Humans Are Even Cheaper Than Statistical Machines? AI Won’t Take Your Job, But That’s Far From Good News

God Translation Bureau is a translation team under 36Kr, focusing on fields such as technology, business, workplace, and life, and mainly introducing new technologies, new ideas, and new trends from abroad.

Editor's note: The AI computing power has been exhausted, and the budget has been burned out ahead of schedule. It turns out that the most expensive thing is not humans, but that crazy "token - maxxing" plagiarism machine. This article is from a compilation.

Previously, I've talked over and over again about why artificial intelligence can't replace human employees due to its low accuracy and numerous loopholes. From frequent "hallucinations" to screwing up basic tasks, it is simply unqualified, let alone being a replacement for human employees with cognitive abilities. But what if I'm completely wrong? What if a miracle happens, and Altman and Amodei perform a magic act, enabling artificial intelligence to miraculously achieve the functions they've been boasting about? Well, it turns out that your job is probably still secure. Because believe it or not, we humans are much cheaper than this huge plagiarism machine!

Axios recently reported that Uber announced that it had burned through its entire 2026 artificial intelligence budget in just four months! Moreover, the vice - president of applied deep learning at NVIDIA also told Axios that "for my team, the computing cost far exceeds the labor cost of employees." This sounds extremely ironic: these artificial intelligence systems are actually much more expensive than the humans they were originally supposed to replace.

But why is this the case?

In response, some people have pointed out a stupid new trend called "tokenmaxxing". Just hearing the name is enough to prove that the entire business world has suffered from a severe "brain - dead syndrome".

The so - called "tokenmaxxing" refers to a company evaluating employees' performance based on the number of Tokens (text units) consumed when they use artificial intelligence. In other words, the more frequently an employee uses AI, the higher their "performance" is considered to be. This is not some niche idea. Industry giants like Spotify, Meta, NVIDIA, and Uber have all joined this "tokenmaxxing" frenzy.

But this is a completely absurd idea. "Tokenmaxxing" can't measure how employees use AI at all - for example, their usage efficiency may be extremely low. Similarly, the work results they produce with AI may be terrible, imposing a huge burden on their colleagues to clean up the mess. In addition, we all know that over - relying on AI tools can seriously erode a person's professional skills and knowledge. Therefore, maximizing exposure to AI before or at the beginning of one's career may well indicate a dangerous decline in the intellectual level of talents.

In short, this indicator values the "working" posture rather than the actual results. Intriguingly, all the companies promoting "tokenmaxxing" have a personal stake in the monetization of the artificial intelligence boom. This implies that they can extract more commercial value from the concept of "excellent AI performance" than from their actual employee teams.

So, is this what's happening at Uber and NVIDIA - is it "tokenmaxxing" that is causing the cost of these AIs to exceed that of the employees they were supposed to replace? Is it because they are using these tools so inefficiently that the costs are completely out of control?

Well, "tokenmaxxing" should definitely take the blame!

But we've long known that even when properly deployed, artificial intelligence is almost always a more expensive solution than human employees.

Take a 2024 study by the Massachusetts Institute of Technology (MIT) as an example. The study found that, on a cost - effective basis, artificial intelligence can only replace 23% of the salaries of front - line employees. In other words, AI is so expensive that in 77% of cases, hiring humans is a cheaper option. In addition, researchers also found that even if the price of AI drops by 20% every year, it will still take decades for AI to become a cheaper option for most jobs.

Amazon has kindly proven the correctness of this study by bringing it from theory to the real world. Their "Just Walk Out" convenience stores use facial recognition cameras, shelf sensors, and artificial intelligence to track what items customers take and directly deduct the money from their Amazon accounts when they leave, eliminating the need for cashiers or self - checkout machines. They replaced retail employees with simple AI, and it wasn't even the more expensive, complex, and inaccurate large language models (LLMs). But as expected, things didn't go as planned. A report found that Amazon had to hire more than a thousand remote workers to monitor the video footage and verify 70% of customers' purchases because the AI system kept making mistakes. Although the salaries of these workers were as low as cheap labor, the costs added up. Eventually, the operating cost of Amazon's "Just Walk Out" AI system became much more expensive than directly hiring ordinary cashiers. As a result, Amazon failed to sell the system to third - parties, and most of these stores ultimately closed down.

And the truly embarrassing reality is that artificial intelligence is not getting cheaper; it's getting more expensive.

First of all, current artificial intelligence is enjoying a huge amount of subsidies. For example, through a series of extremely convoluted transactions, Microsoft is currently renting computing power to OpenAI at an astonishingly discounted price. Other artificial intelligence data center companies like Coreweave are also renting computing power at extremely low discounted prices, to the extent that they are still in a state of extreme loss. However, despite the high - level discounts on the infrastructure, AI companies are all making huge losses. OpenAI loses a large amount of money on each customer, and even its $200 package is a money - losing deal. No wonder their losses in 2025 are likely to exceed $10 billion.

As companies like OpenAI, Anthropic, and xAI/SpaceX are all gearing up for IPOs, coupled with the increasing economic risks, all these companies are now facing more stringent financial scrutiny. Therefore, they must reduce their losses, which means that the subsidies will gradually decline, and the cost of using AI will skyrocket.

However, there are other factors driving up the cost of AI. As the models become larger and larger, the energy and infrastructure costs (i.e., inference costs) required to run them are increasing sharply. In fact, OpenAI's inference costs seem to be soaring right now. Moreover, the "Rampocalypse" caused by these AI companies themselves is starting to bite back - the cost of purchasing the required computer chips has skyrocketed. The costs of other data centers, such as electricity and construction fees, are also rising.

All these factors are driving up the cost of AI. For example, the price of OpenAI's ChatGPT5.2 per 10,000 words (or per a certain amount of Tokens) has reached $1.75, while its predecessor, ChatGPT5.1, was only $1.25. This represents a 40% increase in cost in just a few months! And it will only get more expensive in the future.

Therefore, considering all this, artificial intelligence is simply too expensive to replace the vast majority of human employees, even if it has the ability to perform the same work to the same standard.

So, is this a good thing?

In a way, yes. It means people's livelihoods are secure. After all, at present, AI doesn't seem to have actually triggered any wave of layoffs. Considering the catastrophic damage that such a large - scale unemployment could bring, the poor performance of AI has objectively become a good thing.

However, if a statistical machine trained with a large amount of stolen data - a marionette that can only give a hollow, wrong, and clumsy imitation of us - is more expensive than directly hiring real people, what does this say about the low level of our wages? What does it reflect about the current situation of "the deprivation of labor value" in today's society?

In today's social climate, anyone who seriously considers replacing human employees with AI is simply blindly worshipping the fantasy of "robot slave labor" to avoid the responsibility of giving their fellow humans the dignity they deserve. Why? Because replacing employees with AI simply doesn't make sense commercially! The fact that this topic can even be a point of debate itself proves that our society is sick.

Anyway, this doesn't seem like something to celebrate. It's as heartbreaking as finding that the orcas in SeaWorld have gotten used to their aquariums and can no longer return to the ocean. The fact that human employees are cheaper than a hollow plagiarism machine designed to clumsily imitate them exposes how wantonly modern workers are exploited, ignored, and mistreated.

Translator: boxi.