AI is replacing jobs — with more jobs

I wouldn’t blame you for being confused about AI’s impact on jobs. One minute, big tech says AI will replace workers; the next, it’s hiring thousands to implement it. So which is it, or could both be true? 

Let's start with the premise that AI is hard for everyone and few companies are effectively deploying it. Study after study, including one from McKinsey in March, confirms that roughly a third of companies are able to scale AI across their organizations. Most others are struggling to find value except in pockets. Many projects continue to be stuck in proof of concept or experimentation phases. 

Last month, GitHub, which is owned by Microsoft, suspended new signups for Copilot Pro, its AI coding tool, because it was too expensive to operate on a subscription model. A week later, it announced it was switching to a usage-based pricing model. But perhaps it's not surprising when you look at the CapEx numbers required to build the infrastructure to support AI, and the cost required to run it. 

The big three alone – Amazon, Google and Microsoft – collectively projected around $250 billion in infrastructure spending in 2025, and actually blew past that figure. Their investment is expected to balloon to almost $600 billion this year. What's really wild here though is that these companies are making the biggest infrastructure bet in history, yet their enterprise customers still can't figure out how to use AI well.

Then consider the big AI labs (OpenAI, Anthropic, Google and Meta) are all subsidizing usage well below cost, essentially paying you to use their products, at least for now. Clearly subscription fees aren't paying the cost of operating the underlying infrastructure. If you doubt it, I have a monthly Claude subscription, and still get throttled regularly, where I am shut off for hours at a time with the invitation to pay for extra tokens to keep running uninterrupted (no thank you).

Yes, subsidizing early usage to build market share is a time-honored tech playbook, but the massive ongoing infrastructure investment required to run these models makes the path to profitability far less certain than it was for traditional software companies.

And that subsidy could be short-lived anyway as more companies follow GitHub's usage pricing lead. No big surprise that the cost floats downstream to the consumer eventually. Any way you slice it, the current approach is unsustainable.

So how are companies paying for these huge costs? Meta, Microsoft, Oracle and Amazon have slashed tens of thousands of jobs. While they claim the jobs are a victim of AI efficiency, it's more likely they are actually a victim of AI inefficiency. It costs a fortune to run this technology, and someone has to pay for it.

Hiring to fix what AI was supposed to fix

So we are losing jobs with executives claiming it's AI that's causing the job loss, that they are so much more efficient, they need fewer people to do the same amount of work. Even if that's true, and it's a debatable claim, how does it explain why tech companies are suddenly hiring thousands of engineers to…wait for it…help implement AI.

AI is so hard, and companies are finding it so difficult to demonstrate ROI, that the vendors have decided the way to resolve this is to build their own mini Deloittes and Accentures and send out an army of "forward deployed engineers." OpenAI just launched a separate consulting company called OpenAI Deployment Company with $4 billion in funding, and acquired a startup with 150 of these specialists. And OpenAI isn't the only company taking this route. These are essentially consultants who they hope can march in and save the day. The problem is that those engineers, like those big consulting firms, cost a pile of cash. 

The good news is there are a bunch of new jobs for engineers who actually communicate with customers, hear about their problems, and maybe get to the root of the consistent AI project failure. The bad news is the price tag for these projects gets even more expensive, and it doesn't solve the jobs issue for non-engineering roles.

So if you take a cool look at what we have here, it’s ghastly expensive, few deployments are actually successful, and massive numbers of employees are paying the price of deploying it. None of this makes sense, and it raises a hard question: how long can we keep cutting jobs to pay for a technology that then forces us to hire even more expensive people just to get it to work?

~Ron