Hook: The Phantom Rally
In the last 72 hours, Wolfspeed, STMicro, and Onsemi have collectively added over $12 billion in market cap. The narrative is clean: Nvidia's Vera Rubin ramp is fueling power chip demand. But let me be clear—this is a second-order effect being traded as a first-order catalyst. The market is pricing a SiC (Silicon Carbide) renaissance based on a GPU launch. That’s a structural error. In DeFi, liquidity is the only truth that matters. In semiconductors, the truth is in the order flow and the supply chain bottlenecks. This rally is built on a story, not on confirmed P&L.
Context: The Vera Rubin Power Architecture
Vera Rubin is Nvidia's next-generation GPU architecture, expected to succeed Blackwell. The industry consensus is that its power draw will exceed 1kW per GPU. This forces a fundamental shift in server power delivery: from 12V backplane architectures to 48V or even high-voltage DC distribution. This shift creates demand for a new class of power chips: GaN (Gallium Nitride) for board-level DC-DC converters, SiC for facility-level UPS and PSU, and high-efficiency silicon MOSFETs for intermediate stages.
The three companies in focus—Wolfspeed, STMicroelectronics (ST), and Onsemi—are all IDMs (Integrated Device Manufacturers) with significant SiC exposure. Wolfspeed is the purest play: it owns the substrate, epitaxy, and device fabrication. ST and Onsemi are broader, with automotive and industrial revenue dominating their books. The market is assuming that Vera Rubin’s power needs will directly translate into SiC orders for these three.
Core: The Order Flow Analysis — A Structural Mismatch
Let me dissect this with the precision of a DeFi audit. I’ve spent years analyzing tokenomic structures for hidden vulnerabilities. The same framework applies here: we need to trace the intended flow of capital and power.
1. SiC is for Infrastructure, Not for the Board.
The immediate power need for a Vera Rubin GPU is not 1200V SiC. It’s low-voltage, high-current conversion: 48V to 1V or 0.8V at hundreds of amps. This is the domain of GaN and high-performance silicon. SiC is great for the facility’s UPS (Uninterruptible Power Supply), the main power supply unit (PSU) converting AC to 48V DC, and the high-voltage distribution. But the "ramp" for Vera Rubin, which implies volume production of servers, will primarily drive demand for board-level power chips.
Who makes those? EPC (Efficient Power Conversion), Navitas Semiconductor, MPS (Monolithic Power Systems), and Infineon. Not Wolfspeed. Not ST. The market is conflating "power chip demand" with "SiC demand." That’s a dangerous oversight.
2. The 200mm SiC Yield Problem.
Wolfspeed’s Mohawk Valley fab is the first 8-inch (200mm) SiC facility at scale. Its ramp has been plagued by low yields and high costs. Based on my experience auditing DeFi protocols, I know that a "promising" technology with a 30% failure rate in execution is a short-term volatility play, not a long-term hold. The industry benchmark for 6-inch SiC yields is 70-80%. 8-inch SiC is still in the high 50s to low 60s, and that’s being generous. If Vera Rubin demand surges in 2026, Wolfspeed will need to deliver thousands of 8-inch wafers. If the yield is stuck at 60%, the effective cost of goods sold will be 40% higher than the ASP (Average Selling Price) they can negotiate with Nvidia. Nvidia is a ruthless negotiator. They will push the price down to the marginal cost of the next best alternative. That alternative is GaN or silicon.
3. The Nvidia Supply Chain Dominance.
Nvidia is not a passive buyer. They are the most powerful customer in the tech supply chain. They will drive the architecture. The rumor is that Nvidia is developing its own power management ICs (PMICs) and controllers. If they control the controller, they can specify the exact power FETs (Field-Effect Transistors) required. This commoditizes the power FET. The value shifts from the "SiC module" to the "Nvidia-validated PMIC." This is a classic "vertical integration" move that squeezes the IDM’s margins. In DeFi, we call this a "rug pull" on the yield. The same principle applies here.
4. The Inventory Cycle is a Tailwind, Not a Catalyst.
The global power semiconductor market is in a late-stage inventory correction. The automotive and industrial sectors have been weak for 18 months. The AI data center demand is a genuine green shoot. But it’s not a tidal wave. It’s filling a capacity gap. The market is pricing this as a new growth cycle. It’s not. It’s a demand normalization for a specific, high-value niche. The total addressable market for AI server power chips in 2026 is likely $3-5 billion. That’s a fraction of the $50 billion+ power semiconductor market. The stock reactions are implying a disproportionate contribution.
Contrarian: The Retail vs. Smart Money Divergence
The retail narrative is simple: "Nvidia is building a new supercomputer → power chips will be needed → buy Wolfspeed, ST, Onsemi." The smart money is more nuanced.
I see a classic "sell the news" event forming. The actual Vera Rubin product launch is months away. The supply chain contracts for power chips are likely already awarded. The market is extrapolating a future order book that is already partially priced in. The contrarian angle is that the true beneficiaries of Vera Rubin’s power architecture are not the legacy SiC IDMs, but the GaN specialists and digital power IC companies.
The hidden risk is that the SiC content in a Vera Rubin server is modest. The high-value power chips are the GaN FETs and the controller ICs. The SiC is used downstream in the facility’s power distribution. The data center operators (Google, Amazon, Microsoft) are the buyers of the data center infrastructure, not Nvidia. The procurement cycle for facility-level power is slow and not directly tied to a GPU launch. The correlation between Vera Rubin’s ramp and a Wolfspeed’s revenue is weak.
Furthermore, the geopolitical risk is underappreciated. China controls 80% of the global gallium supply. If the U.S.-China trade war escalates, GaN supply chains could be disrupted. This would create a short-term spike in SiC demand as a substitute. But that’s a reactive, negative catalyst, not a positive growth story. The smart money is hedging via GaN supply chain diversification, not buying SiC on the idea of a structural deficit.
Takeaway: The Price Levels to Watch
The rally is a liquidity event. It’s a chance to offload positions that are overvalued relative to their fundamental order flow. For Wolfspeed, the $30-$35 level is a selling zone. The stock is priced for a yield that doesn’t exist. For ST and Onsemi, the moat is broader, but the AI exposure is a distraction. The real test comes when Nvidia announces its next-generation supply chain partners. If the market discovers that the bulk of the power chip demand is flowing to GaN companies like Navitas or to power management firms like MPS, expect a 30% correction in these names.
The question is not "Will Vera Rubin need power?" The answer is "Yes." The question is: "Which power chip vendors will capture the margin and the volume?" The evidence suggests it won’t be the legacy SiC IDMs. In DeFi, we look for the protocol that captures the fees, not the one that provides the underlying liquidity. The same logic applies here. The smart money is already rotating out of this narrative.
Greed is a variable; discipline is the constant.