Finance
Technology
AI

The AI Bubble Is Starting to Crack: Profit Myths, Market Risks & What Happens Next 

Welcome back. If you read Part 1, you already know we’re not here to dismiss AI as a technology — it’s genuinely impressive and genuinely useful. But there’s a growing disconnect between what AI companies are worth and what they’re currently earning, and that disconnect deserves serious attention.

In Part 2, we’re going deeper. We’re going to talk about the accounting questions that serious analysts are raising, what happens to the broader economy when market sentiment on AI shifts, and — maybe most importantly — what practical steps you can take right now to protect yourself while still participating in the AI opportunity.

The AI bubble is starting to crack. Not from the outside, but from within — from the financial engineering, the overstated revenues, the understated costs, and the circular money flows that make today’s AI economy look more fragile than the headlines suggest.

Let’s get into it.

The Accounting Questions Wall Street Doesn’t Want to Discuss

Here’s something that doesn’t get nearly enough mainstream coverage. While AI firms have been enthusiastic about telling the world about their growing revenues and massive valuations, some serious investors have started digging into the actual books — and what they’re finding raises legitimate concerns.

The Depreciation Problem

AI companies are investing enormous sums in specialized hardware — primarily high-end GPUs and custom chips. These chips don’t last forever. They depreciate. They become obsolete. And the question of how fast they depreciate matters enormously for calculating true profitability.

Prominent investor Michael Burry — yes, the Big Short guy — has made claims that AI firms are significantly understating the depreciation of their existing chips. According to his analysis, the understatement could amount to around $176 billion between 2026 and 2028 across the industry. If accurate, that’s not a rounding error. That’s a fundamental misrepresentation of profitability that would make many AI companies look far less financially healthy than their reported numbers suggest.

Now, Burry has been wrong before and his timing calls are notoriously hard to pin down. But the underlying question he’s raising — are AI infrastructure assets being depreciated fast enough given how quickly the technology is evolving? — is a completely legitimate accounting question that deserves real scrutiny.

The Revenue Inflation Problem

There’s a practice in corporate reporting called annualizing revenue. If a company earns $100 million in a single quarter, they might present that as “$400 million annual run rate” in investor materials. That’s not inherently dishonest — it’s a standard way of projecting trajectory. But it becomes misleading when the revenue being annualized is experimental, one-time, or unlikely to repeat at the same level.

As Business Insider has noted, a significant portion of current AI revenue falls into exactly this category — experimental projects, pilot programs, and corporate trials that haven’t yet committed to long-term contracts. Presenting this as reliable recurring revenue gives a distorted picture of the business model’s actual sustainability.

When you combine understated depreciation with overstated revenue annualization, the gap between reported financial health and actual financial health can become quite large. And the AI bubble is starting to crack partly because sophisticated investors are starting to close that gap in their mental models.

The Circular Money Problem Nobody’s Naming Clearly

Here’s a thought experiment. Imagine Company A buys AI services from Company B. Company B uses that revenue to buy more chips from Company C. Company C uses that revenue to expand its AI research and buy services from Company A. Everyone’s revenue is growing. Everyone’s reporting growth. The whole system looks healthy.

But where is the value actually being created for end customers? Where is the external, real-world demand that justifies this circular flow?

This is essentially the concern that Nvidia’s critics have raised — that a meaningful chunk of AI spending is tech companies buying from other tech companies, with the ultimate customer value creation still very much in question. Nvidia has denied this characterization, and there’s clearly genuine external demand from businesses genuinely trying to use AI. But the circularity concern isn’t entirely wrong either.

Nvidia’s fortunes also depend heavily on a small number of massive customers — the major cloud providers and AI labs. If even one or two of those customers slows their spending, develops competing chips in-house (which Google, Amazon, and Microsoft are all actively doing), or simply decides they’ve built enough capacity for now, Nvidia’s growth story changes dramatically.

Broadcom’s recent results illustrated this perfectly. They announced a tripling of AI revenue and a 40% increase in sales — numbers that should have been cause for celebration. Instead, their stock initially fell. Why? Because their profit forecasts came in below expectations, and they warned that Google was actively seeking cheaper chip alternatives. The market saw past the revenue headline to the margin and competitive dynamics underneath.

That’s a sign of a market getting more sophisticated about AI financials. And more sophisticated scrutiny is exactly what happens before a repricing event.

