The 'Gemini 5' Stress Test: Why the Market Is Wrong About AI Chipmakers and Big Tech Stocks

An amazing, almost paradoxical picture is unfolding in the stock market. Outwardly, the indices are doing perfectly fine. But if we look under the hood, in my view, we can see a severe imbalance and complete distortion of fundamental logic. Investors in a state of euphoria are buying up shares of chipmakers and data-center hardware manufacturers, inflating their multiples as a result. Meanwhile, the companies that are actually paying for this entire infrastructure spending spree are suddenly being valued by the market as dull dividend stocks.

Let's look at the dry figures — specifically, the forward price-to-earnings (P/E) multiples of four of the largest Big Tech firms. Meta Platforms (META) trades at 19.1 times forward earnings, Microsoft (MSFT) at 22 times, Alphabet (GOOGL) at 24.7 times, and Amazon (AMZN) at 31.1 times. For tech giants with such dominant market positions, this is not just cheap, but a historical anomaly. These Big Tech names have started to be priced like classic value stocks.

The reason for the selloff lies in Wall Street's fear of giant capital expenditures. Investors are seeing how billions of dollars are being burned in the furnaces of data centers to maintain the artificial intelligence (AI) race, and they are dumping shares in turn. But the market has forgotten to look at what this system holds onto at all — and has lost sight of the fundamental physics of technology development.

The Armor of Income: Who Actually Pays for the AI Spending Spree?

To understand the scale of the delusion, we need to separate the wheat from the chaff — basic income from expenses. Current multiples indicate that the market is pricing in the permanent destruction of Big Tech's margins.

But let's be honest. The revenues of these companies have no cyclicality and do not depend on how hyped AI will be in the next quarter. Meta and Google have monopolized global attention, and their advertising revenues are protected by the real economy. Small- and medium-sized businesses will not stop buying targeted advertising and context in search engines. Moreover, the implementation of AI already allows companies like Meta to target advertising incredibly accurately, optimize internal costs, and rake in cash hand over fist. These firms have colossal profits.

Meanwhile, Microsoft is deeply embedded in global corporate infrastructure — from operating systems to cloud services. Not a single corporation will refuse subscriptions to its software for the sake of savings. Microsoft's cash flow is as stable as gravity.

Yes, Big Tech is now spending fabulous amounts on purchasing chips from the likes of Nvidia (NVDA) and memory from Micron (MU). These firms are sitting on the money of Big Tech, but nothing threatens the income of the giants themselves. They generate real money from real sectors. At the same time, chipmakers are depending on Big Tech's desire to keep building new data centers.

Here, we come to the main blind spot of the current market consensus. In my opinion, the hardware market is overheated because everyone is sure that the need for computing power will grow exponentially forever. But is that true? To find out, let's conduct a logical “stress test.”

The Physics of the Process and the Limits of ‘Brute Force’

Why does the world need so many huge data centers right now? The answer lies in the current paradigm of AI development. We are in a “brute force” phase of computing.

The logic of AI developers at this stage is straightforward. Each new model becomes “heavier” than the last one. In order for AI to become smarter, more and more parameters are shoved into it as well as more terabytes and petabytes of information. The data cloud is growing continuously, and as AI advances, it proportionately needs more memory, more advanced chips, more servers, and more energy. But this extensive growth has its physical and logical limits.

Investors should understand one thing: The entire fundamental world library has already been downloaded into the brains of current neural networks. From the basic laws of physics to thermonuclear reactions, classical literature, the principles of psychology, macroeconomic trends, and historical chronicles — AI has already read all of it. The base has been formed. So why are the models swelling now? Junk data.

Neural networks vacuum up the internet in real time. They record huge amounts of data — dialogues with millions of users, chat bot logs, and endless variants of interactions of the same type. The information cloud is swelling from data, much of which does not carry any real, fundamental value.

The Inevitability of Optimization

Now, let's imagine a completely realistic scenario. Suppose a hypothetical future AI model — we'll call it Gemini 4 — comes out at the peak of this extensive phase. It will be incredibly heavy, eat up even more memory, and require new mega data centers. Chipmakers will report record profits.

But what will happen in a year or two, when developers hit the ceiling? A realization will occur: Keeping this giant array of junk data in RAM permanently is economic and architectural madness. That's when the development of AI will turn toward optimization. Instead of storing billions of raw dialogues and endless logs of interactions, algorithms will learn to do what the human brain does. They will analyze the array, deduce logical chains, draw conclusions, and delete the original raw material as unnecessary.

