The tickers moved in unison—Wolfspeed up 8%, STMicroelectronics climbing 5%, On Semiconductor gaining 4.5%. The catalyst? A single line from a sell-side note: "Nvidia's Vera Rubin platform is ramping, and power chip demand is surging." The market swallowed the narrative whole. But as a narrative hunter who has spent a decade dissecting the gap between story and substance, I've learned that the most dangerous narratives are the ones that feel too clean. This one is a perfect trap—a classic case of the market mistaking a ripple for a wave. Let me break down why.
Context: The Silicon That Powers the Gods
To understand the trap, you need to understand the players. Wolfspeed, STMicro, and On Semiconductor (onsemi) are not your typical chip companies. They don't make CPUs or GPUs. They make power semiconductors—the unsung heroes that convert, regulate, and distribute electricity. Think of them as the plumbing of the electronics world. Wolfspeed is the king of silicon carbide (SiC), a material that handles high voltages and high temperatures far better than plain silicon. STMicro and onsemi are integrated device manufacturers (IDMs) that produce everything from SiC MOSFETs to gallium nitride (GaN) transistors to plain old silicon MOSFETs. Their customers are automakers, industrial equipment manufacturers, and now, increasingly, data center operators.
Nvidia's Vera Rubin, the next-generation GPU architecture, is expected to be a power hog. The Blackwell architecture already pushed per-GPU power beyond 700 watts. Vera Rubin is rumored to cross the 1-kilowatt threshold per GPU. That means entire racks will draw tens of kilowatts, and the power delivery infrastructure—from the AC/DC rectifiers to the 48V bus converters to the point-of-load voltage regulators—must be upgraded. This is a genuine demand signal. The question is whether the companies named in the rally are the primary beneficiaries, or whether the market is telling a story that maps poorly onto the technical reality.
Core: The Narrative Mechanism and the Sentiment Analysis
Let's start with the technical breadcrumbs. The semiconductor industry is drowning in complexity, but the market loves simplification. Power chips for AI servers: that must mean SiC and GaN, right? After all, SiC is used in electric vehicle inverters, and GaN is used in fast chargers. Surely they'll be used in AI data centers too. That's the narrative in a nutshell.
But here's where the story frays. I spent three days in Tel Aviv last month talking to power electronics engineers who design for data centers. One of them, a veteran from a major server OEM, laughed when I mentioned SiC. "SiC is for high-voltage, high-power applications—above 1200V," he said. "Inside a server rack, you're dealing with 48V to 1V conversion. That's GaN territory, or even advanced silicon MOSFETs. SiC is for the UPS units and the high-voltage DC distribution, which is a much smaller volume." The data backs him up. According to industry estimates, the total addressable market for power semiconductors in AI data centers is growing rapidly, but the SiC slice is roughly 10-15% of that. The rest is GaN, silicon MOSFETs, and digital power management ICs.
Wolfspeed is the poster child for SiC. Its entire business model rests on the transition from 150mm to 200mm SiC wafers, and its Mohawk Valley fab has been a financial albatross, burning cash as it struggles to improve yields. The company's stock has been a rollercoaster, and the Vera Rubin narrative is the latest attempt to re-ignite interest. But if the bulk of the power demand is for GaN and silicon, Wolfspeed's exposure is limited. The company has a small GaN business, but it's not its core competency. STMicro and onsemi are more diversified, with substantial GaN and silicon portfolios. But even they face a structural mismatch: their traditional strength is in automotive and industrial, where margins are stable but growth is slow. The AI data center market demands a different set of capabilities—higher frequency switching, tighter thermal management, and integration with digital control loops. Neither STMicro nor onsemi has a dominant position in the server power supply ecosystem. That crown belongs to companies like Infineon, Monolithic Power Systems (MPS), and Vishay.
Now, let's talk about the hidden infrastructure. The Vera Rubin platform will require a fundamental shift in data center power architecture. The industry is moving from 12V to 48V bus voltages to reduce copper losses, and eventually to 800V or even 1.2kV for rack-level distribution. That shift opens the door for SiC and GaN, but the timeline is longer than the market assumes. The transition from 12V to 48V is already underway, but it's a gradual process. The move to high-voltage DC distribution is still years away, requiring new standards, new connectors, and new safety certifications. The market is pricing in a demand surge that won't materialize until 2027 or 2028.
