Broadcom’s latest earnings report delivered record revenue but triggered a sharp selloff across global semiconductor stocks—not because of weak performance, but due to its conspicuous omission of full-year AI chip guidance. This technical silence laid bare investors’ deep anxiety over the sustainability of AI infrastructure spending. More alarmingly, the shockwave rippled far beyond data centers, crashing into seemingly unrelated power semiconductor firms: Germany’s Infineon saw its shares plunge 12.8% in a single day, its worst drop in months. Despite Infineon’s clear disclosure that it has minimal exposure to high-end AI training chips—deriving over 70% of revenue from automotive and industrial power systems—the market punished it as if it were another AI-dependent play. This “guilt-by-association” sell-off reveals a brutal truth: under today’s valuation regime, every semiconductor company is forced into the same AI narrative crucible.
Infineon’s predicament is not isolated. As Europe’s power semiconductor leader, its fundamentals are anchored in long-term megatrends like electric vehicles and renewable energy inverters—domains untouched by Meta or Microsoft’s data center capex cycles. Yet when Broadcom signaled potential moderation in AI spending, investors instantly repriced the entire sector’s risk premium. Warburg Research downgraded Infineon from “Buy” to “Hold,” citing valuations that had “priced in two years of growth.” This exposes a structural mismatch: power semiconductors should benefit from secular electrification trends, yet they are now hostage to short-term AI sentiment swings.
Meanwhile, Chinese GaN (gallium nitride) startup Innoscience is quietly rewriting the rules of power electronics. While Infineon relies on legacy silicon-based IGBTs, Innoscience focuses on high-volume 8-inch GaN-on-Si wafers, targeting fast chargers, data center power supplies, and 5G base stations with superior efficiency and switching frequency. Crucially, this aligns with AI servers’ urgent need for higher energy efficiency—industry estimates suggest GaN-based power modules can reduce data center energy consumption by 15–20%. Though Innoscience isn’t yet in mainstream AI supply chains, its materials-level innovation is reshaping the hidden cost structure of AI compute.
Broadcom itself sits in paradox. Its custom AI chips are heavily concentrated among a few clients like OpenAI, while VMware integration consumes significant management bandwidth. As markets realize AI infrastructure investment is constrained by hard limits—power availability, thermal dissipation, and geopolitical policy—Broadcom’s “AI story” shows fragility. That fragility transmits through sentiment channels, distorting valuations of non-core players like Infineon.
A deeper fracture lies in how geopolitics distorts technology roadmaps. The U.S. CHIPS Act and EU Chips Act prioritize logic chip manufacturing, treating wide-bandgap (WBG) semiconductors like GaN and SiC as “secondary strategic assets.” Yet reality dictates otherwise: every incremental TFLOP of AI compute exponentially increases demand for efficient power delivery and cooling. Innoscience’s 8-inch GaN fab in Zhuhai, China, now boasts 70,000 wafer starts per month—far ahead of any comparable GaN-on-Si capacity in the West. This manufacturing asymmetry is quietly shifting power semiconductor influence eastward.
Memory makers like Micron face a similar “certification paradox”: even after securing NVIDIA’s HBM3E validation, their shares get sold off. This confirms markets no longer reward technical credentials alone—they demand clear profitability paths and disciplined capacity expansion. If Infineon cannot convert its automotive and industrial strengths into recession-resilient cash flows, it will remain trapped in AI-driven volatility.
I judge that the next phase of semiconductor valuation will feature a “de-AI-ification” divergence: companies with structural demand (e.g., EVs, smart grids) will decouple from AI hype, while concept-tethered names face harsh repricing. If Innoscience cracks automotive-grade GaN certification, it could become the wedge that breaks Western dominance in power electronics. But the larger question remains: in the global race for compute supremacy, have we over-indexed on processing power while systematically underestimating the strategic bottleneck of power delivery? As AI server farms approach nuclear-reactor-scale energy demands, the true value of power semiconductors may only now be coming into focus.