Enphase Enters the SST Race With GaN BDS-Based Distributed Power Platform 

Enphase Energy, the world’s leading supplier of microinverter-based solar and battery systems, is bringing its distributed power architecture to AI data centers with the IQ Solid-State Transformer (IQ SST). Enphase pioneered the solar microinverter and is the first to commercialize a microinverter built on gallium nitride bi-directional switch (GaN BDS) technology. Announced in April 2026, the platform puts Enphase in direct competition with a growing field of SST startups—including DG Matrix, Heron Power, and Amperesand—all racing to solve the same problem: how to deliver megawatt-scale power to next-generation AI racks efficiently, reliably, and fast enough to handle workloads that swing from idle to full power several times per second. 

Where competitors have largely built on silicon carbide (SiC), Enphase is using GaN-based distributed power conversion and drawing on 20 years of experience with nearly 90 million microinverters shipped to make its case. Full system demonstrations are targeted for late 2026, with customer pilots planned for 2027 and volume shipments expected in 2028.

SSTs are a nascent technology that directly target the two main components that currently connect the medium-voltage (MV) 13.8 kV and 34.5 kV grid to the compute rack: line-frequency transformers (LFTs) and sidecar racks. The SST takes grid power and converts it directly to low-voltage (LV) 800 VDC (Figure 1). Infineon estimates that SSTs will replace a portion of the conventional transformer market—a market currently valued at $15 billion—specifically small power transformers, with SSTs expected to reach a market value of $1 billion by 2030. 

Most SST manufacturers favor fewer, higher-power modules. The argument is essentially: fewer modules, less complexity. This is fundamentally reliant on the wide bandgap (WBG) power semiconductor used: SiC. 

Enphase is taking a different approach, instead using GaN BDS at scale, where each 1.25 MW SST rack combines 342 intelligent 4-kW power modules operating in a coordinated series/parallel delta configuration (Figure 2). This is a direct extension of the distributed architecture Enphase has refined across nine generations of microinverters. Along with the GaN BDS, each power module utilizes Enphase’s 5th-generation Kestrel ASIC, a custom 22-nm control chip first deployed in the IQ9 Microinverters. 

According to Enphase, the choice of GaN BDS over the SiC devices favored by most SST competitors is central to their performance claims: 

The IQ SST power modules are rated at 4 kVA continuous peak power, with GaN BDS switching at an average of over 250 kHz and peaking at 500 kHz. The higher operating frequency will shrink the passives, e.g., common-mode choke, inductors, etc., “When they get smaller and smaller, it really directly drives the cost down,” says Raghu Belur, co-founder and chief product officer (CPO) of Enphase Energy in an exclusive interview with Power Electronics News. 

The magnetics are embedded inside the power module and, according to Belur, “it’s a very sophisticated piece of magnetics, because we have to deal with partial discharge.” One of the more demanding engineering challenges in SST design is isolation, specifically, how to safely bridge the gap between MV and LV domains in a high-frequency transformer without triggering partial discharge, a progressive insulation breakdown that can ultimately lead to failure. 

However, since the 342 power modules are connected in a delta configuration (114 modules per phase), each module on the AC side sees only ~300 V (300 × 114 ≅ 34.5 kV), not the full 34.5 kV of the MV network. On the DC side, each module sees 800 V to support both NVIDIA’s 800 VDC and OCP Mt. Diablo ± 400 VDC standards. This makes both the magnetics and isolation problems significantly less challenging. “We don’t have any creepage and clearance issues outside of the transformer. What we have to ensure is that the transformer is completely partial-discharge-free. And that is the productization work we are doing—making sure things like that are addressed,” says Belur.  

Similar to the IQ9 Microinverters, Enphase’s IQ SST uses a dual-active-bridge (DAB) series-resonant single-stage topology with fully soft-switched commutation. “Since we’re using GaN and doing soft-switching with a resonant topology, it allows us to do ZVS [zero-voltage switching] and ZCS [zero-current switching], and our EMI signature is extremely low. So our whole power module is in a plastic enclosure.” 

A plastic enclosure is a bold move as it requires very quiet power conversion with low power loss since there is no metal enclosure to dissipate heat. But Enphase is well beyond the experimental phase with this design, as it is also used in the IQ9 Microinverters. “We make a couple of million of these a quarter, on a standard pick-and-place machine with an automated line. Very traditional manufacturing, very low cost, nothing exotic,” says Belur. 

Even more strikingly, the entire SST is ambient air-cooled, “Divide and conquer on the thermal. You only have to deal with 4 kW for thermal management. So we put it in a silicone encapsulant, just like we do with all our micros [microinverters]. And our micros are in a much harsher environment—they sit on a roof in Arizona and still have to work well. So that drives the cost down on thermal.”

The 342 modules are divided into three groups of 114 modules, each group connected in series across one phase pair of the three-phase medium-voltage input: line 1 to 2, line 2 to line 3, line 3 to line 1. This delta configuration means each module sees only a fraction of the total medium voltage, keeping per-module voltage stress low.

It also means that there is inherent redundancy built in, “You don’t need all 342 of them to 1.25 MW. I actually need 10% less.” The 342-module count includes a built-in 10% redundancy buffer; in other words, the system can deliver its full 1.25 MW with roughly 309 modules, leaving around 33 as headroom against failure. “And I don’t expect to get anywhere near a 10% failure rate, because today our failure rate is 500 defective parts per million. That’s 0.05% failure rate per year. In 10 years, we expect no more than two of those 342 modules to fail,” says Belur, referring to their microinverter field data. 

