Uber’s chief technology officer walked into 2026 with a budget for AI tools. By April, it was ash. Praveen Neppalli Naga had burned through the company’s entire annual AI allocation, mostly on coding assistants, specifically Anthropic’s Claude Code, spreading across Uber’s 5,000-engineer organization faster than anyone planned for. “I’m back to the drawing board because the budget I thought I would need is blown away already,” he told The Information.
That story tells you everything about where we actually stand on AI being cheaper than human labor: not at some clean tipping point where robots suddenly replace workers, but deep inside a messy, expensive, confusing transition where the math keeps surprising everyone, including the people signing the checks.
Is AI Actually Cheaper Than Human Labor Right Now?
The short answer is: it depends, and far more often than you’d expect, AI is currently the more expensive option. A 2024 MIT CSAIL study analyzed computer vision tasks where AI could technically match human performance and found automation was economically viable in only about 23% of those cases. In the remaining 77%, maintaining AI systems, infrastructure, compute, energy, oversight, and cost more than simply keeping the human on payroll.
That finding stings when you consider how aggressively companies have been pitching AI-driven efficiency. Bryan Catanzaro, vice president of applied deep learning at Nvidia, put it plainly: “The cost of compute is far beyond the costs of the employees.” This is Nvidia saying it, arguably the company that benefits more than anyone from AI adoption going wider and faster.
But here’s where it gets complicated. “Right now” is doing a lot of work in that sentence. AI costs are falling, and they’re falling fast. What costs a dollar today may cost a dime in three years. The question isn’t just whether AI is cheaper than human labor today, it’s what happens when it finally is.
The Jobs Already Feeling the Pressure
Even if AI isn’t universally cheaper yet, it’s already reshaping demand for certain kinds of work in measurable ways, and the shift isn’t random.
Analyzing nearly all U.S. job postings from 2019 through March 2025, Harvard Business School research shows that openings for routine, automation-prone roles fell 13% after ChatGPT’s launch, while demand for more analytical, technical, and creative jobs grew 20%. That’s not a modest signal. That’s a structural rerouting of what companies think they need from people.
Industries with rich, abundant data, software development, customer support, and financial trading are seeing AI adoption rates of 60–70%. Sectors with sparse or unstructured data are struggling below 25%. Software development is arguably the most transformed: three-quarters of developers now use AI coding assistants, and tools like GitHub Copilot have had 420 million repositories to train on. Customer service is moving quickly, too, with IBM reporting AI tools cutting support costs by 23.5% in some deployments.
A separate 2025 MIT study found something even more striking: AI systems are now advanced and affordable enough to perform work tied to roughly 11.7% of the U.S. labor market, representing about $1.2 trillion in wages, spanning cognitive and administrative tasks across finance, healthcare, and professional services. That’s technical feasibility, not a timeline for mass layoffs, but it tells you the floor is dropping faster than most workforce planning models anticipated.
Why Companies Keep Spending Despite the Math
So if AI is more expensive in most cases, why are companies investing as if it’s the last train out?
Big tech firms have committed around $740 billion in AI capital expenditures this year, according to Morgan Stanley, a 69% increase from 2025. McKinsey projects AI spending could reach $5.2 trillion by 2030. That’s not a small bet hedged by a few optimistic executives. That’s a civilizational wager.
Part of it is competitive pressure. If your rival automates and you don’t, you’re not just slower, you’re structurally disadvantaged in ways that are hard to undo. Part of it is the trajectory of AI costs, which have historically fallen faster than almost any technology before them. And part of it, if we’re being honest, is hype doing what hype does.
Experts predict AI costs could drop by over 90% in four years, but only if infrastructure, pricing models, and reliability all improve in tandem. The bet companies are making isn’t on today’s economics. It’s tomorrow. And given how quickly inference costs have declined over the past two years, that bet has a real chance of paying off.
There’s also a cultural dimension. At Uber, engineers were ranked on internal leaderboards by AI usage, essentially gamifying token consumption. When tools are free to experiment with but billed by usage, you end up with budgets that evaporate faster than anyone modeled.
Which Industries Are Feeling It First?
The disruption from AI, cheaper than human labor, isn’t hitting every sector equally. It’s clustering, and that clustering follows data density and task repeatability.
In the first half of 2025 alone, nearly 78,000 tech jobs were directly tied to AI-driven layoffs. Wall Street banks have signaled plans to eliminate approximately 200,000 roles over the next three to five years, targeting entry-level and back-office positions in particular.
Manufacturing is accelerating, too. Robotics and AI-driven automation could displace millions of additional factory workers as systems become cheaper and more capable. A customer service center that once employed 500 people might transform into 50 AI oversight specialists working from a single location. The arithmetic is stark, even when the timeline stays uncertain.
Finance, always efficient in its love of efficiency, is moving aggressively. AI reads thousands of financial reports in minutes, spots patterns human analysts miss, and executes trades at speeds that make human reaction time irrelevant. High-frequency trading already accounts for roughly 70% of U.S. equity market volume. The question in finance isn’t whether AI is arriving; it’s how much of the human workforce remains meaningful once it does.
What Happens to Workers When the Math Finally Shifts?
Here’s the part most think pieces quietly skip over: the transition is the hard part, not the endpoint.
The WEF’s Future of Jobs Report 2025 projects that 170 million new roles will be created globally by 2030, while 92 million existing jobs are displaced, a net gain of 78 million positions. That sounds encouraging until you realize those aren’t direct swaps. The new roles aren’t in the same places, don’t demand the same skills, and don’t pay the same wages as the ones disappearing. A factory worker in Ohio isn’t automatically positioned to become an AI systems engineer in Austin.
The WEF has flagged that some companies are already skipping entry-level hiring entirely, since AI can now perform many of the tasks those roles once handled. That matters for reasons that don’t show up on a balance sheet. Entry-level positions aren’t just jobs; they’re how people accumulate judgment, develop relationships, and eventually become the decision-makers organizations depend on. Hollowing them out creates a skills pipeline problem that only becomes visible once it’s already a crisis.
The Skills That Survive
The jobs that hold up when AI becomes genuinely cheaper than human labor share something in common: they require something AI can’t yet replicate affordably at scale. Contextual judgment. Genuine relationship-building. Creative direction that involves real stakes. Ethical accountability. The ability to navigate ambiguity without a clean training dataset to lean on.
That’s not as narrow as it sounds, but it does mean workers who’ve built their entire careers around repeatable, data-driven tasks without developing judgment and communication alongside them face real displacement risk. Data literacy is now considered the new workplace currency, and 75% of U.S. employers now list lifelong learning and upskilling as a top organizational priority.
The Honest Truth About This Transition
AI is cheaper than human labor; it isn’t a binary event, a switch flipping, and humans becoming obsolete overnight. It’s a slow, sector-by-sector, role-by-role repricing of what human work is worth. And right now, we’re in the part where costs are very real, productivity gains remain murky, and the people making decisions are betting heavily on a future that hasn’t quite arrived.
What’s clear is that companies aren’t waiting for certainty. They’re building now for a world where AI costs keep falling. And as Keith Lee of the Swiss Institute of Artificial Intelligence’s Gordon School of Business put it, it’s not just about AI becoming cheaper than humans; it’s about becoming both cheaper and more predictable at scale.
When that moment fully arrives, the winners won’t be the ones who worry about AI the most. They’ll be the ones who figured out, well ahead of time, how to work alongside it, direct it, and build things it can’t build alone.
Uber’s CTO blew his budget. But he didn’t fold; he went back to the drawing board. That instinct to adapt, recalculate, and keep moving is probably the most AI-proof skill of all.
