Top Gallium Nitride Stocks 2026: AI Power Goes GaN

Gallium nitride's first killer app was making your phone charger less brick-like.

Nice trick. Not much of an investing thesis.

The 2026 version is more interesting. Power GaN is moving from low-watt consumer electronics into the places where efficiency is worth real money: AI data centers, EV power conversion, industrial drives, energy infrastructure, and eventually robotics. Yole Group now sees the power GaN device market growing at a 35% CAGR to $3.5 billion in 2031, with data centers and automotive among the fastest-growing end markets.

The AI catalyst deserves special attention. Next-generation racks are shifting toward higher-voltage DC distribution, including 800V architectures, because hauling gigantic amounts of power around at lower voltages means thicker copper, higher losses, and more heat. GaN's fast switching is useful in the middle and lower-voltage stages of that power tree; silicon carbide tends to compete harder at the highest-voltage front end.

That does not mean every company with a GaN slide deck is a GaN stock. The clean public universe is surprisingly thin. Transphorm, once one of the obvious U.S.-listed names, was acquired by Renesas in 2024.^[1] Meanwhile, giants like Infineon, Texas Instruments, and onsemi can win major GaN sockets without GaN becoming a huge percentage of company revenue.

We're slicing Gallium Nitride stocks by how the theme can hits the stock: high-beta GaN, integrated power, foundry exposure, and private bellwethers. It focuses on power GaN, not the separate RF GaN market used in radar and communications.

These are the names where a successful GaN adoption cycle can meaningfully change the company. High exposure, high reward.

Navitas SemiconductorNASDAQ: NVTS

Navitas Semiconductor is the closest thing the U.S. market has to a high-beta public bet on next-generation power semiconductors. The company sells GaNFast GaN devices and GeneSiC silicon-carbide products, but management has deliberately walked away from low-end mobile and consumer business to chase AI data centers, grid infrastructure, and other higher-power markets.

The transition is visible but not finished. Q2 2026 revenue was $10.5 million, still below the year-earlier quarter, yet it grew 22% sequentially, while high-power markets grew more than 50% year over year. Management expects AI infrastructure to represent more than one-third of sales by year-end and says selected hyperscaler and XPU programs should ramp in 2027.

Two fresh developments make NVTS especially worth watching. First, Navitas says its first U.S.-manufactured Gen 5 GaNFast wafers from GlobalFoundries' 200mm GaN-on-silicon line begin shipping in September 2026, with customer samples planned before year-end. Second, Navitas signed an agreement in August to acquire Claros for an estimated $232.8 million, adding vertical power delivery and integrated voltage-regulator technology. The strategic logic is clean: sell more of the power path from grid to processor. The risk is just as clean: a small revenue base is trying to digest a meaningful acquisition while still proving its AI pivot.

Wolfspeed filed a patent-infringement suit in July covering Navitas GaN and SiC product families. For this stock, demos are no longer the scoreboard. Watch production qualifications, actual 2027 ramps, gross-margin durability, and litigation progress.

InnoscienceHKEX: 02577

Innoscience is the scale story. It describes itself as the first IDM to mass-produce GaN-on-silicon on 8-inch wafers at large scale, with products spanning 15V to 1200V. In H1 2026, revenue rose 50.6% to RMB833.6 million, gross margin reached 11.6%, and operating cash flow turned positive. It still posted a RMB309.1 million net loss, so "scaled" does not yet mean "finished."

The mix is moving in the right direction for this theme. H1 shipments into AI and data-center applications rose 183% year over year; automotive-grade chip shipments rose 103%. Innoscience says it secured 747 new customer design-ins in the half, with more than 100 high- and medium-voltage products for 800V/48V data-center architectures designed in at more than 20 major cloud-service providers. It has also disclosed significant design-ins on Google AI hardware platforms and a supply agreement.

Then there's the patent war. Infineon says a U.S. ITC determination, confirmed after the presidential review period, bars Innoscience products found to infringe an Infineon GaN patent; Innoscience says its redesigned current U.S. commercial products fall outside the affected claims and can continue selling. The companies are also fighting in Europe and China.

