The Engineering Core

Same Footprint, Double the Kilowatts: DG Matrix Hits 400 kW SST on ST Gen3 SiC


Hook: The usual SST story is “add more cabinets.” On September 24, 2026, DG Matrix said the opposite out loud: keep the module footprint, double the kilowatts—from 200 kW to 400 kW—by swapping in STMicroelectronics’ latest-generation silicon-carbide MOSFETs (Business Wire via MarketMinute, Sept 24, 2026; secondary: Electronics Buzz, Sept 25, 2026).

Key takeaway: This is not another “we invented SST” slide. It is a device-and-package density argument: lower RDS(on) plus a thermally optimized package lets the same Interport power-module envelope carry twice the rated power while the company claims efficiency stays above 98.5%.

What happened

DG Matrix’s Interport is a programmable multi-port solid-state transformer—one conversion fabric meant to route power among grid, on-site generation, storage, and compute loads instead of chaining separate AC/DC boxes. ST has supplied SiC into Interport since the platform’s early development. The Sept 24 announcement is the density step: 200 kW → 400 kW in essentially the same power-module footprint.

  • CTO Dr. Subhashish Bhattacharya credits ST Gen3 SiC MOSFETs: lower on-resistance plus thermally optimized packaging.
  • Press materials put conversion efficiency above 98.5% at the higher power class.
  • ST also provided application support on voltage, current, thermal behavior, and surge-current handling for AI load steps.
  • Roadmap language points to higher-voltage medium-voltage platforms and volume builds for hyperscale, neocloud, and electrification sites.
  • A public Interport Flex Series sheet for a 400 kVA configuration lists a DC port window of roughly 300–920 VDC (800 V nominal) and AC around 400–480 V three-phase—useful context for “native 800 VDC readiness,” not a substitute for site-specific single-line diagrams (DG Matrix Interport Flex 400 kVA sheet).

Engineering mechanism: why the die, not the cabinet, grew

AI halls are pushing operators toward fewer conversion stages and higher DC bus voltage. An SST that can sit closer to an 800 VDC distribution story is attractive because every avoided AC↔DC hop is copper, heat, and floor you can give back to compute. The hard part is not the block diagram. It is how much power you can push through a fixed thermal and isolation envelope.

  1. 1) Conduction loss budget
    At a given current, lower RDS(on) cuts I²R heat in the switches. Gen3 SiC’s pitch is exactly that: more amps before the junction temperature budget is spent—so rated power can rise without enlarging the heatsink stack or the outer module.
  2. 2) Package as a thermal part
    DG Matrix explicitly pairs the die generation with a thermally optimized package. In power modules, the package is not cosmetics; it sets thermal resistance from junction to coolant and how hard you can switch before parasitics and voltage overshoot eat the margin.
  3. 3) Multiport SST is a routing problem
    Interport’s product story is software-defined routing among ports—not only “MV in, 800 VDC out.” Density matters twice: once for conversion watts per cubic meter, and again because each port still needs current loops, protection coordination, and control bandwidth when GPUs slew.
  4. 4) Surge / transient headroom
    ST’s devices are described as strong on surge current. That is the right vocabulary for AI racks: not only steady kilowatts, but pulse loads that punish DC-link and switch SOA if the converter is sized only for average power.

Numbers to keep straight (press + public sheet):

  • Module class: 200 kW → 400 kW, same footprint (company claim)
  • Press efficiency: >98.5%
  • Flex sheet efficiency band: 96–98% (see caveats)
  • DC window on Flex sheet: 300–920 VDC, 800 V nominal
  • Vendor-cited SemiAnalysis framing: ~39 GW of new capacity tied to 800 VDC by 2030; GPU racks “past 600 kW” in the same press narrative

How this differs from last week’s SST headlines

Misconception: “Every SST announcement is the same box with a new logo.”

Reality check: Enphase’s recent IQ SST story is a manufacturing bet—clone many 4 kW bricks. Portfolio plays such as ABB Infinitus sell a source-to-rack chain. DG Matrix’s Sept 24 news is a die-level density bet: same Interport envelope, higher SiC generation, more kilowatts. Confusing those three RFQs gets you the wrong datasheet.

Limits / 待核:

  • Press >98.5% vs Flex sheet 96–98% — treat as different measurement points, load/port configurations, or marketing vs datasheet bands until DG Matrix publishes a single comparable test condition.
  • “Same footprint” refers to the power module in the press release; full skids still include switchgear, cooling, and clearances the sheet itself notes as outside the listed footprint.
  • SemiAnalysis 39 GW / 600 kW+ racks appear as vendor-cited market framing—spot-check the primary SemiAnalysis report before treating them as measured fleet facts.
  • Which exact ST Gen3 part number, voltage class, and parallel count sit in the 400 kW module is not in the materials reviewed here.
  • Multiport “route every electron in one stage” is an architecture claim; protection selectivity and grounding still need a full DC distribution design—SST density does not erase that layer.

Takeaway judgment

If you are scoping 800 VDC AI power in 2026–2027, put DG Matrix’s announcement on a specific checklist—not “another SST vendor,” but:

When the SiC generation jumps, does rated power rise inside a fixed module envelope—or do you still buy more cabinets?

Watch three checkpoints before you treat 400 kW Interport as rack-ready:

  1. Published efficiency and thermal maps at 400 kW under a stated port mix and coolant condition—reconciling the >98.5% press line with the 96–98% sheet band.
  2. Measured behavior under GPU-like current slews and surge events, not only steady-state nameplate.
  3. How the multiport SST interfaces to DC protection and site grounding without reintroducing the conversion stages the architecture claims to remove.

AI power is not waiting for a perfect single SST. It is waiting for conversion hardware whose watts per liter keep up with rack power. Doubling kilowatts without growing the module is one coherent way to get there—if the efficiency and transient data survive contact with real feeders and real GPU loads.

Related: @TheEngineeringCore-v · @Vkinng · vkinngworld.blogspot.com

Sources: DG Matrix / STMicroelectronics Business Wire (2026-09-24); Electronics Buzz (2026-09-25); DG Matrix Interport Flex Series 400 kVA specification sheet (public PDF). SemiAnalysis figures are vendor-cited—verify against the primary report before quoting as settled industry stats.
— The Engineering Core