My entire job for the last decade was to create those spice model files for transistors that you put up a screen shot of. It's a degenerate job of editing text files for a living, and dealing with simulators with terrible documentation. Thank god for Substack.
The broader power electronics/semiconductor market (projected to reach ~$65 billion by 2036 at a 10% CAGR) relies on AI/data centers for a significant portion of incremental upside. A slowdown here would moderate the category's growth rate, especially for AI-optimized SiC/GaN and intelligent power modules. How much of this optimism is already priced into WOLF. AIXA and IFX?
I am not smart enough to comment on power semis, so here is Claude 😂:
Good read — I've got the full article. Given your deep semiconductor knowledge, I'll give you a thorough breakdown. Overall it's **largely technically sound**, but there are a few places worth scrutinising.
---
## ✅ What's Correct
**SiC vs GaN tradeoffs**
The core framing is accurate. GaN enables higher switching frequencies at lower voltages; SiC handles higher voltages and power density. This is well-established physics. The article correctly notes GaN is currently limited to ~650V class for lateral devices while SiC extends to 1200V, 1700V, and beyond.
**Rds_on as the key figure of merit**
Correct — this is the dominant driver of conduction losses. The article's emphasis on it over headline Vds ratings is the right engineering instinct.
**Gate charge and parasitic capacitance**
The explanations of Qg, Qoss, and Eoss are technically accurate and correctly linked to switching losses and efficiency.
**Negative gate drive requirement for GaN**
Accurate. Lateral GaN HEMTs (depletion-mode devices) have very low threshold voltages and can suffer spurious turn-on without a negative gate bias during off-state. This is a genuine design pain point that TI and Navitas address through integration.
**TI and Navitas GaN integration advantage**
Correct that integrating the gate driver, negative bias generation, and protection circuits into the same package significantly reduces design complexity. This is the "GaNFast" approach Navitas pioneered and TI has followed. It's a real differentiation.
**Infineon's SiC strength at 1200V**
Generally well-supported. Infineon's CoolSiC MOSFET technology is widely regarded as among the best-in-class, particularly in 1200V. Their trench-based SiC MOSFETs have industry-leading Rds_on per unit area.
**Wolfspeed 10kV SiC**
The device exists and the characterisation of it as niche but potentially transformative is fair. The Rds_on sensitivity to temperature is a real challenge, and the article correctly flags that custom packaging and cooling are necessary.
**Solid-State Transformers (SST) and reactive power**
The underlying concepts are correct. SSTs do offer active power factor correction, load buffering, and reactive power management that passive transformers cannot. The argument that AI datacenters create destabilising transient loads is valid — the PyTorch flag reference is real and well-known in the industry.
**Multi-sourcing complexity**
The point that qualifying a new transistor in an existing design is non-trivial (especially for efficiency-sensitive applications like datacenters) is accurate and often underappreciated.
**Aixtron MOCVD near-monopoly for GaN**
Correct. Aixtron dominates GaN epitaxy tool supply (MOCVD reactors), which is a genuine bottleneck. This is a legitimate semicap angle.
---
## ⚠️ What's Questionable or Needs Nuance
**"GaN is limited to 650V only"**
This is partially outdated. Vertical GaN is actively being developed by multiple players (including ON Semi, Transphorm, and others) and is designed to breach the 650V ceiling. The article itself acknowledges ON Semi's vertical GaN but dismisses it as "hopes and dreams on slides." That may be too harsh — Transphorm, for instance, has been selling vertical-architecture GaN at 650V and is working on 900V+ devices. The ceiling is more like a current commercial limit than a hard physical one.
**Navitas SiC being "shit"**
Directionally fair for now, but Navitas acquired GeneSiC in 2022 specifically to build a credible SiC portfolio. GeneSiC had strong technology, particularly for higher-voltage SiC. Calling Navitas SiC categorically poor without engaging with the GeneSiC heritage is a bit superficial.
**"Nobody cares if both GaN and SiC come from the same company"**
This is probably true at the discrete transistor design level, but less true at the system/module level where vendors selling complete reference designs and power modules benefit from offering a unified portfolio with validated cross-family designs. It's not purely a silly marketing claim.
**Infineon patent case against Innoscience**
Mentioned briefly and presented as straightforwardly bullish for Infineon/GloFo. The reality is patent disputes in semiconductors take years and outcomes are uncertain. Innoscience is also a large, well-capitalised Chinese IDM — treating the patent case as a settled negative for them is premature.
