A Crash at the Heart of the Boom

On October 30, Intel's stock plunged 29% in a single session, its worst day since the 1970s. It thus closed October at -45%, the steepest monthly decline ever recorded for the company, before shedding another 8% the following Monday. Yet just months earlier, the stock had multiplied sixfold in ten months, driven by the U.S. government's 10% stake in August 2025 and a $5 billion investment from Nvidia. The market saw it as a winner of the agentic AI wave, whose central processors are an indispensable cog.

The trigger was a forecast: Intel announced quarterly revenue $1 billion to $2 billion below analyst expectations. The stated cause is not a lack of overall demand, but soaring costs for memory and other components, which are deterring consumers from upgrading their computers. PCs still account for roughly half of the company's revenue.

The Paradox of Demand Outstripping Supply

The contrast with the server segment is striking. Intel's data center division grew 78% year-on-year in the third quarter to $7.3 billion, and CEO Lip-Bu Tan sees the global server processor market potentially quintupling in five years to around $200 billion. But the company acknowledges it cannot produce enough chips to meet demand, citing supply constraints affecting the entire industry: wafers, substrates and memory.

These, the company notes, are many of the same factors weighing on PC sales. The mechanism is therefore twofold. Where the customer is a cloud player willing to pay, scarcity translates into a shortfall in volumes to deliver. Where the customer is a price-sensitive consumer, it translates into forgoing the purchase. The same bottleneck fuels both excess professional demand and retreating consumer demand.

Other signals point the same way. Apple has also warned that the memory chip shortage is driving up its costs. The report does not detail the scale of that impact, however: the signal matters more than the measure.

Foundry as the Adjustment Variable

Faced with this shortage, Intel made a consequential choice: shifting part of the capacity at its Oregon site, previously dedicated to wafers for its advanced 18A technology, to the older Intel 3 node used for its data center products. In the short term, the decision serves the most profitable demand. In the medium term, it undermines the ambition to run a foundry for third parties, which relies precisely on 18A to attract outside customers.

David Yoffie, a professor at Harvard Business School and former Intel board member, sums up the problem: a potential customer fears that a foundry with no excess capacity beyond its own needs is "not a very credible supplier." In his view, if Intel does not land a major customer within a year, the strain on its economics will become more visible. He adds a second obstacle: players like Broadcom or Qualcomm may hesitate to manufacture at a competitor in the product arena.

The effort to date is considerable. Intel has spent roughly $108 billion in capital expenditure on this venture over five years, without a flagship customer for its most advanced chips, while its Ohio project is nearly five years behind schedule. Under Lip-Bu Tan's leadership, the strategy has turned more cautious: limiting spending until demand is under contract, with the announcement of "a few customers" in the next two quarters. The stakes extend beyond the company: according to Yoffie, TSMC could create only 20% to 25% of its global capacity in the United States, and that would take several years.

Capital, for Its Part, Is Not Retreating

Intel's decline does not signal a general drying-up of financing. On the same day, a major consolidation deal was announced: Marvell is acquiring laser supplier Lumentum for more than $100 billion ($102.6 billion according to one of the excerpts). Anthropic, for its part, disclosed spending commitments approaching $900 billion ahead of its IPO. Meanwhile, a TSMC campus valued at $265 billion is fueling a real estate boom in the Phoenix suburbs.

These amounts trace a logic of concentration: money is flowing to data center infrastructure components and to players capable of committing to colossal volumes. Companies that depend on a depressed consumer market, or that cannot demonstrate excess capacity, are being punished without mercy.

Adoption Continues, Keeping the Pressure On

End demand shows no sign of flagging. In France, 67% of companies with 20 or more employees report using generative AI, and 48% of the population uses it, up 15 percentage points in a year. ChatGPT claims one billion weekly users, with roughly seven in ten messages unrelated to work. The broader the usage, the more it weighs on computing capacity that, as shown, is itself constrained by component availability.

What the Data Does Not Allow Us to Assert

Several limitations apply. The report establishes an industry-wide supply constraint, but not that AI is its direct cause: the correlation is plausible, not demonstrated here. The scale of Apple's warning is not quantified, and the amounts of the two major financing deals are known only partially or differ slightly across excerpts. Views on Intel also diverge: an analyst at Mizuho Securities, Jordan Klein, notes that by mastering both the manufacturing and assembly of its advanced chips, Intel has "more control over its destiny" than most rivals, and that leaving capacity idle would mean leaving money on the table. Finally, nothing in the report measures the effect of a possible tightening of interest rates on these investment plans.

A Sorting Phase Rather Than a Reversal

The Intel episode illustrates less the end of a cycle than its transition into a selection phase. When scarcity becomes the limiting factor, value shifts to those who control capacity and captive customers, and penalizes those whose margins depend on components that have become expensive. The coming months will provide two concrete tests: the possible identification of foundry customers for 18A, and the industry's ability to ease constraints on memory and substrates. As long as both remain open, emerging technology signals will be less legible in stock prices than in factory capacity order books.

This article was written with the assistance of artificial intelligence and verified by our editorial team.