The AI infrastructure investment wave is sweeping the globe; however, the market is now entering a new turning point. In its latest issue of “Semis Memo: Supply Chain Inheritance,” market research firm Citrini Research pointed out that the investment logic for AI infrastructure has evolved from the intuitive thinking of “buying GPUs” to the more complex stage of “re-pricing power infrastructure and supply chains.”
This article will deep-dive into Citrini’s core views and explain how investors should build their portfolios based on them.
From the first wave to the second wave: how has AI supply chain investment logic changed?
Citrini first reviewed the logic behind the first wave of AI infrastructure deals: LLMs need GPUs; the growth in computing capacity drives demand for optical interconnects; and ultimately, AI demand flows to memory suppliers. This logic is intuitive and easy to understand, but Citrini believes the market is now undergoing a fundamental change.
To truly outperform the market in the coming years, investors must understand not the “obvious first-order beneficiaries,” but the second-order bottlenecks in the supply chain:
How will the power architecture be upgraded? Which components will become the new points of supply-demand tightness? Which production capacity or supply chains have already been expanded by other industries and are simply waiting for AI to take over?
What is Supply Chain Inheritance?
The most important part of this report is the core concept of “Supply Chain Inheritance.” Citrini’s view and logic are as follows:
AI data centers are moving toward 800V DC rack power architectures to meet ever-increasing power density demands.
The key materials relied on by this architecture—“wide-bandgap semiconductors”—mainly include silicon carbide (SiC) and gallium nitride (GaN). Their related capacity and supply chains have already been heavily pushed forward and are fairly mature thanks to the electric vehicle and solar power industries.
Therefore, when AI capex becomes the next wave of demand, it does not need to wait for EVs or solar power to recover in sync—AI can directly “inherit” these existing manufacturing ecosystems and cost-curve advantages.
Citrini emphasizes that weakness in EV and solar demand does not mean SiC and GaN supply chains lose their value. AI may provide stronger second-wave demand, allowing these key raw material suppliers to be re-priced. Examples include Ajinomoto, a seasoning manufacturer and the world’s largest ABF embedded layer thin-film supplier, and the well-known bathroom fixture brand TOTO, among others.
(ABF substrate price hikes are coming! Ajinomoto plans to raise prices by 30%, and the three Taiwan-listed substrate players set new April revenue highs first)
Upstream bottlenecks: long lead times for traditional transformers—solid-state transformers (SST) as the solution
Citrini points out that as AI data center construction accelerates, the delivery lead times of traditional iron-core transformers have become one of the key bottlenecks for hyperscale cloud providers expanding at massive scale.
It proposes “Solid-State Transformers (SST)” as the next potential technological inflection point: SST uses SiC high-frequency switching to directly convert medium-voltage AC into 800V DC. Compared with traditional equipment, it offers advantages such as shorter lead times, smaller size, and support for bidirectional power flow. Major companies in the U.S. such as Eaton (NYSE: ETN), Swedish ABB (SWX: ABBN), and German Siemens (ETR: SIE) are also advancing in this area.
However, Citrini also notes that the majority of targets that can truly benefit purely are still mostly in private placements or startups, and are not yet fully reflected in public-market pricing.
Downstream bottlenecks: VRM faces the “thousand-amp problem”—the value contained in each GPU will rise significantly
Inside the rack, Citrini focuses on the “Voltage Regulator Module (VRM).” As GPU power keeps increasing and core voltages continue to drop, the VRM must output extremely high current at very low voltage while maintaining system stability and integrity.
This is similar to the “thousand-amp problem” in electromagnetics; its prediction is that by 2027 it will evolve into a “three-thousand-amp problem.”
Citrini believes the solution is to adopt multi-phase parallel power delivery. But this also means the number of components required per AI server will increase substantially, including controllers, inductors, and passive components—thereby raising the value embedded in each GPU or server.
(Guide: Fusion industry research report—will 2026 passive components replicate the storm of AI memory shortages?)
How to find investment opportunities? Citrini lays out a four-tier supply chain structure
Citrini divides investment opportunities in AI power infrastructure into four tiers, giving investors a systematic stock-picking framework:
Tier 1 (most core): wide-bandgap semiconductor materials (SiC/GaN) and related capital equipment. Citrini believes this segment has not yet been re-priced by the market through “AI value transfer,” and has higher potential for excess returns.
Tier 2: broadly defined power semiconductors and specialized VRM suppliers. Upgrades for each generation of AI servers will increase the demand density for these components.
Tier 3: power system components, including power supplies (PSU), busbars, interconnect modules, and more. Citrini believes this tier has already been relatively identified by the market, but demand is still growing.
Tier 4: data center power systems and medium-voltage infrastructure, including power supply modules, switching equipment, and SST replacement themes.
How to screen investment targets? Clues from the Citrini report
In the report, Citrini also mentions several individual stocks and cases. If you classify them according to its supply chain tiers and only pick Tier 1 and 2, you get the following list:
Wolfspeed (NYSE: WOLF): A typical AI supply-chain value transfer example for Tier 1. The company actively expanded SiC capacity during the electric vehicle trend, but demand did not meet expectations, leading to bankruptcy reorganization. Citrini believes its existing physical production capacity could regain value under AI infrastructure demand, and the pricing logic could shift toward becoming an “irreplaceable AI infrastructure supplier.”
Monolithic Power Systems (NASDAQ: MPWR): One of the main suppliers in the Tier 2 VRM market, viewed as a core beneficiary of AI server power modules.
Vicor (NASDAQ: VICR): Provides factorized and vertical power architecture solutions. This is another technical path for VRM—more flexible, but market penetration needs to be monitored.
STMicroelectronics (NYSE: STM) / Onsemi (NASDAQ: ON) / Infineon (ETR: IFX): All are major suppliers of SiC/GaN wafers and power semiconductors, and they also fit Citrini’s narrative logic that EV supply chains pivot to absorb AI demand.
How can investors build a portfolio using this report?
At the end of the piece, Citrini also provides investors with the following operational suggestions and logical principles:
Don’t just bet on AI chips—look for beneficiaries of 800V power lines: the real structural opportunities are in the power conversion paths, not only in compute power itself.
Prioritize wide-bandgap semiconductors and the power transmission chain (SiC/GaN, VRM, power conversion infrastructure): these suppliers’ positioning has historically been tied to electric vehicles and industrial demand, but AI capex is becoming a second source of demand.
Control risk by Tier layering: Tier 1 still has the greatest opportunity for re-pricing, but it is also more volatile. Tier 3 and 4 are relatively already priced in by the market, making them more suitable for steadier long-term allocations.
AI infrastructure is being redrawn
The most important takeaway from Citrini’s report is this: the investment direction for AI infrastructure has shifted from “who owns the most compute power” to “who can solve the real-world bottlenecks in power transmission and conversion.” Production capacity and complete supply chains left behind by the EV era are expected to be taken over and re-used by the AI industry in a way that the market has not yet priced.
For investors, the true opportunity for excess returns may require first putting aside the hot AI flagship stocks in hand, and instead looking for key power infrastructure and semiconductor material suppliers that are quietly waiting for the “AI supply-chain value transfer” wave to arrive.
(This article is compiled from a Citrini Research report and does not constitute investment advice.)
Where will the next AI infrastructure wave be? Citrini report reveals “SiC, GaN, and power infrastructure” as a new investment direction. First appeared on Chain News ABMedia.
