NextFin News - Morgan Stanley is warning that one of the AI trade’s most attractive assumptions is starting to fray: chipmakers may not keep the pricing power that justified their outsized rally. Lisa Shalett, the firm’s chief investment officer for wealth management, said the AI data-center stack is being re-engineered around lower-cost proprietary chips designed by hyperscalers, a shift that could move more of the economic upside away from merchant semiconductor suppliers and toward the big buyers that control cloud platforms, software and customer access. The warning lands after a powerful semiconductor run that left investors asking a more uncomfortable question: if AI demand is still expanding, why should the chipmakers automatically keep the margin windfall?
The answer matters because pricing power is not the same thing as demand. A company can sell more chips and still see the value of each chip compressed if its customers are getting better at building substitutes, negotiating harder or integrating more of the stack internally. That distinction sits at the center of Morgan Stanley’s call. The firm is not saying the AI cycle is over. It is saying the profit pool may be migrating. In the short run, that can look like an ordinary rotation away from the most crowded winners. In the longer run, it can become a structural change in who captures the returns from AI infrastructure spending.
That is why the market read is more complicated than a simple semiconductor bearish note. The chips still matter. The question is whether they matter as a scarce tollbooth or as a more fungible input. If hyperscalers keep designing more of their own silicon, the chipmakers’ pricing leverage becomes easier to challenge, even if overall AI capex remains elevated. The tension between rising demand and weakening bargaining power is the story.
Market Reaction: A Rotation Story More Than a Demand Story
The immediate market context fits the rotation thesis. Morgan Stanley previously argued that recent weakness in U.S. semiconductor stocks pointed to a broadening in market leadership and a possible move toward hyperscalers, consumer discretionary, transport and biotechnology shares. That framing matters because it treats semiconductor softness not as a pure collapse in AI demand, but as evidence that the first beneficiaries of the AI buildout may have already been repriced. In other words, investors are no longer paying the same premium just for being in the supply chain.
That shift is important because it changes the transmission mechanism. When AI spending first accelerated, investors paid up for scarcity: advanced nodes, packaging capacity, memory and fabrication bottlenecks. Once customers start designing more proprietary chips, that scarcity premium weakens. The market then starts to ask which firms own the demand, not just the components. Hyperscalers can potentially turn the same AI spending into lower unit costs, faster deployment and stronger platform margins. Chipmakers, by contrast, become more dependent on keeping the manufacturing side tight enough to offset the customer’s growing leverage.
“We are seeing the AI data center tech stack, if you will, being re-engineered to include lower cost proprietary chips that many of the hyperscalers are now designing themselves,” Lisa Shalett said Friday on Bloomberg Television’s Surveillance.
The second-order implication is where the real market debate sits. If the biggest cloud buyers can internalize more of the silicon design process, then the next leg of AI profits may not belong to the companies making the chips at all. It may belong to the companies that can spread the same AI workload across software, subscriptions, cloud services and custom hardware. That is an earnings-multiple question as much as it is an industrial one. A supplier exposed mainly to hardware pricing will struggle to defend a premium if the customer can credibly threaten to build around it.
This is why the current call looks partly cyclical and partly structural. The cyclical piece is familiar: a crowded sector that has run hard can correct when investors rotate toward the next perceived winners. Semiconductor shares have done that before, especially after periods when AI enthusiasm outran near-term fundamentals. The structural piece is more durable. If hyperscalers keep shifting design work in-house, the merchant chipmaker’s pricing power does not recover just because the next quarter’s orders are strong. The customer relationship itself has changed.
There is no contradiction in saying both things at once. In the short term, semiconductor shares can rally again on earnings beats, supply constraints or guidance upgrades. In the medium term, however, the market may begin to treat custom silicon as a way for hyperscalers to capture more of the AI value chain. That would leave chip suppliers fighting to defend volume and margin at the same time, which is a harder equation than simply growing unit sales.
Recent trading has also highlighted a broader market issue: when the most visible AI winners wobble together, the move can matter beyond semiconductors. Morgan Stanley’s earlier call that market leadership may broaden out implies that the chip trade is no longer pulling the rest of the market on its own. If semiconductor upside stops being the default signal for AI optimism, then the market has to reprice what “AI exposure” actually means. A hyperscaler with custom chips, cloud revenue and an application layer may deserve a different multiple than a supplier that sells a narrowly priced piece of the stack. That is the kind of re-rating that does not arrive all at once; it begins as sector rotation and ends as a change in the hierarchy of returns.
There is also a second-order effect on capital spending. If hyperscalers believe they can lower cost per unit of compute by designing their own silicon, they can keep investing in AI while quietly shifting bargaining power away from outside chip vendors. That means the market can still see large capex numbers and still experience margin compression in the hardware suppliers. The headline growth rate may stay high while the profit mix changes beneath it. That is precisely why pricing power is a better analytical lens than demand alone.
On that point, Morgan Stanley’s warning is less about a quarter and more about a channel. The channel is customer substitution: the more a hyperscaler internalizes design, the more it can pressure the merchant chipmaker’s terms. The first-order effect is lower per-chip pricing. The second-order effect is a narrower profit pool for the entire semiconductor supply chain. The third-order effect is that investors start asking whether the AI winners are the companies shipping the hardware or the companies that can absorb the hardware into a broader platform with recurring revenue.
