BRUCE 브루스

X-FAB has been written about all year. There is nothing new to learn about X-FAB that is not already reflected in the price. Yet the stock moves up as much as 70% today because of this post. So either the market has been completely wrong until now, or the claims being made are wrong. Spoiler, it is the claims that are wrong. X-FAB is being pitched as “photonics + power semis” with four supposed AI hooks. The reality is that these segments together account for only about 23% of revenue, and almost none of that has meaningful AI exposure. Microsystems and photonics are mainly tied to automotive sensors, medical microfluidics, and industrial sensing, while integrated photonics contributes only about $5–6M in NRE and remains in the R&D phase. Wide-bandgap exposure is still overwhelmingly automotive and renewables, with datacenter only described on the Q1 call as a recent add-on market. Real AI datacenter revenue is likely well below 2%. The remaining 77% is classic analog and mixed-signal CMOS, heavily concentrated in automotive at around 62%, and that core business is currently shrinking with automotive down 10% YoY. X-FAB is today a European specialty foundry for third-party designs running at roughly 60% utilization, FY2025 revenue around $770M, with only marginally positive EBIT, $291M net debt, and negative owner FCF. This is a high fixed-cost contract manufacturer sitting at the bottom of a cyclical downturn. The claimed 800V power semi exposure through Navitas and Power Integrations is simply wrong. There is no pathway from NVIDIA’s 800V push into X-FAB. Navitas runs through Powerchip and, from late 2025 onward, GlobalFoundries for U.S. capacity. TSMC exits GaN entirely by mid-2027. Power Integrations manufactures PowiGaN internally. Infineon, Innoscience, MPS, ROHM, STM, and TI are vertically integrated. X-FAB’s GaN remains developmental with no commercial revenue, and its production-capable dMode platform is not aligned with the eMode datacenter mainstream. There is no NVIDIA, Navitas, or Power Integrations design win. The silicon photonics argument is half true but massively overstretched. NVIDIA is indeed part of X-FAB’s photonixFAB consortium, but only as an application development partner evaluating technology. That does not imply high-volume manufacturing. NVIDIA’s actual CPO supply chain runs through TSMC and Tower. X-FAB’s photonics business is still pre-revenue, with product commercialization likely not before 2027–2028, and LIGENTEC controls the direct customer relationship. Consortium membership is not a production contract. The MEMS claim is technically correct but irrelevant for AI. X-FAB is a meaningful pure-play MEMS foundry for automotive sensing, medical microfluidics, and industrial applications. But the MEMS segments that matter for AI datacenters, optical circuit switch micro-mirrors and MEMS timing oscillators, are controlled elsewhere. The total addressable pure-play MEMS foundry market is only around $1B globally. X-FAB is a top-five player in a niche that is simply too small to move the valuation narrative. The “only high-volume SiC foundry in the U.S.” statement is also true but misleading. The global SiC market is dominated by integrated manufacturers like Infineon, STM, Wolfspeed, onsemi, and ROHM, which control over 90% of industry revenue. X-FAB’s share is around 1% globally and still operates on 150mm wafers while the industry transitions to 200mm. The headline 152% YoY growth is coming off a tiny base and is constrained by consignment substrate supply, limiting margin expansion. Government support is real, but it is not an earnings backstop. The funding is tied to growth capex for future capacity, with commercial ramp from Fab4Micro only expected around 2029. The U.S. support remains politically uncertain. These subsidies imply years of additional capital outflow before any return, not reduced near-term risk. second post ➡️