Intel Can Supply Only About Half of What Its Customers Want. That's a Better Problem Than It Sounds.

Intel CEO Lip-Bu Tan said the company can fulfill only about 50% of customer processor demand, driven by AI-inference CPU needs. Q2 revenue rose 25% to $16.1B, data center and AI revenue jumped 59% to $6.3B, adjusted gross margin expanded to 41.8% from 29.7% a year earlier, and adjusted EPS reached $0.42 versus a prior-year loss. Intel guided Q3 revenue to $15.8B-$16.8B, or roughly 19% growth at the midpoint, while forecasting a 42% gross margin; 14A production is expected to begin in Q1 2027. Despite improving fundamentals and supply-constrained demand, the article views shares as fully valued at roughly 53x next year's expected earnings after a 6% weekly gain and more than tripling year over year.

Analysis

The key investment question is whether the supply shortfall reflects durable, high-value demand or a transient product-transition bottleneck. If Intel can allocate scarce output toward data-center SKUs, mix and utilization can support incremental gross-margin upside over the next 1-3 quarters; however, the same shortage gives hyperscalers and enterprise OEMs a direct incentive to qualify AMD EPYC and ARM-based alternatives. Lost design wins are more damaging than deferred shipments because server platform decisions typically persist through multi-year refresh cycles.

The market is likely capitalizing a clean capacity ramp and sustained operating leverage simultaneously. That leaves little tolerance for a yield issue, delayed equipment installation, substrate constraint, or customer migration: a 2-3 point gross-margin miss or a deceleration in data-center growth would challenge the premium earnings multiple and could drive a 20-30% de-rating over 6-12 months. Management's demand assertion should be treated as unverified until backlog, lead times, average selling prices, and Q4 guidance demonstrate that demand converts rather than merely queues.

Near term, constrained supply can create an upside earnings setup if pricing and mix offset foregone unit volume, so an outright short ahead of the next report is unattractive. The more asymmetric view is that AMD benefits from Intel's inability to fulfill CPU demand without bearing Intel's execution and capital-intensity risk; ARM server adoption is a second-order beneficiary through AMZN, GOOGL and MSFT internal silicon programs. Conversely, equipment suppliers may see orders, but only ASML has a sufficiently concentrated leading-edge lithography exposure to make Intel capacity spending material, and its valuation already embeds broad foundry demand.

Key Decisions for Investors

• Establish a 3-6 month relative-value position: long AMD / short INTC in equal dollar amounts after the next INTC earnings release if INTC guidance does not rise materially. Thesis is customer substitution and execution-risk asymmetry; target 15-20% relative outperformance. Exit if INTC guides to sustained data-center growth above 30% with gross margin above 43%, indicating supply is converting cleanly into pricing and mix.
• For accounts requiring defined risk, buy INTC Jan-2027 put spreads, centered roughly 10-20% below spot, only following a post-earnings rally that leaves shares above $109. A normalization from a premium execution multiple toward a still-growth valuation offers approximately 2:1 potential payout-to-premium; invalidate on evidence of improving lead times plus upward full-year EPS revisions.
• Maintain a watch alert rather than an equipment-supplier recommendation: increase ASML exposure only if Intel discloses binding leading-edge capacity orders or capex materially above current expectations. Clean-room announcements alone are not enough, since tool revenue recognition and wafer-output contribution lag by several quarters.
• Do not treat NVDA as a direct beneficiary. Greater CPU inference demand can expand total AI infrastructure budgets, but it can also shift spend from accelerators to host compute at the margin; monitor hyperscaler capex commentary and CPU-to-GPU attach rates before expressing that cross-read.