Stock Market Updates · AI & Semiconductors

Who Actually Makes Money in AI? The Memory-Chip Squeeze Is Rewriting the Answer

By Luke  |  EverHealthAI  |  June 2026

There is a useful way to understand what just happened to Micron, and it has nothing to do with semiconductors. Think about airlines and oil.

Airlines can be brilliantly run, beloved by customers, and strategically essential — and still see their profits evaporate the moment crude prices spike, because they depend on an input they don't control and can't easily substitute. Memory-chip makers are now playing the role of the oil producer in the AI economy. Micron, along with Korea's Samsung Electronics and SK Hynix, supply the high-bandwidth memory that AI data centers cannot function without. And this year, that input suddenly became dramatically more expensive — which means Micron's blockbuster profits are, viewed from the other side of the transaction, everyone else's blockbuster costs.

This is not a minor margin story. It is one of the largest and fastest profit transfers in recent market history, and it is quietly redrawing the map of who actually captures value in the AI boom. Investors transfixed by the breadth of the AI rally are missing the more important question underneath it: not whether AI makes money, but which layer of the AI stack makes money — and for how long.

The Mechanism: Scarcity Meets Inelastic Demand

The reason this transfer is happening comes down to a brutal collision between constrained supply and desperate demand.

High-bandwidth memory is extraordinarily difficult to manufacture, and the capacity to make it is extremely limited. Critically, building new production facilities takes years — there is no fast way to add supply in response to a demand surge. So when data-center demand exploded, the only release valve was price. Micron raised DRAM prices more than 60% in a single quarter while shipping only a low-single-digit percentage more product. NAND flash memory prices jumped more than 80%. Over the past year, memory prices have roughly quadrupled.

Sit with that for a moment, because it inverts the entire history of the semiconductor industry. Memory chips are supposed to get cheaper every year — relentless deflation is the normal state of the business. A quadrupling of prices in twelve months is not a cyclical wobble; it is a regime change, driven by the fact that AI demand arrived faster than the industry's physical ability to supply it. The result, for Micron alone, was customers paying roughly $18 billion more in a single quarter. That money came from somewhere. And where it came from is the whole story.

The Crucial Difference: AI Can't Pass the Cost On

Here is where the AI dynamic diverges sharply from the consumer-electronics version of the same shock — and where the real investment insight lives.

In consumer markets, rising memory costs get passed to buyers. Apple raised MacBook prices more than 15% this week. PC builders are watching memory modules cost more than the processors they pair with. That price pass-through is uncomfortable, but it's also self-regulating: higher prices dampen demand, just as expensive gasoline makes people drive less. The system finds equilibrium.

AI companies cannot do this — or at least, they are choosing not to. The leading model producers, OpenAI and Anthropic, are running large losses by design. Their services are priced to acquire customers and capture market share, not to generate profit. Everything in the current AI business model assumes that prices stay low enough to drive adoption, with monetization deferred to some future point of scale and dominance.

So the cost gets absorbed. It lands on the data-center operators or the model producers, squeezing margins in the upper layers of the AI stack while the chip suppliers at the bottom capture the gains.

What the Stock Market Is Already Telling Us

The remarkable thing is that this profit shift may already be visible in share prices — if you know where to look.

The divergence sharpened after April, when the memory stocks accelerated upward and the hyperscalers — along with Nvidia, which packages memory with its GPUs — fell. There are two ways to read this. The skeptical interpretation is that it's just momentum traders rotating out of last year's AI winners and into this year's hot trade. That's plausible — the speculation in memory shares is extreme enough that it's genuinely hard to separate fundamental repricing from pure momentum.

But the divergence fits the profit-transfer thesis almost too neatly. If margins are genuinely migrating from the model and cloud layers down to the memory suppliers, then memory stocks soaring while hyperscalers stall is exactly what you'd expect. The total profit pool of the AI "stack" of businesses stays roughly constant — but its distribution is shifting downward, toward whoever controls the scarcest input.

What the Market May Be Getting Wrong

The dominant AI investment narrative treats the boom as a rising tide — buy AI exposure broadly and ride the wave. This analysis suggests that framing is dangerously imprecise. The AI boom is not lifting all layers of the stack equally. Right now, it is enriching one specific layer at the direct expense of the others.

That matters enormously for portfolio construction. An investor who owns "AI" through hyperscalers and model-adjacent names may be holding the exact part of the stack currently losing the margin war — while believing they own the winning theme. The theme is winning. Their slice of it may not be. This is a sharper version of the concentration fragility that recently rattled the AI complex: it's not just that the market leans heavily on one theme, it's that even within that theme, the profits are migrating to a narrow set of suppliers.

Cyclical or Structural? The Three Pressure Valves

This is where the oil-and-airlines analogy delivers its final lesson. There are only three ways to respond to a sustained input-cost shock, and all three are likely to play out in AI over different time horizons.

First, the short-term response: lower profits. Buyers eat the cost and accept thinner margins. This is happening now, and it's why hyperscaler and model-maker economics are under pressure.

Second, the long-run response: efficiency. Buyers find ways to need less of the expensive input — more efficient model architectures, better memory utilization, smarter inference. This takes time but is already a major focus of AI engineering.

Third, the eventual response: new supply. Fat profits are the strongest possible incentive for capacity expansion. The memory makers will build new facilities. It takes years — which is exactly why prices spiked in the first place — but it will come, and when it does, the scarcity premium compresses.

The implication is that the memory-chip windfall is more cyclical than structural. The current pricing power is real but self-limiting: it is simultaneously the reward for scarcity and the incentive that will eventually end the scarcity. The deflationary gravity of the semiconductor industry has been suspended, not repealed.

That said, there's a genuine counter-argument worth weighing. Some investors note that the increased manufacturing complexity and the consolidated, small number of competitors in high-bandwidth memory mean new supply may come more slowly and with more pricing discipline than in past chip cycles. Fewer players could extend the windfall longer than the usual cycle suggests. Even some bulls who hold this view, though, are hedging — one prominent fund manager has rotated from Samsung's common stock into its preferred shares, which normally track closely but have lagged during this boom, a quiet way of staying exposed while reducing risk if the speculation unwinds.

What to Watch Next

The AI boom is real, but it is not generous to everyone who touches it. Right now, the scarcest input — memory — is capturing a disproportionate share of the profit, at the direct expense of the model makers and cloud providers that investors most associate with the AI story. That transfer won't last forever; the economics that created it will eventually unwind it. But while it lasts, it's a powerful reminder that in any gold rush, the most reliable profits often belong to whoever sells the shovels — until enough shovels get made.

This article is for informational and educational purposes only. It does not constitute financial or investment advice. Always consult a qualified financial advisor before making investment decisions.