That makes Innoscience a very 2026 semiconductor story: huge operating momentum, huge manufacturing ambition, and a legal/IP variable you cannot hand-wave away. Watch overseas revenue, customer concentration, margin expansion, and whether the litigation changes what can actually ship in major markets.

These companies are less "bet the farm on GaN" and more "own the power socket regardless of which wide-bandgap material wins." That lowers theme sensitivity, but it also lowers the odds that one delayed qualification wrecks the whole story.

For investors, this is the segment where GaN adoption can be right even if the pure-play thesis is wrong.

Infineon Technologies (FSE: IFX / OTCQX: IFNNY)

Infineon Technologies may be the most important company on this list precisely because it does not need GaN to be exotic. It sells silicon, silicon carbide, and GaN, giving customers a menu instead of a religion.

The manufacturing angle is the big one. Infineon has pushed GaN onto 300mm wafers, a move it says can yield roughly 2.3 times as many chips per wafer as 200mm production and improve cost competitiveness. Customer sampling began in late 2025. In 2026, Infineon also joined NVIDIA's MGX ecosystem for 800V AI power delivery.

If 300mm manufacturing does what it is supposed to do, this is good for GaN adoption and potentially uncomfortable for smaller suppliers. Lower device cost expands the market while also squeezing anyone whose moat was partly "GaN is hard to manufacture cheaply."

Watch 300mm yield/ramp commentary, AI power revenue, and the pace at which GaN moves from premium designs into ordinary high-volume power supplies.

STMicroelectronicsNYSE: STM

STMicroelectronics is another giant where GaN can matter without owning the whole thesis. Its 2026 PowerGaN push is aimed squarely at high-demand power: AI servers, robotics, industrial systems, smart-grid converters, and advanced consumer gear.

In May, ST put seven new 700V enhancement-mode PowerGaN transistors into production. In March, it expanded its NVIDIA-linked 800VDC data-center power portfolio with 12V and 6V architectures alongside its existing 800V-to-50V stage. That's the important part: ST isn't just selling a transistor. It's trying to own more of the AI power tree.

The obvious catch is dilution. ST is enormous, and GaN is one technology inside a sprawling semiconductor portfolio. Watch PowerGaN product breadth, AI power design wins, and whether GaN becomes a material STPOWER growth driver instead of another very good option in a very large catalog.

ROHMTSE: 6963

ROHM is taking the "make GaN boring enough to manufacture at scale" route. Its 650V EcoGaN devices already have a real AI-server reference point: Murata Power Solutions adopted them in a 5.5kW AI server power supply, and ROHM was still showcasing that design-in at APEC 2026.

In February 2026, ROHM said it would license TSMC's GaN process technology into ROHM Hamamatsu, building an end-to-end in-group production system targeted for 2027. The stated demand drivers are exactly the ones this watchlist cares about: AI servers and EVs. ROHM is effectively trading some foundry dependence for more control over GaN supply and manufacturing.

Watch the technology-transfer timeline, 650V volume growth, and whether ROHM can turn its broader power-and-analog portfolio into system-level GaN wins.

Power IntegrationsNASDAQ: POWI

Power Integrations is easy to miss because it is not marketed like a frontier-tech lottery ticket. It is a high-voltage power-conversion specialist with an established business, cash generation, and a GaN platform called PowiGaN.

It belongs here because of voltage. In August 2026, Power Integrations demonstrated 2200V GaN technology and said its 1700V GaN is already in design-ins for data-center auxiliary power, while 1250V devices can address parts of the 800VDC main power path. That's important because the lazy mental model says "GaN for lower voltage, SiC for high voltage." POWI is trying to move that border.

The stock is less binary than NVTS, but the GaN question is still sharp: can Power Integrations turn very-high-voltage technical milestones into material design wins before larger IDMs close the gap? Watch customer adoption in AI infrastructure, renewables, storage, and HVDC, not just the maximum voltage number in the press release.

Renesas ElectronicsTSE: 6723

Renesas Electronics bought Transphorm in 2024 and inherited its high-voltage SuperGaN platform. Since then, the company has kept filling in the stack: new 650V parts, a 650V-class bidirectional GaN switch for solar/AI/EV applications, and system-level power designs around controllers and drivers.

