AI Is Too Expensive To Replace Us

Who thought this was a 'smart' business idea?

Photo by Sebastian Svenson on Unsplash

Previously, I have rambled on and on about how AI is far too inaccurate and broken to replace human workers (such as this article, for example). From its constant hallucinations to its failure to perform basic tasks, it isn’t really a suitable replacement, let alone a viable alternative, for a cognitive human worker. But what if I am wildly wrong? What if, by some miracle, Altman and Amodei pull a metaphorical rabbit out of the hat and AI can magically achieve what they constantly claim it can? Well, it turns out, your job is likely still safe, because believe it or not, we humans are cheaper than a giant plagiarism machine! But is that really good news?

Axios recently reported that Uber has announced it has already blown its entire 2026 AI budget in just four months! On top of that, the vice president of applied deep learning at Nvidia told Axios that “for my team, the cost of compute is far beyond the costs of the employees”. That sounds an awful lot like these AI systems are quite a bit more expensive than the people they were designed to replace.

But why?

Well, some have pointed to an idiotic new trend known as ‘tokenmaxxing’. That name alone is proof that the entire corporate world is suffering from a debilitating case of brain rot.

‘Tokenmaxxing’ is when a company measures an employee’s performance based on the number of AI tokens they use. In other words, the more frequently an employee uses an AI, the more they will be considered ‘higher performance’. This isn’t a niche idea, with companies like Spotify, Meta, Nvidia and Uber jumping on the tokenmaxxing bandwagon.

But it is an entirely nonsensical idea. Tokenmaxxing doesn’t measure HOW an employee is using AI — for example, they could use it woefully inefficiently. Likewise, the work they produce with it could be so dreadful that it creates an exorbitant load for their colleagues to manage. Additionally, we know that overrelying on AI tools significantly erodes a person’s skills and expertise, so maximising exposure to AI in a pre-professional environment likely indicates dangerous levels of brain drain.

In short, this metric prioritises the idea of work, not results. It is telling that all the companies rolling out tokenmaxxing have a vested interest in the AI boom actualising. It implies that they can extract more value from the concept of AI doing well than from their actual workforce.

So, is that what is happening at Uber and Nvidia — is tokenmaxxing causing these AIs to cost more than the workers they are meant to replace? Are they using these tools so inefficiently that the costs spiral away from them?

Well, tokenmaxxing almost certainly hasn’t helped!

But we have known for a long time now that AI is almost always a more expensive solution than human workers, even when deployed correctly.

Take, for example, the 2024 MIT study, which found that AI could only cost-effectively replace 23% of basic workers’ wages. In other words, AI was so expensive that 77% of the time, hiring a human was the cheaper option. Furthermore, these researchers found that, even if AI prices dropped 20% per year, it would still take decades for AI to become the cheaper option for the majority of jobs.

Amazon kindly proved this study correct by taking it from theory into real-world practice. Their “Just Walk Out” grocery stores used facial-recognition cameras, shelf sensors, and AI to track which items a customer had taken and charge their Amazon account when they left, eliminating the need for a cashier or self-checkout. They replaced retail staff with simple AI, not even a more expensive, complex and less accurate LLM. But, unsurprisingly, it didn’t go to plan. A report found that over a thousand remote workers had to be hired to monitor the video feeds and verify 70% of the customers’ purchases, given that the AI was consistently making mistakes. Despite these workers effectively being paid a slave wage, the costs added up, and Amazon’s “Just Walk Out” AI system became significantly more expensive than simply hiring regular cashier staff. As such, Amazon failed to sell the system to third parties, which resulted in the closure of almost all of these stores.

And here’s the really inconvenient part: AI isn’t getting cheaper; it’s getting more expensive.

Let’s start with the fact that, right now, AI is subsidised to the gills. For example, using a series of very roundabout deals, Microsoft currently rents compute power to OpenAI at a staggering discount. Other AI data centre companies like Coreweave rent out compute at such astronomically discounted prices that they remain insanely unprofitable. But, despite the heavily discounted infrastructure, AI companies all run at colossal losses. OpenAI loses huge sums of money per customer, and even its $200 plans are loss-makers. No wonder they likely reached losses of more than $10 billion in 2025.

With a flurry of AI IPOs from OpenAI, Anthropic and xAI/SpaceX, combined with growing economic risks, all of these companies are now under increased financial scrutiny. As such, they need to reduce their losses, which means easing all this subsidisation and dramatically increasing the cost of using AI.

However, there are other factors increasing the cost of AI. As these models become larger, the energy and infrastructure costs of running them (also known as inference) increase dramatically. Indeed, OpenAI’s inference costs appear to be dramatically increasing right now. But the Rampocalypse these AI companies created is also beginning to affect them, as the cost to buy the computer chips they need has gone absolutely bananas. Other data centre costs are rising too, such as energy and construction.

All of these factors dramatically elevate the cost of AI. For example, OpenAI’s ChatGPT5.2 costs $1.75 per token when its predecessor, ChatGPT5.1, cost just $1.25. That is a 40% cost increase in a matter of months! And it is only set to rise from there.

So, with all of this in mind, AI is too damn expensive to replace the vast majority of human workers, even if it were capable of doing the same job to the same standard.

Now, is that a good thing?

From one perspective, yes. It means people’s livelihoods are safe. After all, it appears as if AI is not actually causing any layoffs at the moment (read more here). When you consider the horrific damage such widespread job losses could cause, AI being this terrible is objectively a good thing.

But what does it say about wages that a statistical machine trained on copious amounts of stolen data and that can only ever hope to be a hollow, inaccurate, badly puppeteered imitation of us is more expensive than just hiring actual people? What does that say about the state of wage theft today?

In our modern climate, anyone seriously considering replacing human workers with AI is just fetishising the idea of a robotic slave workforce that allows them to avoid having to treat their fellow humans with dignity. Why? Because it makes no business sense to replace workers with AI! The fact that this is even a point of contention proves there is a sickness in our society.

Either way, this doesn’t feel like something to celebrate. It’s like finding out the killer whale at SeaWorld has grown accustomed to its tank and so can never return to the ocean. The fact that workers are cheaper than the hollow plagiarism machine designed to badly imitate them shows exactly how abused, neglected and shortchanged the modern workforce is.

---

Thanks for reading! Everything expressed in this article is my opinion, and should not be taken as financial advice or accusations. Don’t forget to check out my YouTube channel for more from me, or Subscribe. Oh, and don’t forget to hit the share button below to get the word out!