is why the claim “AI is cheaper than workers” is often too simplistic. The price of a prompt is not the cost of an AI system. Real AI costs include chips, cloud contracts, electricity, cooling, data pipelines, security, vendor management, legal review, model evaluation, human oversight, error correction, retraining, and workflow redesign. When those costs are included, the economic case becomes much narrower.

Daron Acemoglu’s macroeconomic work has argued that AI’s near-term productivity effects may be modest. His NBER paper estimated that about 19.9% of U.S. labor tasks are exposed to AI, but only a portion of exposed tasks can be profitably automated at current costs; his baseline calculation implied total factor productivity effects over the next decade of less than 1% in total. That does not mean AI has no value. It means the transition from technical capability to profitable deployment is much harder than the public narrative suggests.

Compute is becoming a labor cost of its own

For years, companies treated labor as the main variable cost to be attacked. AI changes that equation by creating a new recurring cost: compute. Unlike traditional software, generative AI systems often become more expensive the more they are used. Every query, document, image, code suggestion, agentic workflow, and model call can consume compute. Training frontier models is expensive, but inference—the day-to-day running of models—also becomes a major expense when deployed at scale.

That is why Catanzaro’s comment was so important. It punctured the fantasy that AI is simply a cheaper digital employee. In some contexts, compute is not replacing payroll; it is becoming a second payroll.

Gartner forecast that worldwide IT spending would reach $6.31 trillion in 2026, up 13.5% from 2025, driven by momentum in AI infrastructure, software, and infrastructure-as-a-service. Gartner also projected that spending on data center systems would surpass $788 billion in 2026. These are not marginal expenses. They are huge capital commitments that must eventually be justified by revenue growth, cost reductions, or strategic control.

The energy burden is also material. The International Energy Agency reported that data center electricity demand surged in 2025 and projected that data center electricity consumption would roughly double by 2030, with AI-focused power use poised to triple. The IEA also noted that capital expenditure by five large technology companies surged to more than $400 billion in 2025 and was set to rise further in 2026.

This is the hidden contradiction in the AI fear campaign. Companies tell workers that AI is cheap enough to replace them, but they tell investors and governments that AI is important enough to justify one of the largest infrastructure buildouts in corporate history. Both cannot be casually true in every workplace. If AI were already a simple low-cost labor substitute, companies would not need to spend so much proving it can scale.

Fear helps companies before AI does

If AI has not yet delivered broad cost savings, why do so many companies talk as though replacement is inevitable? One answer is that fear creates leverage.

Workers who believe they are easily replaceable are less likely to demand raises, resist restructuring, change jobs, unionize, or push back against unreasonable workloads. Economists have long studied how the threat of automation can weaken labor’s bargaining power. Research published by the American Economic Association argues that the threat of automation weakens workers’ bargaining power in wage negotiations. That mechanism does not require automation to have already replaced everyone. The credible threat is enough to change behavior.

This is one reason the AI narrative is so useful to management. If a company says, “We need everyone to do more with less because AI is changing the business,” workers may accept pressure that would otherwise look like plain cost-cutting. If managers say, “Your job will not be replaced by AI, but by someone who uses AI,” the message is not only educational; it is disciplinary. It tells employees that resistance is career risk.

The fear also helps explain why companies can announce layoffs under the banner of AI even when the underlying economics are unclear. In 2025 and 2026, a growing number of companies linked workforce reductions, hiring freezes, or contractor cuts to AI. IBM said in 2023 that it would slow hiring for some back-office roles that could be replaced by AI. Klarna repeatedly promoted its AI assistant as doing the work of hundreds of customer service agents, while later reports indicated the company needed to restore more human support to address quality concerns. Duolingo announced in 2025 that it would gradually stop using contractors for work AI could handle.

Whether each of those decisions was economically justified is less important than the pattern: AI provides a powerful explanation for reducing headcount. It sounds modern, strategic, and unavoidable. A layoff blamed on poor planning can damage management credibility. A layoff blamed on AI can be framed as adaptation.

The investor audience matters

Executives are not only speaking to workers. They are speaking to Wall Street. Since the launch of ChatGPT, AI has become a required part of corporate storytelling. Companies that cannot explain their AI strategy risk looking behind; companies that can tell a convincing AI story may receive more patience from investors, even before profits arrive.

This is visible in earnings calls, annual reports, and risk disclosures. A 2024 Arize AI report found that 64.6% of Fortune 500 companies mentioned AI in their most recent annual reports, and that most generative AI mentions appeared in the context of risk disclosures. By 2025, Fortune reported that 72% of S&P 500 companies disclosed AI as a material risk in 10-K filings, up sharply from 2023.

AI has therefore become both a growth story and a risk story. Companies can say, “We are investing in AI to improve efficiency,” while also saying, “AI creates competitive risks that require further investment.” This dual narrative is convenient. It justifies spending, restructuring, acquisitions, vendor contracts, and workforce changes. If results are slow, the company can argue that the transition is still underway. If costs rise, it can argue that falling behind would be worse.

That does not mean AI investment is irrational. Some companies will build durable advantages. But the broad corporate incentive is clear: it is safer for executives to be seen as over-investing in AI than to be accused of missing the next platform shift. Fear of AI does not only affect workers; it also affects boards, analysts, and competitors.

AI can make work worse while looking like productivity

One of the strongest pieces of evidence against simplistic AI optimism is the rise of “workslop.” BetterUp Labs and Stanford Social Media Lab found that 40% of surveyed U.S. desk workers had encountered AI-generated workslop in the previous month, and estimated a cost of $186 per employee per month. Workslop is polished-looking output that lacks the substance needed to advance a task. It transfers work from the sender to the receiver, who must interpret, correct, or redo it.