The more interesting 2026 move is at low voltage. Renesas signed a licensing and second-sourcing agreement with private GaN pioneer EPC, gaining access to EPC's low-voltage eGaN technology and working toward internal wafer-fabrication capability for those products. That gives Renesas a more complete GaN ladder instead of a single-voltage product island.

This is an integration bet. Renesas does not have to win by selling the world's flashiest transistor; it can win by bundling GaN with the control silicon, drivers, MCUs, and reference designs customers already use. Watch whether that system advantage turns into AI and industrial design wins that matter at company scale.

onsemiNASDAQ: ON

onsemi is running two GaN experiments at once.

The lower-risk path is lateral GaN. Its GaNEXUS portfolio began sampling in June 2026 across 40V to 650V, and onsemi is working with GlobalFoundries on 200mm 650V GaN-on-silicon manufacturing. It also has a collaboration with Innoscience for parts of the portfolio.

The higher-upside path is vertical GaN: current flowing vertically through GaN-on-GaN rather than across a GaN-on-silicon layer. onsemi says 700V and 1200V vertical-GaN devices are sampling to early-access customers, with volume production targeted for late 2026.

If vertical GaN scales with good yield, reliability, and cost, it could attack applications usually handed to silicon carbide. If it does not, it becomes another beautiful semiconductor technology that engineers loved and accountants killed. Watch late-2026 production status, named customer qualifications, and whether vGaN stays a science project or turns into revenue.

Texas InstrumentsNASDAQ: TXN

Texas Instruments is the least pure GaN exposure on the list. That's also the point.

In March 2026, TI and NVIDIA outlined an 800VDC data-center power architecture that includes an 800V-to-6V DC/DC power-delivery board using TI integrated GaN power stages. TI cited 97.6% peak efficiency and power density above 2,000 W/in³ for the design.

Nobody should buy TXN because "GaN goes up." GaN is too small a piece of the whole company. The watchlist thesis is that TI can capture a broad slice of the power-management bill of materials as AI racks re-architect: controllers, isolation, analog, and GaN power stages together.

For a conservative investor, that dilution is a feature. For someone looking for maximum GaN torque, it is a bug.

GaN needs fabs, process technology, qualification, and boring old manufacturing scale. The foundry layer can benefit even when chip designers fight over whose transistor is prettier.

GlobalFoundriesNASDAQ: GFS

GlobalFoundries is the picks-and-shovels name here. Its Burlington, Vermont site is the only U.S.-based 200mm facility focused on GaN-on-silicon production, and its power-GaN platform targets data centers, automotive, industrial systems, and fast charging.

Navitas said on September 1 that first wafers for its U.S.-made Gen 5 GaNFast family are scheduled to ship in September, with internal samples in October and strategic customer samples before year-end. onsemi is also using GF's 200mm e-mode GaN process for 650V products.

GFS will never give you pure GaN exposure. What it can give you is exposure to more companies deciding they need a qualified, resilient GaN manufacturing option. Watch additional customers, volume ramps, and how quickly the platform broadens beyond the first announced partners.

Public GaN investing has an annoying problem: some of the most technically important companies are still private. These three matter because public companies are licensing from them, partnering with them, or watching them push into applications the public market has not fully priced yet.

Efficient Power Conversion (Private)

Efficient Power Conversion (EPC) is one of the foundational names in enhancement-mode GaN. Its sweet spot is lower-voltage, high-frequency power conversion: AI server power, robotics, motor drives, and other applications where switching speed and tiny form factors matter.

In early 2026, EPC put its Gen 7 40V transistor into mass production and then signed the licensing/second-source deal with Renesas. That agreement is the useful signal. A giant semiconductor company isn't just buying parts; it's licensing EPC technology and helping establish another manufacturing path for it.

For public investors, EPC is a bellwether for how valuable low-voltage GaN IP becomes as power delivery moves closer to the processor and inside robotic actuators.

Cambridge GaN Devices (Private)

Cambridge GaN Devices is building ICeGaN, an enhancement-mode GaN platform designed to make the devices easier to drive and integrate. The company raised a $32 million Series C in 2025 to expand internationally and push further into higher-power data-center, industrial, and automotive markets.