What Happens When Market Sentiment Actually Turns

Let’s be very specific about why this matters beyond just AI investing, because the answer affects everyone.

The so-called “Magnificent Seven” — Apple, Microsoft, Alphabet, Amazon, Meta, Nvidia, and Tesla — account for roughly 30% of the entire US stock market by capitalization. That’s an extraordinary concentration. It means that if AI sentiment shifts and these stocks reprice downward significantly, the ripple effects are not contained to tech investors.

They hit pension funds. They hit 401(k) balances. They affect consumer confidence. They influence corporate investment decisions. They impact the economy as a whole.

This isn’t abstract. A meaningful correction in AI-inflated tech valuations — even one that still leaves AI as a valuable and growing industry — could wipe trillions of dollars off paper wealth that ordinary Americans have built up in their retirement savings. That’s the real-world consequence of overvaluation that doesn’t get talked about enough in the excitement about AI’s transformative potential.

The Margin Debt Warning Sign

One indicator that historically shows up near market peaks deserves its own mention: margin debt. This is money that investors borrow to invest — using their existing portfolio as collateral to buy more stocks on credit.

Margin debt tends to amplify market moves in both directions. When stocks rise, leveraged investors gain more. When stocks fall, leveraged investors get margin calls and are forced to sell — which accelerates the decline, which forces more sales, which accelerates the decline further. It’s a self-reinforcing spiral.

Right now, we’re seeing elevated levels of margin debt, particularly concentrated in the kinds of high-growth tech and AI stocks that have led the market higher. This doesn’t cause a crash on its own, but it does mean that if a correction starts for any reason, it could accelerate faster than it would in a less leveraged market.

Past bubbles — the dotcom crash, the 2008 financial crisis, the 1929 crash — all had elevated margin debt as a contributing factor. It’s not a guarantee of disaster, but it’s a yellow flag worth watching.

The Utility Future: What AI Might Actually Look Like in 5 Years

Here’s something interesting though. Even in the scenario where the AI bubble deflates significantly, the underlying technology doesn’t disappear. What likely changes is the pricing model and who captures the value.

There’s a real possibility — some analysts think a near-certainty — that AI compute eventually gets priced like a utility. The way you pay for electricity or water. You use it, you pay for what you use, at a rate that reflects the actual cost of delivery plus a reasonable margin for the provider.

This would be profoundly different from today’s model, where AI is priced partly as a service, partly as a subscription, partly below cost as a customer acquisition strategy, and partly in ways nobody fully understands. A utility model would bring transparency, price competition, and commoditization.

In that world, the companies that win aren’t necessarily the ones building the biggest language models. They might be the ones who own the most efficient delivery infrastructure, the cheapest compute, or the most valuable proprietary data. The application layer would look very different — more specialized, more focused on genuine workflow integration, and far less tolerant of AI tools that are impressive but don’t actually move the needle on real business outcomes.

Pro Tips for Navigating the Current AI Landscape

For Investors

• Look past revenue to free cash flow. Revenue can be dressed up in many ways. Free cash flow — actual money coming in minus actual money going out — is much harder to inflate artificially.
• Ask where the recurring revenue comes from. Is it contracted, multi-year enterprise agreements? Or is it experimental pilots and monthly-cancel subscriptions? The quality of revenue matters as much as the quantity.
• Track customer concentration risk. If a company’s top three customers represent more than 40% of revenue, ask what happens if any one of them pulls back.
• Diversify beyond the Magnificent Seven. AI will reshape many industries. Healthcare, logistics, manufacturing, agriculture — the long-term AI winners may not be the companies dominating the news today.
• Watch the capex cycle. When hyperscalers start talking about moderating their infrastructure spending, take that seriously. The cascading effect on chip makers and data center operators could be rapid.

For Business Owners and Operators

• Measure actual productivity gains, not potential ones. If you’ve deployed AI tools, do you have rigorous before-and-after data on what changed? Feelings aren’t metrics.
• Build pricing sensitivity into your AI cost models. Assume the AI services you use today could cost 2–3x more in two years. Does your business model still work?
• Don’t build critical infrastructure on a single AI provider. Vendor lock-in is real, and switching costs for deeply integrated AI tools can be significant.
• Focus on use cases where AI genuinely reduces labor cost or improves output quality — not just use cases that look good in board presentations.

