The artificial intelligence bubble is about to burst, but the next one is already in the making.

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

Editor's note: Another theory about the AI bubble. The original author uses the input - output ratio of GPT - 5 to argue that artificial intelligence is a bubble, and believes that the financial and technology circles are creating a new "trend" to escape. This article is from a compilation.

Speculation dominates the world. But it wasn't always like this. From the 1980s to 2008, things changed. Investors realized that the returns they got from hype were far greater than those from any legitimate business. After all, this is the information age, and information can be easily manipulated and commodified. This has led to the Internet bubble, the 2008 credit crisis, the cryptocurrency bubbles from 2016 - 2017 and from the end of 2020 to 2021, and the NFT bubble in 2022. The latest craze is the artificial intelligence bubble. In fact, nearly half of the world's private investment is flowing into the artificial intelligence field, and speculation in artificial intelligence is the main driving force behind the recent growth of the S&P 500 index. However, just like other bubbles before their catastrophic bursts, the artificial intelligence bubble also shows signs of imminent rupture. Nevertheless, those in the financial and technology circles have learned their lessons and are creating the next "trend" to siphon off all our money so that they can leave safely when they inevitably need to withdraw. Unfortunately, this new trend is even more of a dead - end compared to artificial intelligence.

So, it's basically a consensus that the artificial intelligence bubble is ripe to burst. Theories such as the "Efficient Computing Frontier" and the "Floridi Conjecture" imply that our existing artificial intelligence models have almost reached their performance limits. Even if OpenAI spends trillions of dollars to expand its model size tenfold, the performance will only improve slightly. The recently released ChatGPT - 5 is a perfect example. Compared with its "little brother" ChatGPT - 4, ChatGPT - 5 consumes significantly more data, training, and funds, but its performance has only a negligible improvement.

This is a big problem! Because at present, generative artificial intelligence models are actually not that useful, and are far from being profitable.

A report from the Massachusetts Institute of Technology found that 95% of AI pilot projects failed to increase a company's profits or productivity. In the remaining 5% of the projects, AI was relegated to the background, performing highly restricted administrative tasks, and even then, it only brought negligible improvements. A report from METR found that AI coding tools actually slowed down developers. The inaccuracy of these models means that they repeatedly create very strange coding errors, which are extremely difficult to find and correct. Logically, it would be faster and cheaper for developers to code on their own. Some studies have even found that for 77% of employees, AI increases their workload rather than improving their productivity. For now, the error rate of generative AI is too high to bring meaningful productivity or profitability improvements in most application scenarios.

In other words, for AI models to truly meet the speculative expectations that drive their huge investments, they need to become much, much better, which requires exponentially increasing capital investment.

This is another huge problem because OpenAI, which has the largest customer base among all generative AI companies to date, is still losing a lot of money with its $200 - per - month subscription plan. In fact, they seem to need to set the price at around $2000 per month to break even.

In the past few years, technology giants supported by venture capital and investment banks have spent hundreds of billions of dollars on artificial intelligence every year. However, this technology is reaching its limits and cannot be improved, and profitability is still a long way off. This is a perfect bubble, where huge amounts of money are used to support baseless or even completely false speculations. Now, with the disappointing debut of GPT - 5, Meta's restructuring and low - key handling of its AI department, and the threat of possible interest rate hikes, the investors who helped inflate this bubble are starting to warn that it is about to burst. Even Goldman Sachs, which has invested a large amount of money in the artificial intelligence bubble, warns that the artificial intelligence bubble is likely to burst soon, and it will also bring down the data center bubble, causing great damage not only to xAI, Meta, Google, Anthropic, and OpenAI, but also affecting third - party players in artificial intelligence infrastructure such as Amazon, Oracle, and Nvidia.

In other words, when this bubble bursts, it will have an incredibly devastating impact on everyone in the technology and financial circles.

Fortunately, they have a plan to avoid this man - made economic apocalypse - quantum computers. And they can't wait to get us on this new "bandwagon".

Unfortunately, I don't have time to explain in detail how quantum computers work, but if you're interested, Veritasium's introduction video is excellent. However, in simple terms, traditional computers use bits (which can be 1 or 0), while quantum computers use qubits, which can be both 1 and 0 at the same time. This means that in theory, quantum computers can have exponential computing power because they can receive a large number of inputs at once and output a large number of results simultaneously. In fact, a mathematical problem that would take our most powerful supercomputer longer than the age of the universe to solve was recently solved by a quantum computer in just a few minutes. Therefore, quantum computing is expected to bring revolutionary progress to the field of chemistry by calculating complex molecular structures and interactions; similarly, in the fields of machine learning and artificial intelligence, its exponential computing power may also break through the limitations of current technologies.

