AI Bubble or Not? Why Your Investment Decision Should Be Independent of It

Peter Berezin says the party is nearly over. Juan Correa says it's only just getting started. Both sit at BCA Research, both are regarded as top strategists, both look at the same numbers - and reach opposite conclusions. This open dispute within a single research house is perhaps the most honest signal the financial world currently offers on the question "AI bubble or not?": nobody really knows. And that's exactly why the question is the wrong one for your investment decision in the Mittelstand.

Wall Street is arguing publicly - with itself

At BCA Research, two camps collide head-on. On one side, the bears, led by Peter Berezin and Arthur Budaghyan. Their core argument: some of the spectacular profits in the AI ecosystem are artificially inflated. When a chipmaker sells to a cloud provider, it shows up as revenue for the seller and as investment for the buyer - a loop that looks impressive on paper without the real cash flow necessarily keeping pace. The bears' actual point is subtler than it first sounds: AI makes companies more efficient, they say, but not necessarily more profitable - because once every market participant has access to the same technology, the competitive edge individual companies hope for from it evaporates.

On the other side stand the bulls, Juan Correa and Noah Weisberger, with a solid counter-argument: there's no oversupply of compute at all - quite the opposite. Rental rates for the latest chip generations have doubled, and the order backlogs of the major cloud providers grew by around $750 billion in just two quarters, according to the BCA bulls. This figure comes from one market party's argument and isn't an independently audited metric - but it clearly shows the direction of the bull case: if demand structurally exceeds supply, that's closer to the opposite of a bubble.

JPMorgan leans toward the bullish side: the bank expects chip stocks to hit new record highs in the second half of 2026, because new manufacturing capacity is barely securable before 2028, and chronic supply scarcity gives manufacturers lasting pricing power.

Regardless of which camp you lean toward, there are facts nobody can get around. The broad US market traded at a price-to-earnings ratio of around 27, according to reports - above the dotcom peak of 26.5 from March 2000. That's not proof of a bubble, but it's not a footnote either.

At the same time, the four big hyperscalers - Amazon, Microsoft, Google/Alphabet, and Meta - are planning combined capital expenditures of around $725 billion for 2026, according to market analyses. That would be an increase of about 77 percent over roughly $410 billion in 2025. The individual figures vary considerably depending on the source and the timing of guidance updates, but the order of magnitude is clear: this is investment at a pace that's historically rare.

According to press reports - including Bitget News - a confidential draft report from the US Treasury Department is also said to exist, warning of a dotcom-like shock should AI monetization fail to materialize as hoped, with spillover effects on data centers, the private credit market, and chipmakers. This report has not been officially confirmed or published. It belongs in the category of rumor, not fact - but it shows that concern about overexuberance is being discussed in official circles too, even if it can't be verified.

The key insight from the Wall Street debate isn't who ultimately turns out to be right - it's that even the sharpest minds in finance reach opposite conclusions from the same data. If you're a Mittelstand business making your AI investment dependent on this dispute, you're waiting for an answer that won't arrive anytime soon.

Why the valuation question isn't your question

Here lies the crucial mistake many entrepreneurs in the Mittelstand fall into: they confuse the question "Is the stock market valuation of Nvidia, Microsoft, or OpenAI sustainable?" with the question "Is an AI investment worth it for my company?" These are two completely independent questions. The causal chain between Wall Street sentiment and your own business case is weak to nonexistent.

Even if it turns out in a few years that the valuations of the big AI providers were overblown and a correction follows, the technology itself doesn't disappear as a result. Just as the bursting of the dotcom bubble in 2000 didn't abolish the internet again. Amazon, Google, and eBay existed before the bubble, during the bubble, and after it - only the stock market valuations corrected. What matters for your company isn't whether Nvidia is still trading at a P/E of 40 in five years, but whether the AI agent you introduce today demonstrably saves time or generates revenue in three months.

The real danger: 95 percent with no measurable ROI

The number you should take seriously instead doesn't come from the stock market, but from practice. The MIT NANDA study "State of AI in Business" from 2025, which continues to be widely cited in 2026, reaches a sobering conclusion: despite global investments of $30 to $40 billion in generative AI, around 95 percent of the enterprise pilot projects studied achieve no measurable ROI. Only about 5 percent extract significant value.

That sounds at first like an argument against AI investment altogether. It isn't. The study also delivers the decisive lever that determines success or failure: externally purchased specialist solutions with specialized partners achieved a success rate of around 67 percent - compared to only about 33 percent for in-house developments built by the companies themselves. In other words: it's not the question "AI, yes or no" that determines ROI, but the question "bought or built in-house." For the Mittelstand, which rarely has its own AI development teams, that clearly favors the path through specialized partnerships over trying to build everything in-house in the early phase.

Bubble-resistant decision criteria

From the Wall Street debate itself, three criteria can be derived that make your AI investment independent of whether Berezin or Correa ultimately turns out to be the better analyst:

First: short payback period instead of a long-term market bet. If an AI agent pays for itself in months rather than years, it doesn't matter whether the overall market corrects in two years. You've already realized your benefit.

