Artificial intelligence has reached that familiar stage in the technology cycle where excitement, fear, money, and confusion all collide at once. Headlines swing wildly between “AI will replace everyone” and “AI is just glorified autocomplete.” Valuations soar, startups multiply overnight, and critics begin to push back—hard. When someone as influential as Linus Torvalds suggests that AI looks like a bubble, the conversation shifts from internet hype to something more serious.
This article isn’t here to sell you on AI as salvation, nor to dismiss it as a scam. The goal is to do something rarer: slow down, zoom out, and look at AI the way an engineer, a historian, and a long-term thinker would. If you’re a builder, a curious observer, or even someone trying to make sense of where to place their attention (or capital), there’s real value in understanding where the bubble narrative comes from, where it’s wrong, and where it’s actually justified.
Why the “AI Bubble” Conversation Is Even Happening
No one calls something a bubble unless three things are present at the same time: rapid growth, massive capital inflows, and exaggerated expectations. AI checks all three boxes.
Over the past few years, artificial intelligence has gone from a niche academic and enterprise topic to a mainstream cultural force. Large language models, image generators, voice systems, and autonomous agents are suddenly accessible to anyone with a browser. At the same time, trillions of dollars in combined market value have flowed into companies connected to AI infrastructure, chips, cloud services, and software.
That kind of acceleration always attracts skeptics—and rightly so.
When Linus Torvalds comments that AI feels bubble-like, he’s not saying it’s useless. He’s reacting to the disconnect between what the technology actually does and what people claim it will do. Torvalds comes from a world of deterministic systems, where you understand how things work down to the metal. Modern AI, by contrast, is probabilistic, opaque, and often poorly explained to non-experts. That alone is enough to make serious engineers uneasy.
The Engineering Perspective: Why AI Feels “Off” to Builders
One reason AI triggers skepticism among veteran engineers is that it breaks traditional intuitions about software. In classic programming, you write rules, test edge cases, and debug failures. With modern AI models, especially large language models, behavior emerges statistically rather than deterministically. You don’t always know why a model failed—only that it did.
To someone like Torvalds, that feels less like engineering and more like controlled chaos.
Add to that the way AI is marketed. Tools are described as if they “understand,” “reason,” or “think,” even though these are metaphors layered on top of pattern recognition systems. The marketing language gets far ahead of the technical reality, and that gap is fertile ground for bubble accusations.
There’s also a practical concern: much of today’s AI ecosystem is built on top of a small number of foundation models. Thousands of startups are effectively renting intelligence from the same sources, building thin layers on top, and racing to scale before pricing, policies, or competition change. From an engineering standpoint, that’s a fragile ecosystem.
Fragility doesn’t mean useless—but it does mean correction is inevitable.
The Money Loop: Is AI Just Capital Chasing Itself?
One of the most common arguments for the “AI is a bubble” thesis comes from the financial side. You’ve probably seen the memes: NVIDIA sells GPUs to AI companies, AI companies raise venture capital to buy more GPUs, cloud providers profit, valuations rise, and the cycle repeats. On the surface, it can look like a closed loop where money generates more money without producing proportional real-world value.
That criticism isn’t entirely wrong—but it’s incomplete.
What’s different about AI compared to many past bubbles is that the money is flowing into infrastructure. GPUs, data centers, networking hardware, and power systems are tangible assets. They don’t disappear when hype fades. Even if dozens of AI startups fail, the underlying compute capacity remains and gets repurposed.
This is a key distinction. During the dot-com bubble, much of the money went into marketing and speculative companies with no infrastructure moat. In the AI era, a significant portion of capital is being sunk into the physical backbone of future computation. That changes the long-term equation.
Where the AI Bubble Argument Is Actually Correct
Being honest matters, so let’s be clear: parts of AI are absolutely overheated.
There are too many startups doing nearly identical things. There are products with no defensible advantage beyond “we used AI.” There are valuation models based on future dominance rather than present fundamentals. And there are bold claims—about replacing entire professions or achieving general intelligence—that are not grounded in current capabilities.
These areas are almost guaranteed to correct.
You can expect consolidation. You can expect many AI tools to disappear quietly. You can expect pricing pressure, regulation, and customer skepticism to weed out weak players. Anyone pretending otherwise is either inexperienced or selling something.
This is the “bubble” part—and it’s real.
Why AI Is Not a Fake Technology Bubble
Now for the other half of the truth.
AI is already delivering measurable, sustained value across multiple industries. Developers use it to write and refactor code faster. Analysts use it to process documents and data at scale. Designers prototype faster. Researchers accelerate experimentation. Businesses reduce friction in customer support, logistics, and planning.
These are not speculative use cases. They’re operational realities.
Another sign that AI is not a hollow bubble is who is investing in it. Governments, militaries, universities, and critical infrastructure providers are all building AI capabilities. These institutions don’t chase trends lightly. When they commit resources, it’s because the technology has strategic importance, not just short-term hype.
AI is also becoming embedded rather than standalone. Instead of being a product you buy once, it’s becoming a layer inside everything else—search engines, operating systems, development tools, enterprise software. Technologies that embed themselves tend to survive hype cycles because they become invisible utilities rather than flashy products.
A Better Mental Model: AI as a Stack, Not a Single Thing
One of the biggest mistakes in the AI bubble debate is treating AI as one monolithic entity. It’s not. It’s a stack, and each layer behaves differently.
At the top is the hype layer: marketing, headlines, social media, and exaggerated promises. This layer is noisy, unstable, and highly bubble-prone.
Below that is the application layer: startups, tools, and platforms built on top of models. This layer is overcrowded and will thin out significantly over time.
Below that is the model layer: foundation models, training techniques, and algorithms. This layer evolves more slowly and benefits from cumulative progress.
At the base is the infrastructure layer: chips, data centers, energy, and networking. This layer is capital-intensive, slow-moving, and extremely durable.
When people say “AI is a bubble,” they’re usually talking about the top two layers. When people say “AI is the future,” they’re usually talking about the bottom two. Both statements can be true at the same time.
What This Means for Long-Term Thinking (Not Advice)
It’s important to be clear: this isn’t investment advice. But understanding patterns helps anyone make better decisions—whether that’s where to work, what to build, or what to learn.
Historically, the biggest winners after a tech bubble are not the loudest players during the hype. They’re the ones quietly building infrastructure, tooling, or deeply integrated solutions that remain useful when attention shifts elsewhere.
In AI, that likely means:
- Companies with proprietary data
- Vertical-specific applications that solve real problems
- Tools that reduce costs rather than add novelty
- Infrastructure and enablement layers
- Open ecosystems rather than closed gimmicks
The riskiest place to stand is at the intersection of hype and thin differentiation.
So, Is AI a Bubble?
Here’s the most honest answer available:
AI is experiencing a hype bubble, not a technology bubble.
The expectations will deflate. The narratives will calm down. Many companies will fail. But the underlying capability—the ability for machines to assist, automate, and augment cognitive work—is not going away. If anything, it’s becoming more deeply embedded in how society functions.
People like Linus Torvalds are performing an important role by pushing back against nonsense. Skepticism forces clarity. But history suggests that when a technology both attracts hype and delivers real utility, the outcome is not collapse—it’s normalization.
The hype dies. The tech stays. And the world quietly reorganizes around it.
That’s not a bubble bursting. That’s a system growing up.