In fact, some people even propose that our brains are quantum computers, which means that quantum artificial intelligence may ultimately create machines with true human - level intelligence, and they will operate more efficiently and at a lower cost than current models.

I'm sure you've seen the pattern. Everyone in the technology and financial circles who has invested huge amounts of money in artificial intelligence now believes the speculative claim that quantum computers can solve all of artificial intelligence's problems, and thus they are pouring their money and influence into it.

AI giants Google, Microsoft, and Amazon are currently developing their own quantum computers. Nvidia is developing hardware and software platforms for quantum computers. OpenAI recently hired some of the world's top photon - based quantum computing physicists. Even Musk has started to propose quantum computing ideas for his AI ambitions. But this isn't just a game for the giants; smaller quantum computing companies are also starting to receive huge investments, and their values are skyrocketing. Take Quantinuum for example. This small quantum computing research startup recently raised $600 million, doubling its valuation to $10 billion.

This seems to be their "escape pod" from the artificial intelligence bubble - forcing money and hype into this technology because it promises to solve all the problems faced by the artificial intelligence community. So, is this just replacing the old bubble with a new one? Are these technology giants and their supporters pouring billions of dollars into false promises again? Or can quantum computing really solve the problems faced by the artificial intelligence industry?

Well, unfortunately, quantum computing isn't as good as it seems on the surface.

First of all, the hardware is still a long way from being fully usable. It will take 10 to 20 years to build a real, general - purpose, and functional quantum computer. They are extremely expensive to build and even more difficult to operate. Although huge investments may speed up this process, as nuclear fusion technology shows, nothing is guaranteed.

To be honest, the hardware isn't the key problem; it's the software.

In most cases, quantum computers are actually much slower than ordinary supercomputers. They can only outperform the latter when dealing with very specific and complex tasks such as factorization. However, due to the special nature of qubits, these computers cannot run standard codes or algorithms. They require specific algorithms, and this is where the problem lies.

Remember when you open the box of Schrödinger's cat, you fix its superposition state as "dead" or "alive"? This is called the collapse of the quantum wave function. Well, when you read a qubit, you're doing the same thing. By fixing its state, it becomes just like a bit in an ordinary computer, either 1 or 0, which makes the entire exponential computing power meaningless. Therefore, the computer needs to run an algorithm that uses quantum wave interference before reading to filter the qubits into useful states; then, when we read it, we can get a useful answer and take advantage of this exponential computing power.

However, it is extremely difficult to find such algorithms, and they can only be used to solve very specific multi - node complex tasks.

We already have some algorithms that can factorize numbers or simulate quantum physics, but that's it. We haven't found any algorithms suitable for chemical simulations or neural networks that drive artificial intelligence, and many researchers even think that the relevant algorithms for these applications may not exist at all. They point out that the data types used for artificial intelligence training are very unstructured, and the mathematical problems actually calculated during the training process are not suitable for quantum computers.

Therefore, even if technology giants can accelerate the delivery of truly usable quantum computers, current science shows that they won't help artificial intelligence at all. In fact, we've found so few quantum algorithms that most of the potential promised by this technology seems completely unrealistic.

If this topic makes your head spin - and let's face it, this is quantum physics, which is always extremely difficult - there's a channel called "Looking Glass Universe" run by a quantum computing doctoral graduate. This video does a great job of explaining how quantum computers work and their limitations.

It's another farce full of hype but lacking in substance.

Even the claim that our brains are quantum computers - the idea that sparked the entire quantum artificial intelligence movement - has basically been disproven by recent research.

But it doesn't matter. Reality no longer matters. This concept has become the trend of the times. Misinformation about what quantum computers can do has spread and is in full swing. And those fools are ready to commodify this concept.

If the technology giants and their supporters can inflate the quantum computer bubble quickly enough, all they're doing is delaying the burst of the artificial intelligence bubble. Sooner or later, after the promised returns fail to materialize, the hype will subside, and the trend of the times will start to align with reality. The hundreds of billions of dollars - which should have been used to raise workers' wages to at least keep up with inflation but were siphoned off into this bottomless pit - will disappear without a trace, and our efforts will yield almost nothing. This is sad and lamentable, and it will hurt all of us, except for those billionaire elites at the top, because they will have withdrawn their money long before everything completely collapses.

Translator: boxi.