**STM buying GaN wafers from Innoscience**
Plausible and has been reported in trade press, but the article presents this as confirmed fact. STM has its own GaN-on-Si development as well. The supply relationship may be partial or transitional rather than a total dependency.
**"Traditional transformers are basically inductors and capacitors"**
A slight oversimplification for the purposes of this article. Traditional power transformers are magnetically coupled inductors — there are no capacitors in the core design. The article may conflate the power transformer with the broader substation equipment (including capacitor banks for power factor correction). Minor but worth noting.
**SST "never took off because it was too expensive"**
True but incomplete. SSTs also face reliability and maturity concerns compared to passive transformers that have a 100-year reliability track record. Utilities are conservative buyers. The cost barrier is real but it's not the only barrier.
---
## Summary Verdict
| Area | Assessment |
|---|---|
| SiC/GaN physics | ✅ Solid |
| Datasheet analysis methodology | ✅ Sound |
| Infineon SiC leadership | ✅ Accurate |
| TI/Navitas GaN integration | ✅ Accurate |
| GaN 650V ceiling | ⚠️ Oversimplified |
| Navitas SiC dismissal | ⚠️ Too harsh (ignores GeneSiC) |
| Innoscience patent/supply claims | ⚠️ Stated with more certainty than warranted |
The author clearly has genuine technical depth — the datasheet walkthrough and circuit intuition sections are better than most sell-side content on this topic. The investment framing is reasonable but benefits from the bullishness being occasionally stronger than the evidence strictly supports.
Fking good article again! Ur promoting Delta electronics sounds good, for they are negotiating the VPD license with $VICR, who is the superstar in $CBRS. But for NVTS, really disagreed with u, pls never doubt the tech capabilities and progress of Innoscience, u also mentioned they are selling GaN wafers to STM & ON semi ! Since Innoscience already came in game of NVDA 800V, hot competition is coming lol. Really concerning NVTS’ gross margin performance in the coming quarters lol!
"Only two claims of the ’481 patent were found valid and infringed, and only with respect to legacy products that are no longer manufactured or sold. As a result, the associated import ban has NO practical impact on Innoscience’s U.S. business. Innoscience will continue supplying its existing GaN power products to U.S. and worldwide customers uninterrupted."
(I'm not an expert by any means. somebody correct me if I'm wrong lol)
I am more interested in the VPD license with $VICR angle of power. Anyone else you hear that Vicr is negotiating with on VPD? A second or 3rd source may open up the design into the big Boyz and not just cars. What are your thoughts?
ITC case no: 337-TA-1370, VICR raised in CY23, won on 13Feb25; ITC case no: 337-TA-1484 , raised on 12Jan.26, pattern no: 12,395,087 , u can check it accordingly lol
I'm not an expert in power semis but to me personally Navitas looks most interesting. Here is my thinking -
1. Navitas seems to be the only one with a GaN IC platform including logic on the same die. They even have their own internal PDK for this AllGaN process. This seems a bit like how MPWR has its own custom BCD processes at fabs. Say what you want MPWR (I own it), but it was the newest company after NVDA to break into SMH last I looked (top 25 US listed names) and has had 10 years of 10-year rolling growth greater than 15% and comes close to clearing the rule of 40. That's a clear sign of a moat. In comparison Navitas is very early in its journey but unlike BCD which MPWR did not invent and others have, AllGaN seems unique to Navitas. Now TI can probably achieve something similar with package level integration but something where Navitas could have an edge perhaps in the future.
2. Focus and agility. All the others are bigger companies with many more fish to fry. This was probably another angle that MPWR exploited. Navitas founders are still around and seem to have a long history of having worked with each other.
3. Smaller cap more volatile name. Yes, this is a plus in my book. Works both ways and actually gives opportunities to buy cheaper in a drawdown. I'm perfectly happy to buy a small position and then buying more when it drops to tax loss sell higher lots later. As long as the story checks out. Small cap also means less intuitional coverage and lack of ability to buy big positions so individual investors have an edge.
Finally, none of this advice and I could be wrong in many ways. I'm fine with that and it will be a journey of following the story and figuring it out.
WOLF just had a violent short squeeze after the Citrini endorsement this week. Probably need to wait to cool off before new entry. Regarding Aixtron which ticker are you buying - AIXXF or AIIXY since AIXTRON SE is a European/non USA company?