The AGI Wildcard: Could Hyper-Optimists Be Right?

It would be dishonest to write a complete analysis of AI without acknowledging the optimist case — and the optimist case for AI is genuinely compelling.

The bull scenario goes like this: current large language models are already impressive, but they’re essentially pattern-matching machines. The next generation of AI — possibly arriving in the next 3–5 years — could achieve something closer to artificial general intelligence (AGI): systems capable of genuine reasoning, creative problem-solving, and autonomous action across a wide range of tasks.

If that happens, the economics change completely. A system capable of doing complex knowledge work autonomously isn’t a cost center — it’s a revenue-generating asset that scales without hiring. The productivity gains wouldn’t be incremental; they’d be exponential. And the current valuations, which look stretched against today’s AI capabilities, might actually look cheap against AGI-era capabilities.

This isn’t science fiction — serious researchers at serious institutions are working on these problems. But it’s also not here yet. And “it might arrive in 3–5 years” has been said about transformative AI breakthroughs for decades. The timeline risk is real.

Betting your portfolio on AGI arriving on schedule is a high-variance bet. It might pay off spectacularly. It might be a very expensive wait.

Practical Action Plan: What to Do Right Now

Okay, enough analysis. Let’s get concrete about what this all means for actual decisions you might want to make.

Common Mistakes in This Market (Part 2 Edition)

Mistake #6: Dismissing the bears entirely

The optimists aren’t wrong that AI is transformative. But dismissing concerns about valuation as “you just don’t understand the technology” is intellectually lazy. Plenty of dotcom-era technology was transformative. That didn’t protect investors who paid peak 1999 prices.

Mistake #7: Assuming the AI narrative will keep working indefinitely

Narratives drive markets — until they don’t. The moment a few major AI earnings reports disappoint significantly, the narrative can shift with surprising speed. Don’t mistake “the story is still working” for “the story will always work.”

Mistake #8: Not separating AI hype from AI use

You can be highly enthusiastic about using AI in your daily work — and you should be, because some of it is genuinely great — while also being skeptical about current market valuations. These two positions are completely compatible. In fact, they’re the most intellectually honest position to hold right now.

Mistake #9: Ignoring the concentration of AI gains

Most of the AI value creation so far has gone to a tiny number of companies and their shareholders. The productivity gains that AI proponents promise for the broad economy haven’t materialized at scale yet. Broad-economy effects may come eventually, but they haven’t arrived yet — and “eventually” doesn’t pay today’s valuations.

FAQs: Part 2 Deep Dives

Conclusion: Informed Optimism Over Blind Enthusiasm

Let me bring this home with the honest version of where I think we are.

AI is real. It’s useful. Some of it is genuinely impressive. The technology will continue to develop and will likely reshape significant parts of how we work and live over the coming decade. None of that is in dispute.

But the AI bubble is starting to crack under the weight of valuations that price perfection, costs that are being underreported, revenues that are being overstated, and a circular investment dynamic that looks sustainable on a spreadsheet but may not survive contact with a growth slowdown or a rising cost of capital.

The internet survived the dotcom crash. AI will survive a potential AI correction. The question that matters for your financial wellbeing isn’t whether AI has a future — it does — but whether the prices you’re paying today accurately reflect the realistic timeline for that future to arrive.

Informed optimism is the right stance. Eyes open. Excited about the genuine possibilities. Skeptical about the financial engineering. Prepared for multiple scenarios. Diversified enough to survive the bad ones.

That’s not pessimism about AI. That’s wisdom about markets.

• Audit AI accounting carefully — depreciation and revenue quality matter
• Watch hyperscaler capex guidance — it’s the leading indicator for the whole ecosystem
• Build pricing sensitivity into every business model relying on AI APIs
• Diversify your portfolio beyond AI concentration
• Stay curious — AGI might still surprise us all on the upside
• Keep using AI for real work — just stop confusing useful with correctly priced

Thanks for reading both parts. If this was useful, share it with someone who needs a reality check — or with someone who thinks you’re being too pessimistic about AI. Either conversation is worth having.

Indian Navy INS Vikramaditya : India’s Floating City of Power — Everything You Need to Know

What Is Quantum Mechanics ? A Beginner’s Guide to the Weirdest Science Ever

OLED vs Mini LED in 2026 : Which Should You Actually Buy ?