Second: no dependence on a single provider or model. The very compute scarcity that bulls cite as an argument for rising valuations is a risk for you as a user: if you rely on just one provider, you're exposed to price shocks or availability bottlenecks. Flexibility in vendor choice is therefore not a nice-to-have, but a solid risk-management criterion.

Third: clearly measurable benefit from day one instead of investment based on future promises. The gap between the planned $725 billion in hyperscaler investment and the actually demonstrated enterprise benefit is exactly the pattern the bears warn about. For you, that means: don't rely on "the big players are already scaling this up, so it will automatically pay off for us too." Instead, choose your own use cases with clearly quantifiable benefits - automated invoice checking, a customer-service agent with measurable response time, a sales pipeline that demonstrably produces more qualified leads.

These three criteria can be applied as a simple filter before every AI investment decision, whether it concerns a single tool or a multi-year digitalization strategy. Ask yourself the same three questions for every planned project: exactly when does this pay for itself - in weeks, months, or only after years? What happens if the chosen provider doubles its prices or discontinues the service tomorrow - would you then be left without an alternative? And: can you show in four weeks, with a concrete metric, that the investment paid off, or are you relying on a "it'll work out somehow"? If you have a clear, solid answer to all three questions, the Wall Street debate no longer matters for your decision.

Germany: the business case works - but only with strategy

A look at the German reality shows that the question "is AI worth it" has long been empirically answerable - even without looking at Wall Street. According to Bitkom-linked reports, around 41 percent of German companies now actively use AI, a share that has more than doubled compared to the previous year. German AI-using Mittelstand companies achieve 76 percent positive revenue returns according to the studies cited, compared to 46 percent for non-users. Important here: this is a correlation, not proven causation - it's entirely possible that already more successful companies are more likely to adopt AI, rather than AI alone explaining the success. Still, the difference is significant enough not to ignore.

According to an IBM study from October 2025, which continues to be widely cited, 62 percent of German companies report noticeable productivity gains from AI. At the same time, the same data also shows the downsides: 33 percent of the German companies surveyed say AI adoption was more expensive than planned, and 43 percent have no concrete AI strategy so far. This lines up exactly with what the MIT study shows globally - success doesn't hinge on the technology itself, but on strategy, cost expectation management, and choosing the right implementation path.

When it comes to expected returns, a closer look is worthwhile, because two numbers are circulating here that seem contradictory at first glance. One survey cites an expected average ROI for German companies in 2026 of around 24 percent, up from 17 percent the year before. Another source cites an average ROI of 210 percent within 24 months for successful AI projects. These two figures come from different studies with presumably different methodology and sample sizes and aren't directly comparable - one presumably measures the average across all projects, the other only the successful ones. But that difference is precisely the real message: the gap between an "average" and a "successful" project is huge, and it comes from the quality of execution, not from market conditions.

While analysts argue over valuations, the big AI providers keep expanding briskly. Anthropic, for instance, raised its US lobbying spend in the second quarter of 2026 to around $1.97 million, up 26 percent quarter over quarter, now surpassing Nvidia in lobbying spend while closing in on Oracle. Together with Blackstone and Hellman & Friedman, Anthropic has also formed the joint venture "Ode With Anthropic," a $1.5 billion undertaking already working with around 100 deployed engineers.

Numbers like these show: regardless of how the bubble debate turns out, enormous capital continues to flow into building out AI infrastructure and regulation. For you as a Mittelstand company, that means above all one thing: the supply of tools, models, and specialized solutions is set to broaden rather than narrow - another argument for not committing to a single provider, but choosing the best solution for each use case.

Another, often overlooked factor comes from regulation: since August 2026, the EU labeling requirement for AI-generated content has been in force, along with expanded enforcement powers for the EU Commission over providers of general-purpose AI systems. At first glance, that may seem to have nothing to do with the stock-market valuation question - but it's actually another bubble-resistant criterion for your vendor selection. Anyone who invests today in a partner or platform that already handles compliance and labeling requirements cleanly reduces the risk of having to retrofit or switch providers in a year or two - regardless of how Nvidia's or Microsoft's share prices develop in the meantime.

Conclusion: decide by your business case, not by the headline

Whether we're currently in an AI bubble or not, we'll probably only know for certain in hindsight - just as with the dotcom bubble, whose scale only really became visible after it burst. The good news: you don't have to answer that question to make a good investment decision. What matters isn't Nvidia's P/E ratio or the next headline about hyperscaler spending, but three simple questions: does the investment pay off in a reasonable time? Do you stay flexible instead of depending on a single provider? And is the benefit measurable from the start, rather than resting on a promise for the future?

Anyone who consistently applies these three criteria makes a decision that holds up regardless of whether Peter Berezin or Juan Correa ultimately turns out to be right. That's exactly the core of a solid AI investment strategy in the Mittelstand: don't bet on market sentiment, bet on your own, clearly demonstrable business case.