First, I think it will be difficult for players to have real differentiation or moat for their SST products. In data center applications you need a box that converts from MV AC to 800V DC (ostensibly). But that's really the only need! The SST use case is much less sensitive to density than the in-rack, sidecar, or point of load power conversion; ostensibly all the different products' efficiencies will be in the same ballpark. So I think this just ends up looking like a commodity. Advantage goes to the companies that can get to market fastest, or are already established OEMs for the data center industry. I think the power semiconductor companies will benefit but I struggle to see which SST makers will capture lots of value.
Second, I buy that data centers are the killer app for SSTs, but what beyond that? There are good reasons why the SST dream has been stuck in limbo for the past decade (til the data center boom). I haven't done the market sizing math but I'm skeptical about broader adoption.
The entry barrier of HV GaN is much easier than high speed Serdes. actually there is no much moats for US players in GaN industry. U bullish $NVTS for u already buying in the ticker? So what their Q1 results? So what their Q2 guidance?
I stopped reading when you said the 2300V module from Infineon is cheating because it uses two transistors. Man, obviously each transistor has a max Vds of 2300V, they are just putting as many dies as they can fit in the package so you could get half-bridge, full-bridge depending on target Rdson. This is clearly out of your circle of competence.
Actually, Wolfspeed has used Carsem Malaysia as its back-end device packaging partner since 2018. The company is also OSAT to Infineon, STMicro, SiTime and a bunch of other analogs.
Same as Unisem - OSAT for MPWR, Renesas, Power Integration
This is the part of the AI buildout that gets underweighted. The market talks about chips and models, but the bottleneck keeps moving back to physical execution: power, interconnection, cooling, permitting, and uptime. The winners will not just be the companies with demand. They will be the ones that can turn megawatts into reliable compute without breaking the local system around them.
My entire job for the last decade was to create those spice model files for transistors that you put up a screen shot of. It's a degenerate job of editing text files for a living, and dealing with simulators with terrible documentation. Thank god for Substack.
The broader power electronics/semiconductor market (projected to reach ~$65 billion by 2036 at a 10% CAGR) relies on AI/data centers for a significant portion of incremental upside. A slowdown here would moderate the category's growth rate, especially for AI-optimized SiC/GaN and intelligent power modules. How much of this optimism is already priced into WOLF. AIXA and IFX?
Are you going to write about it?
I am not smart enough to comment on power semis, so here is Claude 😂:
Good read — I've got the full article. Given your deep semiconductor knowledge, I'll give you a thorough breakdown. Overall it's **largely technically sound**, but there are a few places worth scrutinising.
---
## ✅ What's Correct
**SiC vs GaN tradeoffs**
The core framing is accurate. GaN enables higher switching frequencies at lower voltages; SiC handles higher voltages and power density. This is well-established physics. The article correctly notes GaN is currently limited to ~650V class for lateral devices while SiC extends to 1200V, 1700V, and beyond.
**Rds_on as the key figure of merit**
Correct — this is the dominant driver of conduction losses. The article's emphasis on it over headline Vds ratings is the right engineering instinct.
**Gate charge and parasitic capacitance**
The explanations of Qg, Qoss, and Eoss are technically accurate and correctly linked to switching losses and efficiency.
**Negative gate drive requirement for GaN**
Accurate. Lateral GaN HEMTs (depletion-mode devices) have very low threshold voltages and can suffer spurious turn-on without a negative gate bias during off-state. This is a genuine design pain point that TI and Navitas address through integration.
**TI and Navitas GaN integration advantage**
Correct that integrating the gate driver, negative bias generation, and protection circuits into the same package significantly reduces design complexity. This is the "GaNFast" approach Navitas pioneered and TI has followed. It's a real differentiation.
**Infineon's SiC strength at 1200V**
Generally well-supported. Infineon's CoolSiC MOSFET technology is widely regarded as among the best-in-class, particularly in 1200V. Their trench-based SiC MOSFETs have industry-leading Rds_on per unit area.
**Wolfspeed 10kV SiC**
The device exists and the characterisation of it as niche but potentially transformative is fair. The Rds_on sensitivity to temperature is a real challenge, and the article correctly flags that custom packaging and cooling are necessary.