This is a crucial point. AI can increase output while reducing useful output. A worker who uses AI to generate a long memo may appear productive. But if the memo is vague, inaccurate, or context-free, the cost is pushed onto colleagues. The organization has not saved time; it has redistributed cognitive labor.

The same issue appears in software, legal work, marketing, consulting, customer support, and administration. AI can accelerate first drafts, but many business tasks are not valuable because a first draft exists. They are valuable because someone understands the problem, makes tradeoffs, takes responsibility, and delivers a reliable result. When AI is used to skip that thinking, it creates the appearance of productivity without the substance.

This helps explain why enterprise ROI is weak. Many companies adopted AI at the individual-tool level instead of the system level. They gave workers chatbots, copilots, or writing assistants, then expected measurable transformation. But productivity gains at the individual level can disappear when the organization absorbs more review, coordination, rework, and compliance costs.

Fear shifts responsibility from companies to workers

Another reason companies promote AI fear is that it transfers responsibility. If workers are anxious about replacement, they may blame themselves for not adapting quickly enough. They may spend unpaid time learning tools, accept constant reskilling demands, or tolerate vague performance standards tied to “AI readiness.”

This is especially useful because many companies have not yet figured out how to deploy AI effectively. The MIT findings suggest that failure often comes from poor integration, not worker laziness. The NBER executive survey suggests most firms still have little measurable productivity impact. Yet the public message to workers is often: if you fall behind, it is your fault.

That framing benefits employers. It turns an organizational strategy problem into an individual employability problem. Instead of asking, “Has management built a reliable AI workflow with clear metrics and safeguards?” the worker is pushed to ask, “Am I obsolete?”

Surveys show that this fear is real. Pew Research Center found in 2025 that 52% of U.S. workers were more worried than hopeful about future AI use in the workplace, and 32% thought it would lead to fewer job opportunities for them in the long run. Gallup’s 2026 workplace polling found rising concern that AI or automation could eliminate jobs, even as many workers using AI reported individual productivity benefits.

That mix—useful tool, unclear enterprise ROI, high worker anxiety—is exactly the environment in which fear is most powerful.

The global benefits are uneven

Globally, the cost-effectiveness question is even more complicated. AI adoption depends on infrastructure, electricity, data access, cloud affordability, digital skills, language coverage, regulation, and institutional capacity. Wealthy firms in wealthy countries are much better positioned to experiment, absorb losses, and capture gains.

OECD research on AI and the global productivity divide warns that AI’s benefits may be uneven across countries, with low- and lower-middle-income economies facing adoption barriers even as they may benefit from spillovers over time. Microsoft’s AI Economy Institute similarly described global AI adoption in 2025 as record-setting but uneven, with a widening digital divide.

This matters because the strongest AI narratives often sound universal: AI will transform every economy, every company, every job. But the evidence points to uneven diffusion. Firms with proprietary data, capital, technical talent, and scale may benefit. Smaller firms may pay for subscriptions and consultants without achieving transformation. Workers in some roles may gain leverage; others may face deskilling, surveillance, or wage pressure. Countries with cheap power and cloud infrastructure may attract data centers; others may become customers rather than producers.

So when companies say everyone must be afraid of AI, the question should be: who benefits from that fear? The answer is rarely “everyone.”

The real issue is power, not just technology

AI is often presented as an autonomous force, as if machines themselves decide who gets laid off, monitored, promoted, or underpaid. But companies decide how AI is deployed. They decide whether productivity gains become higher wages, shorter hours, better staffing, lower prices, or higher margins. They decide whether AI augments workers or replaces them. They decide whether employees participate in the redesign of work or merely receive instructions from above.

That is why fear is such a useful management tool. A fearful workforce is less likely to ask who owns the gains. If AI saves ten hours a week, does the worker get time back, or does the company raise quotas? If AI improves customer service, are employees trained for higher-value roles, or are they cut? If AI reduces drudgery, are jobs redesigned democratically, or is every keystroke monitored?

The most honest AI debate is not about whether the technology has value. It does. The honest debate is about distribution. Who pays the compute bill? Who absorbs the rework? Who is blamed when AI fails? Who captures the gains when it succeeds?

Conclusion: skepticism is not denial

AI is not fake. It is not useless. It is already changing many tasks, and some organizations will use it very effectively. But the evidence as of April 29, 2026, does not support the sweeping claim that AI has broadly proven itself more cost-effective than human labor across the global economy. The research shows something more complicated: narrow productivity gains, high infrastructure costs, uneven adoption, weak enterprise ROI, worker anxiety, and strong incentives for executives to exaggerate inevitability.

That is why companies want you to be afraid of AI. Fear lowers resistance before the business case is proven. Fear makes layoffs look strategic. Fear makes unpaid reskilling feel mandatory. Fear shifts accountability from management to workers. Fear supports investor narratives. Fear turns an expensive, uncertain technology rollout into an unavoidable destiny.

The right response is not panic. It is evidence-based bargaining. Workers should learn AI where it genuinely helps them, but they should not accept every claim of inevitability at face value. Boards should demand real ROI, not just AI theater. Policymakers should focus on competition, labor rights, energy costs, and shared gains. And companies should be judged not by how loudly they talk about AI replacement, but by whether their AI systems actually create durable value without simply pushing costs onto workers, customers, and the public.

AI may transform the economy. But fear is already transforming the workplace.

Amit Shrivastava