The Real AI Bubble
Some say artificial intelligence is a bubble. But the problem is that they throw everything into the same pot. They confuse the ones who build it with the ones who just play around it.
Stay with me till the end, because the difference is huge.
By 2025, spending on model training will exceed 400 billion dollars. An unprecedented level that, for some, recalls the Internet rush of the late 1990s. But it’s not the same story. Back then, growth was inflated by circular loans and fake invoices, like in the WorldCom case. Today we’re talking about real infrastructure: data centers, multimodal models, physical GPUs that cost money and actually work.
Of course, even here in the United States, some go too far. Startups calling themselves “AI companies” when they just stick a prompt on a user interface. Apps built overnight with vibe coding and borrowed APIs. That’s the real bubble. A bubble made of useless apps, inflated pitches, and capital that disappears after a month.
But the models, no. The models are the base, the cognitive infrastructure of this new decade. OpenAI, Anthropic, Google, and Meta are spending huge sums not for hype, but for computing power, efficiency, training, and scale. These aren’t speculative numbers. They’re digital factories.
Today ChatGPT has about 800 million weekly users, and 5 percent pay for the premium version. Just like in the early Internet, when most people browsed for free, the value was in the network, not in the single subscription.
The bubble might burst. But it won’t burst in the models. It’ll burst above them, in the startups selling smoke…
#ArtificialDecisions #MCC #AI