**Solid-State Transformers (SST) and reactive power**
The underlying concepts are correct. SSTs do offer active power factor correction, load buffering, and reactive power management that passive transformers cannot. The argument that AI datacenters create destabilising transient loads is valid — the PyTorch flag reference is real and well-known in the industry.
**Multi-sourcing complexity**
The point that qualifying a new transistor in an existing design is non-trivial (especially for efficiency-sensitive applications like datacenters) is accurate and often underappreciated.
**Aixtron MOCVD near-monopoly for GaN**
Correct. Aixtron dominates GaN epitaxy tool supply (MOCVD reactors), which is a genuine bottleneck. This is a legitimate semicap angle.
---
## ⚠️ What's Questionable or Needs Nuance
**"GaN is limited to 650V only"**
This is partially outdated. Vertical GaN is actively being developed by multiple players (including ON Semi, Transphorm, and others) and is designed to breach the 650V ceiling. The article itself acknowledges ON Semi's vertical GaN but dismisses it as "hopes and dreams on slides." That may be too harsh — Transphorm, for instance, has been selling vertical-architecture GaN at 650V and is working on 900V+ devices. The ceiling is more like a current commercial limit than a hard physical one.
**Navitas SiC being "shit"**
Directionally fair for now, but Navitas acquired GeneSiC in 2022 specifically to build a credible SiC portfolio. GeneSiC had strong technology, particularly for higher-voltage SiC. Calling Navitas SiC categorically poor without engaging with the GeneSiC heritage is a bit superficial.
**"Nobody cares if both GaN and SiC come from the same company"**
This is probably true at the discrete transistor design level, but less true at the system/module level where vendors selling complete reference designs and power modules benefit from offering a unified portfolio with validated cross-family designs. It's not purely a silly marketing claim.
**Infineon patent case against Innoscience**
Mentioned briefly and presented as straightforwardly bullish for Infineon/GloFo. The reality is patent disputes in semiconductors take years and outcomes are uncertain. Innoscience is also a large, well-capitalised Chinese IDM — treating the patent case as a settled negative for them is premature.
**STM buying GaN wafers from Innoscience**
Plausible and has been reported in trade press, but the article presents this as confirmed fact. STM has its own GaN-on-Si development as well. The supply relationship may be partial or transitional rather than a total dependency.
**"Traditional transformers are basically inductors and capacitors"**
A slight oversimplification for the purposes of this article. Traditional power transformers are magnetically coupled inductors — there are no capacitors in the core design. The article may conflate the power transformer with the broader substation equipment (including capacitor banks for power factor correction). Minor but worth noting.
**SST "never took off because it was too expensive"**
True but incomplete. SSTs also face reliability and maturity concerns compared to passive transformers that have a 100-year reliability track record. Utilities are conservative buyers. The cost barrier is real but it's not the only barrier.
---
## Summary Verdict
| Area | Assessment |
|---|---|
| SiC/GaN physics | ✅ Solid |
| Datasheet analysis methodology | ✅ Sound |
| Infineon SiC leadership | ✅ Accurate |
| TI/Navitas GaN integration | ✅ Accurate |
| GaN 650V ceiling | ⚠️ Oversimplified |
| Navitas SiC dismissal | ⚠️ Too harsh (ignores GeneSiC) |
| SST opportunity thesis | ✅ Directionally compelling, some nuance missing |
| Aixtron/GloFo angle | ✅ Legitimate |
| Innoscience patent/supply claims | ⚠️ Stated with more certainty than warranted |
The author clearly has genuine technical depth — the datasheet walkthrough and circuit intuition sections are better than most sell-side content on this topic. The investment framing is reasonable but benefits from the bullishness being occasionally stronger than the evidence strictly supports.
Fking good article again! Ur promoting Delta electronics sounds good, for they are negotiating the VPD license with $VICR, who is the superstar in $CBRS. But for NVTS, really disagreed with u, pls never doubt the tech capabilities and progress of Innoscience, u also mentioned they are selling GaN wafers to STM & ON semi ! Since Innoscience already came in game of NVDA 800V, hot competition is coming lol. Really concerning NVTS’ gross margin performance in the coming quarters lol!
https://www.innoscience.com/news/press-releases/10636
"Only two claims of the ’481 patent were found valid and infringed, and only with respect to legacy products that are no longer manufactured or sold. As a result, the associated import ban has NO practical impact on Innoscience’s U.S. business. Innoscience will continue supplying its existing GaN power products to U.S. and worldwide customers uninterrupted."