The strongest bull case for chipmakers is that AI infrastructure remains so demand-heavy that pricing pressure will stay muted for longer than skeptics expect. TSMC’s next-quarter results, due July 16, are expected to show that advanced packaging and leading-edge manufacturing are still central bottlenecks in the AI buildout. Investors will be watching for any update on full-year revenue guidance, capital spending and CoWoS packaging capacity, because those details would indicate whether the supply chain is still constrained enough to support firm pricing.
That argument is real. It says the market has not yet reached an equilibrium where supply catches demand and destroys margins. But Morgan Stanley’s warning is about the direction of travel, not the current level of demand. When the largest buyers have the resources to design their own chips, they do not need to erase merchant chipmakers; they only need to reduce dependence on them. That is often enough to cap future pricing power. Once a customer can credibly threaten substitution, the seller’s ability to push through price increases weakens even before the substitution is complete.
The comparison that matters is not between strong demand and weak demand, but between strong demand and strong bargaining power. Those are different things. Semiconductor firms can still post solid revenue growth while losing pricing power if the incremental demand is increasingly captured by lower-margin custom designs or if the buyers direct more of the economics toward integrated platforms. The market often mistakes volume growth for profitability durability. That is the error Morgan Stanley is flagging.
Why This Looks Structural, Not Just Cyclical
The critical judgment is that the pricing-power challenge is becoming structural, even if the stock reaction is cyclical. A cyclical setback comes from inventory digestion, a temporary pause in spending or a crowded trade getting unwound. Those episodes reverse when the next data point surprises the market. A structural shift is different. It happens when customer behavior, product architecture and profit allocation change in a way that does not automatically snap back.
On the evidence available now, this fits the structural category better. The reason is not just that semiconductor shares have become expensive or that investors are rotating. It is that hyperscalers have a strategic reason to internalize more design work. If they can reduce dependence on merchant silicon, they can lower cost, improve deployment flexibility and keep more of the AI economics inside the platform. That is not a one-quarter inventory adjustment. It is a competitive strategy.
The comparison with earlier cycle peaks is useful. In prior chip downswings, pricing pressure often receded once inventories normalized or demand re-accelerated. Here, the pressure comes from the customer’s ability to redesign around the supplier. That is a different mechanism. It means the supply chain can still be busy while the vendor’s negotiating power weakens. The old cycle playbook assumes that stronger end demand restores pricing. The new playbook suggests that customer self-sufficiency can cap it.
The counter-thesis is that the AI hardware stack is too complex for proprietary chips to displace the merchant leaders in a meaningful way. Advanced nodes, memory, packaging and production scale still sit with a small number of companies, and the most important buyers still need outside suppliers to keep their infrastructure on schedule. If that remains true, the current pricing-power concern may prove cyclical rather than structural: a valuation reset after a crowded trade, followed by another leg higher once demand data and earnings confirm the buildout is still accelerating.
The falsifying signal for Morgan Stanley’s view would be clear and measurable. If chipmakers continue to expand gross margins while hyperscaler-designed silicon stays a niche effort rather than a growing share of AI deployments, the pricing-power erosion thesis weakens. In that case, the market would be right to treat recent softness as a rotation, not a regime change. The burden of proof is on the structural call, because customer substitution has to show up in actual deployment patterns and margins, not just in the language of strategy decks.
That is also why the relevant comparison is not whether AI demand slows, but whether the mix of AI winners changes. If the same cloud companies keep winning software, services and platform share while also reducing their dependence on merchant chips, the AI trade can remain healthy while semiconductor pricing power erodes. The market could keep celebrating AI spending and still under-own the chip suppliers. That is a more uncomfortable conclusion than a simple “AI is over” narrative, and it is the one Morgan Stanley is pressing.
The market is still paying for AI growth, but Morgan Stanley is questioning who will own the profit when that growth becomes more custom, more internal and less dependent on merchant chips. If that shift sticks, the AI trade will stop rewarding pure exposure to silicon and start rewarding control of the platform above it.
What To Watch Next
In the short term, semiconductor shares will keep trading on order books, guidance and any sign that AI capital spending is accelerating or slowing. That is the cyclical layer, and it can reverse quickly if the next earnings round proves demand is still outrunning supply. The medium-term test is whether hyperscalers continue to bring more chip design in-house, because that determines whether pricing pressure becomes a persistent feature of the industry. Over the long term, the issue is whether the AI value chain migrates upward toward the firms that own cloud platforms, software and customer relationships, leaving chip suppliers with a smaller share of the economics even if unit demand stays robust.
Base case: AI spending remains strong, but the winners broaden as hyperscalers and integrated platforms capture more of the value. Upside case for chipmakers: supply remains tight enough that pricing power survives longer, keeping margins elevated. Downside case: custom silicon adoption accelerates, and the market begins to re-rate merchant semiconductor exposure as a lower-quality claim on the AI cycle.
The key watchpoint is not a vague sense of whether AI is “still hot.” It is whether the largest buyers keep describing proprietary chips as a central part of their deployment strategy while chipmakers struggle to defend pricing. If that happens, Morgan Stanley’s warning will look less like a short-term sector call and more like an early read on a lasting shift in who captures the economics of AI.
Explore more exclusive insights at nextfin.ai.