In 2026, CGD entered a long-term collaboration with NXP. NXP plans to build GaN-based system solutions using CGD products and gain early access to next-generation CGD technology, targeting automotive, industrial, and data-center applications.

Watch whether that relationship becomes production programs. Private-company partnerships are cheap to announce. Qualified automotive and data-center sockets are not.

VisIC Technologies (Private)

VisIC Technologies is going after one of GaN's hardest prizes: the main EV traction inverter. Its D³GaN platform is positioned as a lower-loss, potentially lower-cost alternative to silicon carbide in high-power automotive drivetrains.

In late 2025, VisIC announced a $26 million second closing of its Series B round, with Hyundai Motor and Kia joining as strategic investors. That is a stronger signal than a generic venture round because the strategic backers are the kind of customers whose vehicle platforms determine whether traction GaN gets real volume.

This is still a qualification-and-reliability story. Watch OEM validation, inverter partnerships, and any evidence of mass-production vehicle programs.

The first failure mode is sneaky: GaN wins, but GaN stocks don't.

If Infineon gets 300mm economics working, GlobalFoundries ramps 200mm capacity, Innoscience keeps scaling, and large IDMs pile in, device prices can fall fast. That is wonderful for engineers and end customers. It can be brutal for smaller vendors that need premium pricing to fund R&D and operating losses. In other words, a bigger market does not automatically mean better margins.

The second failure mode is architectural. Much of the current excitement assumes AI racks keep moving toward 800VDC and that GaN captures attractive stages in that power tree. Navitas itself warns that its opportunity depends on adoption of those new system architectures. If data-center designs standardize more slowly, silicon improves enough, or SiC wins more of the voltage range than expected, the GaN revenue ramp slides to the right.

Then add the thing semiconductor investors love most: patent litigation. Innoscience versus Infineon is active across multiple jurisdictions, and Wolfspeed has sued Navitas. The technical winners can still lose time, money, customers, or geographic access in court.

So the four things to watch are boring on purpose: production qualifications, shipment ramps, gross margins, and legal freedom to operate. Those will matter more than another efficiency demo at a trade show.

The next 12–18 months should answer whether GaN is graduating from "fast-charger technology" to a mainstream power-platform technology.

The biggest catalyst is AI. Watch 2027 production ramps tied to 800V rack architectures, especially at Navitas, Innoscience, Infineon, STMicroelectronics, ROHM, TI, onsemi, and Renesas. The key distinction is design-in versus shipment. A design-in says an engineer likes the part. Volume revenue says purchasing, qualification, yield, reliability, and economics all survived contact with reality.

The second catalyst is manufacturing. GlobalFoundries is starting the Navitas U.S. 200mm GaN ramp now. Infineon is pushing 300mm GaN to lower cost. ROHM is bringing licensed TSMC GaN process technology into its own production system. onsemi is targeting late-2026 volume production for vertical GaN. Renesas and EPC are building a new low-voltage manufacturing path. Those are not versions of the same story; they are different bets on how GaN becomes cheap and dependable enough to stop being special.

The third catalyst is the voltage ceiling. Power Integrations is pushing GaN into 1700V and 2200V-class territory, while onsemi's vertical GaN targets 700V/1200V. If those technologies earn high-volume customer trust, the old "GaN below, SiC above" map gets messier and more investable.^[2]

We break it down like this:

• Highest GaN sensitivity: Navitas, Innoscience.
• Best positioned to commoditize GaN: Infineon, STMicroelectronics, ROHM, GlobalFoundries, onsemi.
• Most interesting high-voltage boundary push: Power Integrations, onsemi.
• Best system-level diversified exposure: STMicroelectronics, ROHM, Renesas, Texas Instruments.
• Private tells to watch: EPC for low-voltage AI/robotics, CGD for integrated GaN, VisIC for EV traction.

GaN doesn't need to replace silicon or silicon carbide everywhere to make investors money. It only needs to become the obvious answer in a few very large, very power-hungry sockets.

The next phase is about finding out which companies turn that physics advantage into repeatable manufacturing economics.