(I'm not an expert by any means. somebody correct me if I'm wrong lol)
great article as always, btw
I am more interested in the VPD license with $VICR angle of power. Anyone else you hear that Vicr is negotiating with on VPD? A second or 3rd source may open up the design into the big Boyz and not just cars. What are your thoughts?
ITC case no: 337-TA-1370, VICR raised in CY23, won on 13Feb25; ITC case no: 337-TA-1484 , raised on 12Jan.26, pattern no: 12,395,087 , u can check it accordingly lol
Who are the main SST players?
I'm not an expert in power semis but to me personally Navitas looks most interesting. Here is my thinking -
1. Navitas seems to be the only one with a GaN IC platform including logic on the same die. They even have their own internal PDK for this AllGaN process. This seems a bit like how MPWR has its own custom BCD processes at fabs. Say what you want MPWR (I own it), but it was the newest company after NVDA to break into SMH last I looked (top 25 US listed names) and has had 10 years of 10-year rolling growth greater than 15% and comes close to clearing the rule of 40. That's a clear sign of a moat. In comparison Navitas is very early in its journey but unlike BCD which MPWR did not invent and others have, AllGaN seems unique to Navitas. Now TI can probably achieve something similar with package level integration but something where Navitas could have an edge perhaps in the future.
2. Focus and agility. All the others are bigger companies with many more fish to fry. This was probably another angle that MPWR exploited. Navitas founders are still around and seem to have a long history of having worked with each other.
3. Smaller cap more volatile name. Yes, this is a plus in my book. Works both ways and actually gives opportunities to buy cheaper in a drawdown. I'm perfectly happy to buy a small position and then buying more when it drops to tax loss sell higher lots later. As long as the story checks out. Small cap also means less intuitional coverage and lack of ability to buy big positions so individual investors have an edge.
Finally, none of this advice and I could be wrong in many ways. I'm fine with that and it will be a journey of following the story and figuring it out.
WOLF just had a violent short squeeze after the Citrini endorsement this week. Probably need to wait to cool off before new entry. Regarding Aixtron which ticker are you buying - AIXXF or AIIXY since AIXTRON SE is a European/non USA company?
Sounds to me that the best way to play this themes is IFX and to a certain degrees also AIXA?
I am more bearish on the SST market.
First, I think it will be difficult for players to have real differentiation or moat for their SST products. In data center applications you need a box that converts from MV AC to 800V DC (ostensibly). But that's really the only need! The SST use case is much less sensitive to density than the in-rack, sidecar, or point of load power conversion; ostensibly all the different products' efficiencies will be in the same ballpark. So I think this just ends up looking like a commodity. Advantage goes to the companies that can get to market fastest, or are already established OEMs for the data center industry. I think the power semiconductor companies will benefit but I struggle to see which SST makers will capture lots of value.
Second, I buy that data centers are the killer app for SSTs, but what beyond that? There are good reasons why the SST dream has been stuck in limbo for the past decade (til the data center boom). I haven't done the market sizing math but I'm skeptical about broader adoption.
The entry barrier of HV GaN is much easier than high speed Serdes. actually there is no much moats for US players in GaN industry. U bullish $NVTS for u already buying in the ticker? So what their Q1 results? So what their Q2 guidance?
I stopped reading when you said the 2300V module from Infineon is cheating because it uses two transistors. Man, obviously each transistor has a max Vds of 2300V, they are just putting as many dies as they can fit in the package so you could get half-bridge, full-bridge depending on target Rdson. This is clearly out of your circle of competence.
Yeah,me too lol
Great read
Thank you for the work you put into these. Really great read
Actually, Wolfspeed has used Carsem Malaysia as its back-end device packaging partner since 2018. The company is also OSAT to Infineon, STMicro, SiTime and a bunch of other analogs.
Same as Unisem - OSAT for MPWR, Renesas, Power Integration
Any thoughts on Renesas? It might be behind in product specs, but there's a shortage and stock is at half of peer multiples
This is the part of the AI buildout that gets underweighted. The market talks about chips and models, but the bottleneck keeps moving back to physical execution: power, interconnection, cooling, permitting, and uptime. The winners will not just be the companies with demand. They will be the ones that can turn megawatts into reliable compute without breaking the local system around them.
This was an awesome read thx. Already own WOLF. Gonna add IFX/AIXA.