Intel Meets Half of CPU Demand, and Margins Jumped 12 Points

Intel can fill only about 50 percent of the CPU orders its leading-edge customers want, Chief Executive Lip-Bu Tan said during a fireside chat at Splunk's .conf26 conference in Denver, held Sept. 14-17, and several big technology companies' chief executives have called him directly to ask for more chips. "I have to apologize because our capacity cannot keep up," Tan said, according to TradingKey's reporting on the remarks.

Tan attributed the shortage to central processors, not graphics chips, saying demand from AI inference work, the running of an already-trained model rather than the training of one, has outpaced what Intel's fabs can produce. That is a different bottleneck than the one most AI capacity stories describe. GPU scarcity has dominated the conversation since 2023; Tan's math says Intel's more ordinary product, the server CPU, now has the same problem.

Why margins rose while supply fell

A company that cannot fill half its orders would normally be losing customers to a competitor that can. Intel's second-quarter 2026 results say the opposite happened: revenue grew 25 percent year over year to $16.1 billion, its strongest quarter in 15 years, according to The Motley Fool's reporting. Adjusted gross margin reached 41.8 percent, up 12 percentage points from a year earlier, and Intel posted adjusted earnings of $0.42 a share against an adjusted loss in the same quarter of 2025. Scarcity, not efficiency, is doing that work. When a supplier cannot meet demand and customers keep ordering anyway, the price the market will bear goes up, and Intel's margin recovery is the accounting record of that dynamic rather than evidence its costs fell.

Intel, Q2 2026	Figure
Total revenue	$16.1 billion, up 25% year over year
Data center and AI segment revenue	$6.3 billion, up 59% year over year
Foundry division revenue	$5.77 billion
Adjusted gross margin	41.8%, up 12 points year over year

Investors read the shortage the same way. Intel shares rose as much as 10 percent in the days after Tan's remarks, touching their highest level in nearly two months, according to TradingKey. The Motley Fool put the stock around $109 as of Sept. 19, up 6 percent that week and roughly triple its price a year earlier, and cautioned that at close to 53 times forward earnings, the shares already price in the capacity expansion Tan was describing rather than leaving room for it to still happen.

What 18A and 14A actually change

Intel's 18A process node, its most advanced in production, has moved to full-scale mass manufacturing, Tan said. The next node, 14A, enters production in the first quarter of 2027, and its first disclosed outside customer is not a chip company but Elon Musk's TeraFab project: Tesla is building a roughly $3 billion pilot line inside its Texas Gigafactory campus to validate the process at a capacity of a few thousand wafers a month, while SpaceX takes on high-volume manufacturing once the node matures, Tom's Hardware reported. Musk said on Tesla's April 2026 earnings call that the company would use 14A because it is "state-of-the-art and in fact not yet totally complete," a description that cuts against treating 14A as a solved capacity problem arriving on schedule.

1. Total revenue16.1 $ billions
2. Data center and AI segment6.3 $ billions
3. Foundry division5.77 $ billions

Source: The Motley Fool and Calcalist, Sept. 2026, citing Intel's Q2 2026 results

The packaging warning underneath the demand story

Tan paired the capacity apology with a separate warning about where chips get finished, not just fabricated: "Having 95% of global advanced packaging capacity concentrated in a single region is extremely dangerous," he said, according to TradingKey, without naming the region directly. That statement is about a different chokepoint than the CPU shortage; a fab running at full output still needs packaging capacity downstream to turn wafers into shippable parts, and Tan's own 50 percent figure does not say which stage, fabrication or packaging, is the binding constraint today.

What would change this read

Tan also used the appearance to talk about quantum computing, telling the audience "GPUs, CPUs and quantum computing are all forms of computing, so they will run together in certain application areas," and pointing to Intel's Tunnel Falls silicon spin-qubit chip and its Horse Ridge II cryogenic control chip as evidence Intel has kept investing through what he called years when "people also think the change is taking too long." He put a specific number on the timeline for even the infrastructure question, saying cryogenic cooling requirements will remain "a certain impact over the next three to five years," which is a longer horizon than the CPU shortage he was apologizing for in the same conversation. Intel has not said what share of its 50 percent shortfall closes when 14A reaches volume production, or whether the TeraFab wafers Tesla and SpaceX are counting on draw from the same constrained capacity as everyone else's orders. Until Intel publishes that number, the honest read of the .conf26 remarks is that a chip company said no to half its customers and watched its margin improve anyway, the same oversubscribed-capacity story already playing out at Oracle's GPU cloud business, and the same omission Intel made months earlier when it priced a new inference chip without disclosing its tokens-per-watt figure.