How far will the Artificial Intelligence Bubble go?

This article was originally written in Brazilian Portuguese

The truth about the state of the current market, to be honest:

We have been navigating for a few years in a market that is not profitable , at a completely unsustainable speed and without any real prospect of revolution. There are no truly transformative products that people are rushing to buy: all current products are being subsidized by Venture Capital.

There is no plausible justification for consume all the water on the planet when there is not enough profit or real benefit to compensate for the amount of money being thrown at startups. This level of irresponsibility and risk has never happened! All the latest updates have been incremental and not much different from what we have been seeing in the last 3 years!

And yet, during the week of SXSW 2025 , we were forced to listen to Dario Amodei , Anthropic's CEO, saying:

"I think we'll be there in 3 to 6 months where AI will write 90% of the code. And in 12 months, we could be in a world where AI will basically write all the code possible" - HOW?!

Anthropic will burn more than $3 billion by 2025, after having already spent $5 billion in 2024, and the model is being challenged by more promising Reinforcement Learning strategies, such as Chinese alternatives . There is no evidence so far that the upcoming models will be revolutionary, and the best we can hope for is increased productivity and possibly better benchmark results. This year, the company projects to earn $1 billion, with no plan for profitability. Assuming that investors expect gigantic multiples, how does this account add up?! Are we all dreaming?! How does it work?

At the end of February, Microsoft scaled back its plans to expand computing capacity in more datacenters , reducing its target by more than 100 megawatts of power. To give you an idea, that's equivalent to all the datacenters in Tokyo and London. Doesn't sound like much? It must be much more. The entire leasing process has been reduced: and that's a big deal!

What does this signal? Despite still planning to spend $80 billion on AI this fiscal year , one of the main market leaders is signaling that Computing Power will not grow in the coming years. Even though it has already declared 20 billion in additional spending on AI, we can expect the company to push the platform across all its product lines. In fact, this past week during the MVP Summit, even Vibe Coding is gaining attention in the company's presentations 🤦♂️! What could go wrong?!

To be fair, it could also mean that MS overestimated consumption, and to avoid future expenses, is readjusting. In the end, the message is the same:

In a market that expects accelerated growth, if it doesn't multiply revenue, it's no good.

As much as I want to believe in a promising future in the AI market, I only see misfortune. But for now I will focus on what I believe is the biggest driver of market destruction in the coming years.

When the Bubble bursts, will we hear the noise?!

As more information emerges about Project Stargate , which absurdly proposes to invest $500 billion in the construction of dantesque data centers in Texas, it becomes more difficult to believe in such an absurd number, that not even the companies involved are able to organize a coherent explanation.

• The first expectation is to invest 100 billion to produce 1 Gigawatt of energy and hundreds of thousands of GPUs

• What exists today in Texas is a single contract, which should come into operation in 2026, and evolve in phases until it reaches expectations.

For this first stage to happen, the project will build a Natural Gas Plant with a capacity of 360 Megawatts. The project is still in the stage of obtaining approval to begin construction. If there is one thing we know, it is that every construction project is successful and completed on time. Right?

With Microsoft taking its foot off the gas in terms of building more datacenters, everything is up to these two companies and their ability to build datacenters - which, let's face it, neither of them has mastered .

Both OpenAI and SoftBank have pledged $18 billion each to start the Stargate project. The problem? That money doesn’t exist yet!

OpenAi is currently seeking $40 billion in funding at a valuation of $260 billion, with part of that money going to fund Stargate. SoftBank is managing the entire funding round , even announcing that it was close to closing the round. This is much less than what was thought at the end of last year ($360 billion).

There's just one detail: apparently, SoftBank doesn't have the money either! 🤡

Masayoshi Son, CEO and famed investor of WeWork, that decade-old craze , has announced plans to borrow an additional $16 billion to invest in AI , and an additional $8 billion next year, on top of the $18 billion already committed to the Stargate project. Confused? Well, it’s about to get more.

SoftBank is huge, but it’s not $40 billion in cash on hand. In fact, according to investor reports, it will need to borrow money to fund OpenAI and also its acquisition of Ampere, a chip design company . The entire premise of the Stargate project is based on SoftBank’s ability to:

• First get $40 billion, which will come with 90% of the money borrowed from banks and insurance companies

• Make the remaining $450 billion viable

How does this stand? How does it work? To raise this money, in addition to borrowing, SoftBank will have to sell stakes in some of its most valuable positions, which is a risky path. In this situation, SoftBank will lead the future of OpenAI, and all the processing potential would migrate from Microsoft to Stargate.

I believe the market has become desensitized to this type of investment: 50 billion dollars in cash to invest in building construction, network cabling and power plants? And that is ONLY 10% of the investment projection? For what product? Expecting a return of 10x or more? How does it work?!

To arrive at the absurd calculation of how much SoftBank is proposing to spend:

• Stargate's initial $18 Billion

• $3 Billion in OpenAI Product Revenue

• $30 Billion to OpenAI paid in 12 and 24 months (on top of the $40B OpenAI needs)

Thus, OpenAI depends on SoftBank to survive, while at the same time it depends on Stargate to cover the growth demand, which is also dependent on SoftBank's ability to raise investment which in itself needs to borrow money to give it back to OpenAI: WTF ?!

All this with a technology where the only certainty is that next year it will need more money to invest in model training and computing power?! With no expectation of return?!

OpenAi projects to burn $10 billion by 2025. Assuming OpenAI doesn't expect to become profitable until 2029, and that they project they will burn $40 billion per year by 2028 : HOW DOES IT WORK?!

We are facing the biggest scam of our time, with a company that hopes to raise more money than any other startup has ever raised in history, and as Ed Zitron said :

"It's better to set fire to that amount of cash, because at least it generates heat!"

Nothing compares to what we are seeing right now. And even with OpenAI's vision of focusing on applications using its platform, no "killer" product has emerged:

• Deep Research is mediocre

• Automator is expensive and horrible

• GPTs have not gained traction and Agent Store is sinking

• GPT 4.5 is a joke, and is far from the promised performance leap

Even in China, there are already indications that the investment in computing infrastructure may have been exaggerated, and without proof of truly innovative and useful products, all the investment may have been in vain.

I've already said here that in this gold rush, the ones who make money are the ones who sell shovels.

Apparently, the deposit discovered in 2020 is starting to dry up, and all we can do is wait: in the end, all that will be left is scorched earth.