Another factor to consider is that, since the system is lower-voltage and lower-power, it is hot-swappable, so failed modules can be replaced. According to Belur, each module is 4 lbs, “So you can just pop one out and pop another one in—exactly like blade servers.” Note that this may not be as simple with larger, monolithic converters that rely on HV SiC. And while using fewer modules may encourage a more efficient solution, it may not encourage modularity and hot-swappability as readily. 

The trend toward solutions with “lower-level modular sub-blocks” seems to also be encouraged by the OCP to achieve high data center reliability. However, the importance of the efficiency metric cannot be overstated—mitigating loss from the grid down to the AI processor is the entire rationale for shifting toward the 800 VDC architecture. The challenge lies in striking a balance between efficiency and reliability, although these aren’t necessarily competing design goals. 

“We’re going to warrant this for 10 years,” says Belur, “this is how we achieve the five 9s reliability [99.999% uptime] we are targeting. The system itself is a very reliable unit; it has phenomenal uptime.

Perhaps counterintuitively, Enphase achieves this level of reliability with an open-loop system. “It’s a single-stage resonant converter with predictive control. That predictive control means we synthesize control vectors in nanoseconds. Predictive control also means it’s not a dual-stage, closed-loop system—it’s actually a completely open-loop system,” says Belur. 

Control vectors are essentially the set of instructions sent to the switching devices, e.g., when to turn on/off, phase angle, and timing. These are done on the order of nanoseconds to track and manage load changes that occur at 4 to 5 Hz without any perceptible lag. And with ~90 million units in the field, Enphase has the data; the predictive control model of the single-stage resonant converter is extremely well characterized.

The topology allows for a sub-millisecond response, according to Belur, “If you look at a training data center—that big dynamic load issue—it swings 10% to 100% at 4 to 5 Hz.” That rapid fluctuation (Figure 3) has to be absorbed somewhere, and feeding it back to the grid is not an option as grid operators won’t permit an interconnection with that level of variation. So data center operators must install local energy storage, i.e., capacitor backup units (CBUs) and battery backup units (BBUs), at the rack level to buffer those swings, giving it several seconds of absorption or hold-up. Absorption refers to the ability to take in or supply a sudden surge of energy, while hold-up is the amount of time a power supply can continue to deliver output power after the input source is interrupted or removed. 

BBUs come with their own design challenges, as they provide extended hold-up when the primary utility power goes down, which calls for batteries rated at high C-rates, from 15 C to 20 C. At these C-rates, the battery degrades rapidly. Plus, it is physically impossible to make a battery large enough to store meaningful energy while fitting it in a sidecar rack. 

“Because of our sub-millisecond response time, we only need anywhere from 1 to 10 ms of hold-up before the remote BESS can respond,” says Belur. In other words, since the IQ SST can react to load swings faster than they can propagate, those transients can instead be reflected back to a BESS located on the medium-voltage campus network, potentially a mile away from the data center building. Belur continues on the topic of the remote BESS, “it can do what it’s meant to do, with a normal C-rate—like 0.5 C or 0.2 C.” 

This means a much smaller CBU and elimination of the BBU: “I can truly eliminate the sidecar and shrink the CBU by 10x, or maybe even 100x. A very small CBU can just be part of the compute rack itself, because I only need one to 10 milliseconds of hold-up,” says Belur. 

Belur also notes a potential compromise some SST companies may be making by relocating batteries from the sidecar rack to the SST itself. “It opens up a whole other can of worms because you have very high circulating currents, you’ll have big EMI issues, and you’ll have to deal with faulted batteries that can source infinite current.” The cleaner solution, he argues, is to remove the BBU from the low-voltage network altogether.

It’s important to note that other SST companies have achieved sub-microsecond response times by integrating a BESS into their SST. However, this can be seen as analogous to relocating the BBU function rather than eliminating it. 

As a pioneer of implementing GaN BDS technology, Enphase has built a safety net by adopting a multi-vendor approach. “We put them through our long-term reliability testing [LTR] and our reliability demonstration test [RDT], which is a 110-day test under pretty severe, harsh conditions. Every vendor we pick has to go through that test cycle,” says Belur. At this point, the vendors are placed on the company’s approved vendor list (AVL). Belur continues: “All the good GaN players are on the list.” It’s already known that Infineon is in their AVL, as it’s officially used in the IQ9 Microinverters. While the voltage rating of the devices used within the IQ SST is unclear, 650-V GaN BDS are already commercially available, with a move to 900 V. There will almost inevitably be more GaN BDS announcements this year. Belur echoes this sentiment, “We’ll get 900 V shortly. That development is coming along very predictably.”

Enphase’s commitment to GaN BDS predates the SST market entirely. Belur noted the company began pushing GaN BDS technology 6 to 7 years ago, working directly with manufacturers to develop devices that met their specific requirements — and only released it commercially with the IQ9 Microinverter once they were confident it performed as needed. “We really curated the GaN [BDS] industry,” he said. “We pushed them towards exactly what we wanted because it was a near-perfect fit for our application.”

“We have made it so that we can do pretty much any voltage in, any voltage out, any frequency— AC, DC, anything. We have made this really powerful, general-purpose platform, but that general-purpose platform has an incredible amount of IP in it.” Belur continues, “That’s the beauty of our architecture—there really isn’t an upper limit to what power we can get to with that specific topology. It’s completely scalable. So you’re not re-architecting, re-engineering, redesigning a whole new product every time.” 

Enphase’s years of GaN BDS development and field deployment with the IQ9 Microinverter now form the technical foundation of the IQ SST, giving the company a manufacturing and reliability track record with the technology that newer entrants in the SST space are likely still working to establish.