Let's go one step further and imagine a hypothetical Gemini 5 model. It will receive this entire giant array of information accumulated over the years, analyze it, compress it 10 times in the form of high-level generalizations and fundamental patterns, and simply throw away 90% of the garbage. The intellect of the system will grow due to the quality of connections, not the volume of stored bytes. The model will slim down. Artificial intelligence, which for years had only grown heavier, will suddenly become light and elegant.

If the focus does shift from throwing hardware at the problem to the software optimization of code and memory, a real nightmare will begin for chip manufacturers — and a golden age will begin for Big Tech.

Anatomy of a Disaster for Hardware Manufacturers

As soon as the paradigm shifts and our hypothetical Gemini 5 proves that a "light" model can be many times smarter and more efficient than a heavy one, a tectonic shift will occur in the hardware market. Investors who today are buying shares of memory and GPU manufacturers at high prices will face a harsh reality. Chipmakers by their nature are in a deeply cyclical business, which is now at an artificially inflated peak.

Let's take off our rose-colored glasses and examine how companies like Micron earn their billions right now. These firms are winning not because they physically started to produce 10 times more silicon wafers; increasing physical production is years of work and requires huge investments in factories. Instead, their current huge profits are the result of a price supercycle. A severe shortage of HBM and accelerators has allowed these companies to raise prices many times over.

Now let's put the puzzle together. Currently, succumbing to the euphoria, chipmakers are building new factories all over the world. These facilities will begin to be mass-commissioned in a year or two, forming a so-called capital lag. But at the same time, AI models are poised to slim down due to the optimization of algorithms. This setup invites a collapse of demand: Big Tech will suddenly realize that it no longer needs to double memory purchases every year, because the industry's already-built data centers will become more than enough.

The result? A classic crisis of overproduction. In our hypothetical scenario, as demand slows down sharply, new chip factories will just be beginning to come online and pour millions of new chips onto the market. Supply will exceed demand, and memory and GPU prices will drop like a rock, returning to historical averages. In turn, the multiples of Micron, Nvidia, and others will fall many times over, with shares going into a prolonged nosedive. The market will remember the harsh reality of technological cyclicality.

AWS Is the Weakest Link Among the Giants

In this scenario of software slimming down AI, Big Tech firms are liable to react differently. Here, we should draw a clear distinction between Amazon and the likes of Google, Meta, and Microsoft. While Meta and Google use AI for internal needs — so that their own services work more efficiently and advertising is sold at higher prices — the business model of Amazon's AWS is fundamentally different.

AWS is a digital landlord. Amazon has built giant data centers and leases the computing power to millions of third-party companies, startups, and corporations. When the era of AI optimization comes, AWS clients will find they need less server power to solve their tasks. The demand for cloud computing within Amazon's infrastructure will drop sharply, leading to a tough price war between cloud providers and the depreciation of computing power itself.

Amazon's data centers, built at the peak of the hype for heavy models, will turn out to be redundant. That is why Amazon is the most vulnerable of our four Big Tech companies, as it depends the most on the preservation of extensive hype.

A Cash Tsunami for Big Tech

For the rest of the giants — Google, Meta, and Microsoft — this potential reversal in AI will become a moment of absolute triumph and colossal enrichment. What happens to a company with stable and growing basic income when its main expense item suddenly shrinks? Explosive growth of free cash flow (FCF).

As soon as AI proves its lightness, the capital expenditures of these companies should drop sharply — and the cash that previously went to chipmakers will stay with Big Tech.

At the same time, the incomes of these Big Tech firms will remain. Google will continue to skim the cream off search traffic and YouTube. Meta will spin highly efficient, AI-targeted advertising to billions of users. Microsoft will keep selling subscriptions to global businesses.

The net profits of these firms will likely take off into the stratosphere — and then the market will finally realize the full depth of the mistake being made today. Investors will see that Big Tech, which they once buried for “inefficient” spending on AI, simply built the infrastructure of the future at the expense of its own cash, optimized it, and turned into an even more efficient printing press for money.

A Reassessment of Value

The current forward P/E ratios of Alphabet or Meta are a historical gift for investors, disguised as market panic. Right now, the market is showing short-sightedness, overestimating the temporary shortage of hardware and underestimating the power of software engineering.

The problem with the crowd is that it extrapolates charts in a straight line. But technological development never goes in a straight line. A linear function is generally not applicable to reality, as real progress always occurs through qualitative changes and sharp turns.

A turn toward lighter, advanced AI models — in which algorithms objectively clean up and delete large amounts of junk data — seems like an obvious and inevitable next step in my view. This qualitative leap will radically change the entire current alignment of the market. For investors, it's important to understand not only dry quarterly reports but also these deep technological — and in some ways even ideological — forces at work.