This article digs into a weird paradox in tech right now: we’re seeing huge layoffs and a tidal wave of AI investments at the same time. But if you look at the numbers—cost, reliability, productivity—AI isn’t actually a simple swap for human workers yet.
Official job cuts keep making headlines. Corporate budgets are tilting hard toward AI. The result? A messy cost curve, reliability headaches, and shifting business models that could reshape work for a long time.
AI economics in the real world: layoffs, capex, and the cost delta
Recent layoffs and hiring pauses at big tech companies, like Meta’s plan to axe about 8,000 jobs and freeze hiring for 6,000 more, make it look like AI is taking over. But insiders say AI still costs more than people. Nvidia’s Bryan Catanzaro told Axios that his team’s compute bills outstrip what they’d pay for labor. So, while capital spending on AI is exploding, the job market is still all over the place.
Looking at the bigger picture, there’s real tension between spending and results. A 2024 MIT study found AI automation only makes financial sense in 23% of visually-focused jobs. For the other 77%, humans are still cheaper. That’s a huge detail—AI can handle some stuff, but most roles that need eyes or brains are still better done by people, at least with today’s tech and energy prices.
Are machines cheaper? The MIT view on job types
The MIT study really drives home a point: not everything can be automated, and costs swing wildly depending on the task. That 77% of jobs—things like making decisions or dealing with tricky customer problems—often need flexibility, context, and reliability that AI just can’t nail at scale yet. So, companies are mixing things up: using AI for boring, repeatable stuff and keeping humans in the loop where judgment and accountability actually matter.
Reliability and risk: real-world failures and trust
Reliability is still a big sticking point. There are stories, like one engineer who watched an AI agent wipe out a database and a network, that highlight trust and safety issues. These aren’t just embarrassing—they cost real money and create risks that companies have to handle with human oversight, lots of testing, and better rules. Because of these gaps, most teams roll out AI slowly and still lean on people for the mission-critical stuff.
Capital flows: how much money is actually going into AI
The financial side is just as tangled. Big Tech is pouring money into AI—Morgan Stanley says there’s been $740 billion in announced AI-capex this year, up 69% from 2025. At the same time, layoffs are everywhere—Layoffs.fyi counts more than 92,000 job cuts in 2026 so far, already beating 2025. Keith Lee from the Swiss Institute of Artificial Intelligence’ Gordon School of Business calls this a short-term mismatch caused by high hardware and energy costs. In other words: the money is flowing, but the productivity gains haven’t caught up yet.
Outlook: costs, models, and the path to scale
Looking ahead, AI spending could skyrocket. McKinsey thinks AI-related spending might hit $5.2 trillion by 2030—or even $7.9 trillion if things speed up—mostly going into data centers and IT gear. Gartner predicts inference costs for 1-trillion-parameter models could drop by 90% in four years. That could flip pricing models from subscriptions to pay-as-you-go. If AI gets cheaper and more reliable, it’ll open up more use cases. But for now, AI needs to get both cheaper and more reliable—with fewer hallucinations and less babysitting—before it can really take over from people in most jobs.
Implications for businesses and policy
- Cost parity is not the default: AI usually isn’t cheaper than hiring people for most tasks right now.
- Reliability matters: Real-world failures keep showing us why we need governance, testing, and human-in-the-loop setups.
- Capital is racing ahead: Investors keep pouring money in, but productivity gains haven’t quite caught up. There’s a weird tension between spending and what you actually get out of it.
- The economics could shift: If inference costs drop and pricing changes, AI could get a lot more affordable. But that only works if reliability and scalability actually get better.
Here is the source article for this story: ‘The cost of compute is far beyond the costs of the employees’: Nvidia executive says right now AI is more expensive than paying human workers