Yield wasn't the only metric that mattered. The semiconductor report I analyzed earlier this week highlighted a critical point: the yield rates for 8-inch SiC wafers are still significantly lower than 6-inch. Wolfspeed's Mohawk Valley fab has been running at an estimated 60-70% yield for its 8-inch line, compared to 80%+ for 6-inch. That means every wafer costs more, and the capacity ramp is slower. The market is ignoring this. It's betting on volume, not on the cost structure. If yields don't improve, the economics of SiC for AI data centers will be marginal, and the companies that invested billions in 8-inch capacity will struggle to generate returns.
Let me bring in a personal experience. In 2020, during DeFi Summer, I wrote a series of reports on the yield farming craze. I interviewed women in Lagos and Rio who were using Aave to earn yields that their local banks could never offer. The narrative was that DeFi would democratize finance. But I also warned that the underlying infrastructure—the gas fees, the liquidity fragmentation, the oracle risks—was not ready for mass adoption. The same dynamic is playing out here. The narrative is that AI demand will save power semiconductor companies. But the infrastructure—the yields, the supply chain bottlenecks, the geopolitical risks—is not ready.
Contrarian: The Blind Spots the Market Refuses to See
Here's the contrarian angle that most analysts are missing: the real beneficiaries of the Vera Rubin power wave are not the IDMs that make the power transistors, but the companies that make the digital power controllers, the GaN-on-Si epiwafers, and the advanced packaging solutions. Think of it this way: the GPU is the star, but the power supply is the supporting cast. The supporting cast is about to be disrupted by a new set of players.
First, GaN. The dominant GaN power semiconductor companies are startups like Navitas Semiconductor, EPC (Efficient Power Conversion), and GaN Systems (now part of Infineon). These companies have been working for years to crack the data center market. Navitas, for example, has a GaN-based power IC that integrates the driver and the transistor in a single package, reducing parasitic inductance and improving efficiency. The Vera Rubin platform, with its high current demands, is a perfect fit for GaN. But Navitas is not a household name, and its stock has been battered by the bear market. The market is sleeping on it.
Second, the supply chain for GaN is fragile. The raw material for GaN wafers is gallium, a byproduct of aluminum refining. China controls over 80% of global gallium production. In 2023, China imposed export controls on gallium, citing national security. The price of gallium spiked, and the supply chain for GaN epitaxial wafers tightened. The companies that can secure gallium supply—or develop alternative substrates—will have a massive advantage. The market is ignoring this geopolitical risk. It's betting on a smooth ramp, but the reality is that the GaN supply chain is a single point of failure.
Third, the packaging. The power modules for AI servers are moving from wire-bonded packages to copper clip and silver sintering technologies. This is a specialized area dominated by companies like Heraeus, Indium, and a few boutique OSATs. The IDMs have their own packaging capabilities, but they are optimized for automotive and industrial, not for the high-frequency, high-density requirements of server power supplies. The shift to vertical power delivery—where the power stage is integrated directly under the GPU—will require a new generation of packaging that is not yet mature. The companies that crack this will be the unsung heroes.
Yield wasn't the only metric that mattered. The second hidden insight is about the narrative itself. The market is using the SiC story to hype AI demand, but the two are only loosely connected. It's a classic case of narrative arbitrage: the most exciting story (AI) is attached to the most exciting technology (SiC), even if the technical fit is weak. This is exactly what happened with NFTs and the metaverse in 2021: the narrative detached from the fundamentals, and when the hype faded, the floor collapsed.
Let me give you a specific example from my own experience. In 2022, I covered the LIQUID collapse. I saw how the narrative of "algorithmic stability" blinded investors to the flaws in the underlying mechanism. The same thing is happening here. The narrative of "AI power demand" is blinding investors to the flaws in the supply chain, the yield issues, and the competitive dynamics. The stock prices of Wolfspeed, STMicro, and onsemi have rallied on the back of a story that is only partially true. The real demand is there, but it's not a rising tide that lifts all boats. It's a selective current that rewards the companies with the right technology, the right supply chain, and the right customer relationships.
Takeaway: The Next Narrative Pivot
So where does this leave us? The market is currently pricing in a bright future for power semiconductor companies based on AI demand. But the narrative is ahead of the fundamentals. The next pivot will come when the market realizes that the real beneficiaries are the GaN startups, the digital power IC designers, and the advanced packaging specialists. The big IDMs will benefit, but not as much as the market expects. The takeaway for investors and crypto enthusiasts alike is the same: narratives are powerful, but they are not reality. The truth is always in the details—the yield rates, the supply chain dependencies, the customer concentration. The power chip narrative is a ZK proof that has not been verified: it looks convincing on the surface, but the underlying proof is still being challenged.
Yield wasn't the only metric that mattered. The next time you see a rally in power chip stocks, ask yourself: who is really selling the picks and shovels? The answer might not be the companies you think.