Jukan
Semiconductor & AI Infrastructure Research at Citrini
Foundries, not CPU designers, are the better CPU bottleneck trade. A rush to make CPUs would tighten TSMC's wafer capacity and strengthen its bargaining power, while Intel's own fabs should let it expand CPU supply over time and draw ASIC and CPU customers to Intel Foundry. AMD is hard to recommend because hyperscalers increasingly use in-house CPU ASICs internally and AMD must compete for TSMC capacity. Arm is not a pure CPU trade, as the memory crunch likely hits its smartphone royalties. x.comÂ
Intel's server CPU supply constraints should ease considerably in the second half, more than Mizuho allows, as demand shifts to newer parts like Clearwater Forest and Intel 3/Intel 7 bottlenecks ease once the Fab 34 takeover materializes. AMD is better placed with TSMC, but N3 wafer demand is ramping now, which could pose trouble if hyperscalers want newer generations; its early lead came from N4/N5 supply freed by Qualcomm and MediaTek. Mizuho still sees strong Turin and Genoa demand. x.comÂ
US sovereign AI and defense demand for domestically made compute will bring Intel most, if not all, US semiconductor designers as customers. PC weakness also lets Intel shift wafers to higher-margin server CPUs, but that takes a few quarters; capacity to ship more server CPUs is the limiter. The $240 billion 2030 server CPU market forecast is a stretch case, not the base case, versus roughly $200 billion consensus. x.com x.com x.com x.com x.comÂ
Intel's production was reportedly affected by a pause at its Israel fabs, which should benefit competitors Arm and AMD. Both look good on that basis. x.comÂ
Insufficient substrate supply may cap AMD's growth because the company has not secured enough capacity. The author says limited substrate availability drove AMD's lowered guidance, making supply-chain execution the constraint on how far it can grow. This adds an execution risk to the earlier concern that AMD is too expensive to buy despite strong results. x.com x.comÂ
TSMC likely benefits most from chip cost inflation because its advanced-node pricing power pushes higher foundry costs through the supply chain. Per ChannelGate, AMD told partners to expect about 10% higher Q4 pricing after TSMC reportedly raised advanced-node wafer prices 5-10%, though AMD has not confirmed it. Advanced nodes are now most of TSMC's wafer revenue, and the foundry layer keeps capturing more value as the AI supply chain gets more expensive. x.comÂ
AMD is not crazy expensive at 32x 2027 estimated EPS, because estimates look much too low and the real multiple may be in the low-to-mid 20s. It can probably add more capacity than Intel over the next 6-12 months despite TSMC's leading-edge constraint, and pricing is set to be a tailwind. Some semis look frothy, and AMD may be due for a breather, especially if the Iran war restarts. x.com x.comÂ
AMD and Nvidia face a growing capital burden from substrates without the benefit of foundry margin stacking. Each is committing $1 billion to the substrate ecosystem; AMD has also publicly committed $10 billion to its Taiwan supply chain, more than three times its Q2 net income. That capital is unavailable for R&D and product development, while their gross margins are flat or shrinking. Intel's gross margin is growing as customers prepay its foundry for substrates. x.comÂ
AMD can't expand capacity to meet demand, which will become clear toward year-end. TSMC makes its chips, and TSMC isn't going to kick out Apple and Nvidia to make room for AMD. x.com x.com x.comÂ
AMD stands to gain as CPUs for agentic AI could exceed 50% of the total CPU market by 2030; @BenBajarin estimates a 62% CAGR for agentic CPUs. The risk is supply: next year ASIC makers and AMD may struggle with supply chain capacity constraints, as Nvidia locks up tier-one suppliers with hard supply agreements, which per @lokoyacap signals worsening shortages. That lockup is good for tier-two suppliers. x.com x.comÂ
Intel is benefiting from this upcycle as a CPU maker and as an overflow and swing producer in foundry, and swing producers always rally more than the low-cost producer in an upcycle. TSMC, Nvidia and ASML showed restraint on capex and pricing, which extended the cycle but cost them volumes that bled to Intel. Stocks run on rate of change: Intel going from dead to maybe not dead was a better long than TSMC going from alive to more alive. x.com x.comÂ