Hook:
Over the past 72 hours, Wolfspeed, STMicro, and On Semiconductor each rallied 8–12% on a single narrative: Nvidia's Vera Rubin platform is ramping and it needs power chips. Smart money doesn't trade the headline; it trades the block time. The real question isn't whether power demand is real — it's whether the supply chain can deliver without bleeding cash. I've been tracking this space since 2020 when I ran a 45% APY arbitrage on Compound, and I can tell you: the semiconductor play here is more nuanced than the retail crowd thinks.
Context:
The three names — Wolfspeed (SiC substrate + device), STMicro (IDM with SiC/GaN), Onsemi (power management + SiC) — are not the only beneficiaries of the AI data center power boom. The market is pricing them as "AI power plays," but the underlying demand vector is split: Vera Rubin's GPU board pulls 1kW+ per chip, requiring a shift from 12V to 48V bus architecture, which favors GaN for DC-DC conversion and high-voltage silicon MOSFETs for the PSU. SiC, while critical for UPS and high-voltage distribution, may not see immediate volume in the GPU board itself. This is a classic case of "sentiment buys the dip; data fills the position." I learned this the hard way in 2017 when I manually audited 50 ICO contracts and found reentrancy bugs that saved my firm $2M — code first, narrative later.
Core Insight:
Let's break down the order flow. The power semiconductor market is ~$50B, growing at 5–7% CAGR historically. AI data centers could lift it to 8–10%, with SiC/GaN growing 15%+ annually. But the real alpha is in the structural mismatch: existing 150mm SiC fabs (Wolfspeed's Durham, ST's Catania) run at ~70-80% yield for mature 6-inch lines. The new 200mm lines (Mohawk Valley, Onsemi's Hudson) are still ramping, with yields in the 50-60% range. That means the effective output of high-quality SiC devices is constrained until 2026 at least. Meanwhile, blockchain mining — Bitcoin ASICs and Ethereum PoS validators — also consume significant power conversion hardware. Each Antminer S19 uses ~3kW of power, requiring robust PFC and AC-DC stages that rely on silicon superjunction MOSFETs. As miners migrate to more efficient units (S21, M60), the demand for GaN-based power modules in the 48V-to-1V conversion stage will spike. Based on my experience designing a yield optimization strategy in DeFi Summer, I can tell you that the capital efficiency of a mining farm depends on uptime and power efficiency — and the semiconductor bottleneck directly impacts that.
Here's the contrarian piece: most retail investors are buying the SiC narrative as a pure AI play. But the 800-pound gorilla, Nvidia, is likely designing its own power management ICs for Vera Rubin, leaving only the discrete power FETs to external suppliers. This means the gross margin for ST and Onsemi on those parts could be compressed to 20-25% — far below their corporate average of 35-40%. Wolfspeed, with its high depreciation on the 200mm line, is even more vulnerable. In 2022, when my portfolio dropped 60% in the bear market, I pivoted to stablecoins and shorted altcoins — I learned that preserving capital requires questioning every narrative. The same applies here: the power chip demand is real, but the profit pool allocation is shifting from IDMs to the platform owner (Nvidia) and the substrate suppliers. The real winners might be the GaN-only players like Navitas and EPC, or the digital power controller companies like MPS and Renesas.
Takeaway:
The Vera Rubin ramp is a catalyst, not a guarantee. For DeFi yield strategists looking at this as a cross-asset signal: monitor the 200mm SiC yield data from Wolfspeed's quarterly reports. If yields break above 70% by Q3 2025, the supply glut will crush SiC prices, squeezing margins. If they stay below 60%, the narrative will hold but the stock will be a rollercoaster. Either way, the smart money is already positioning in GaN ETFs and digital power names. The block time doesn't lie — trade the data, not the hype.

