Analyst Warns AI Bubble Dwarfs Dot-Com Crisis by Seventeen Times

Let's talk about this AI bubble analysis from Julien Garran, because whether he's right or catastrophically wrong, the implications are massive for every technology executive listening to this.

Here's what makes this bubble claim technically interesting. Garran isn't just looking at stock prices—he's analyzing the fundamental architecture of how AI companies generate value, or more accurately, how they're failing to. The core technical problem is this: we've built an ecosystem where large language models require massive computational infrastructure, but we haven't cracked the code on applications that generate revenue exceeding their operational costs.

Think about it from an engineering perspective. When you're running inference on these models, you're consuming significant GPU resources for every query. The cost structure doesn't scale the way traditional software does.

A SaaS company can add users with minimal marginal cost, but AI applications are burning compute with every interaction. That's why companies like OpenAI are reportedly losing money on ChatGPT despite having hundreds of millions of users. The technical bet everyone's making is that we'll achieve either massive efficiency gains—we're talking orders of magnitude improvements in inference costs—or we'll discover that killer application that justifies premium pricing.

Garran's argument is that neither is materializing fast enough. He specifically claims there will never be a commercially successful app built on the back of LLMs, which is a nuclear take, but his reasoning centers on the fact that LLMs are best at tasks where accuracy doesn't matter much, what he calls "bullshit jobs." The technical limitation he's pointing to is the hallucination problem and the inability of current architectures to reliably handle high-stakes decisions.

Now let's follow the money, because this is where things get genuinely concerning. Garran cites nearly one trillion dollars in market value creation across ten AI startups with zero profits. That's not theoretical—that's real capital being deployed based on future expectations.

But here's the structural problem: venture capital funding for AI startups is already drying up because valuations have become absurd. When you're talking about companies raising at billion-dollar valuations before they've proven product-market fit, you've created a funding gap that only the largest players can bridge. Who's left writing the big checks?

You've got SoftBank, foreign sovereign wealth funds, and NVIDIA—which has its own strategic reasons for keeping the ecosystem alive since they're selling the picks and shovels. But this creates a house of cards scenario. If one major player pulls back, the entire funding cascade could collapse.

Compare this to the dot-com bubble, which was seventeen times smaller according to Garran's analysis. The dot-com bubble was largely driven by retail investor speculation in public markets. This AI bubble is different—it's institutionalized through corporate balance sheets.

Microsoft, Google, Amazon, and Meta are each spending tens of billions on AI infrastructure. If the return on investment doesn't materialize, these aren't failed startups we're talking about—these are write-downs on Fortune 100 balance sheets that will ripple through the entire economy. The revenue model problem is stark.

OpenAI reportedly needs to reach forty billion in revenue just to justify its current valuation trajectory. They're currently around three to four billion. That's not a growth curve—that's a vertical cliff they need to climb.

And they're supposed to be the success story.

The disruption angle here is fascinating because it cuts both ways. If Garran is right and we're in a bubble, the disruption isn't what the AI will do to other industries—it's what the bubble bursting will do to tech itself. We're looking at a potential reshuffling of the entire venture capital ecosystem, a massive correction in tech valuations, and potentially a lost decade for AI innovation as funding becomes scarce and skepticism becomes the default position.

But let's steel-man the counter-argument, because there's a legitimate bull case here. The bears said the same thing about cloud computing in the early 2000s, about mobile apps in 2009, about crypto, about every major platform shift. Sometimes markets correctly price in future value creation that isn't immediately obvious.

The difference is that those platforms found their killer apps relatively quickly. Cloud computing had AWS and enterprise migration. Mobile had the App Store ecosystem.

AI is still searching. The competitive dynamics are also unusual. Normally in a bubble, you see massive fragmentation with hundreds of competitors.

In AI, we're seeing rapid consolidation. The foundation model layer is dominated by maybe five serious players. The application layer is fragmented, but increasingly, those apps are just wrappers around the same underlying models.

This creates a strange market structure where the companies burning the most cash—the foundation model builders—are the ones with the most defensible positions, while the potentially profitable application layer has no moat.

Here's where this gets really interesting from a societal perspective. If we are in a bubble and it bursts, the cultural backlash against AI could be severe and lasting. We've already seen the pattern with crypto—a boom-bust cycle that left the entire technology culturally tainted, making it harder for legitimate use cases to gain traction.

An AI bust could create a similar effect, but with broader consequences because AI has been positioned as the solution to everything from climate change to disease research. The social contract around AI is already fragile. We've got artists and writers protesting their work being used for training data.

We've got concerns about job displacement. We've got regulatory pressure building globally. A bubble burst could accelerate the backlash, leading to stricter regulations that make future innovation harder, not easier.

But there's another cultural angle Michael Buckley raised in one of the linked pieces—the fear of missing out on AI is currently overshadowing the fear of losing our humanity. That's a profound observation. Right now, every company feels compelled to have an AI strategy because they're more afraid of being left behind than they are of implementing technology they don't fully understand or trust.

If the bubble bursts, that calculation flips. Suddenly, the fear of having wasted resources on AI becomes dominant, and we could see a massive pullback in adoption even for legitimate use cases. The Channel 4 documentary with the AI host is a perfect example of this tension.

It's simultaneously a demonstration of technological capability and a cultural inflection point. The reaction to it will tell us a lot about whether society is ready to accept AI in roles we previously considered inherently human, or whether we're going to draw harder lines about where AI belongs.

So what do you actually do if you're a technology executive processing all of this? Here are three concrete actions to consider. First, audit your AI investments with a focus on time-to-value.

If you're making significant AI infrastructure investments, you need a clear thesis on when those investments generate returns. Not "eventually" or "when the technology matures"—specific milestones and dates. Garran's analysis suggests the market is running out of patience for long-term bets.

If your AI initiatives can't show meaningful ROI within eighteen to twenty-four months, you need to seriously reconsider the scale of investment. This doesn't mean abandon AI—it means be surgical about where you deploy resources. Focus on use cases with clear efficiency gains or revenue generation, not exploratory research projects that might pay off in five years.

Second, diversify your AI strategy beyond LLMs. This is critical. If Garran is even partially right that LLMs won't produce commercially successful standalone applications, then betting your entire AI strategy on GPT-4 wrappers is existential risk.

Look at specialized models, look at traditional machine learning for specific use cases, look at hybrid approaches. Google's quantum computing breakthrough is a reminder that the AI story is broader than just large language models. The companies that survive a potential bubble burst will be the ones that weren't over-indexed on a single technological approach.

Third, build optionality into your AI vendor relationships. Right now, many companies are getting locked into specific platforms and providers. Given the financial instability Garran is highlighting, you need strategies that allow you to pivot if your primary AI vendor gets acquired, pivots their business model, or simply disappears.

This means maintaining in-house expertise, using open-source models where possible, and ensuring your data and workflows aren't completely dependent on any single provider's API. The next twelve to twenty-four months could see significant consolidation in the AI space, and you don't want to be the customer left stranded when your vendor gets absorbed or shut down. The meta-lesson here is about narrative versus fundamentals.

The AI narrative is incredibly compelling—we're building machines that can think, reason, and create. But narratives don't pay salaries or satisfy shareholders. At some point, the technology needs to generate value that exceeds its cost.

Whether that happens before or after a market correction will define the next decade of technology development. As an executive, your job is to position your organization to benefit from AI's genuine capabilities while being resilient enough to survive if Garran's prophecy comes true.