Episode Summary
Executive Summary: Harun Inam, CEO of DG Matrix, explains why solid-state, multiport transformers are becoming critical for AI data centers and electrification. He argues that legacy AC-to-AC transformer infrastructure cannot scale fast enough for multi-gigawatt, behind-the-meter loads, while DG Matrix’s software-defined power architecture can consolidate multiple components, improve efficiency, and enable faster deployment of DC-native power systems.
Main Topics: Harun Inam’s background and DG Matrix origin (Priority: 5/5): Inam traces his career in power electronics from Duke to work on aerospace, solar, computer room power, transmission control, and SmartWires, then explains how those experiences led to DG Matrix with co-founder Dr. Bhattacharya. Why the grid and transformer model is bottlenecked (Priority: 5/5): The conversation frames the traditional centralized, AC-based power model as increasingly inadequate for modern AI data centers, where load growth, interconnection delays, and transformer lead times constrain deployment. What a solid-state, multiport transformer does (Priority: 5/5): Inam defines solid-state transformation as chopping power into high frequency to shrink magnetic components, and multiport as a single device that can coordinate multiple AC and DC inputs/outputs to replace a large legacy power train. AI data centers and 800V DC architecture (Priority: 5/5): NVIDIA’s 800V DC rack vision is presented as a validation of DG Matrix’s approach, since high-density AI racks require far more efficient power delivery than conventional low-voltage AC systems can support. Product strategy and market evolution (Priority: 4/5): DG Matrix started in fleet and building electrification, then expanded into AI data centers after early pilots proved the technology’s value. The company now has multiple product lines for low-voltage and medium-voltage applications. Manufacturing, supply chain, and reliability (Priority: 4/5): Inam emphasizes that the company is designed to scale manufacturing quickly, avoid rare-earth dependence, and meet stringent reliability requirements through heavy testing, telemetry, and root-cause analysis. Business model, partnerships, and long-term implications (Priority: 4/5): DG Matrix sells through neo-clouds, hyperscalers, EPCs, and integrators, while building alliances with battery, generator, and turbine partners. Inam argues the model could eventually enable distributed microgrids for broader electrification.
Key Arguments: Legacy AC-to-AC transformers are not the right architecture for modern high-density loads; the value comes when solid-state power conversion is integrated with upstream and downstream system functions. AI data centers need behind-the-meter power because transmission and distribution systems cannot deliver multi-gigawatt loads fast enough in many locations. Natural gas, storage, and other on-site resources are currently the most practical way to supply large AI loads in the United States. Solar and storage are important, but solar’s land density limits make it unlikely to serve large nearby AI campuses at gigawatt scale in most suburban areas. Multiport power electronics can dynamically allocate power among grid, batteries, supercapacitors, fuel cells, and generators, reducing stranded power and increasing usable compute. DG Matrix’s approach improves reliability by reducing component count and integrating functions that would otherwise be spread across multiple vendors and firmware stacks. The biggest near-term value in data centers is not just efficiency but eliminating stranded power, which directly increases revenue-generating compute capacity. A software-defined power platform can be sold as hardware plus recurring software/licensing, especially for energy management and firmware-driven use cases across markets. The company is intentionally avoiding commoditized distribution-pole transformer markets and focusing on higher-value behind-the-meter applications where its architecture fits best. DG Matrix believes it can hyperscale manufacturing with modular work cells, U.S.-based final assembly, and diversified sourcing without rare-earth dependence.
Data Points: Conventional transformer lead times: 2 to 4 years - Set up by the host as one reason AI data centers are bottlenecked. DG Matrix Series A: $60 million - Recently closed, led by Engine Ventures. Years of industry experience: about 40 years - Inam describes his personal career in power electronics. Relationship with co-founder Dr. Bhattacharya: 15 years - They first met in 2011 and have worked together since. Engineering effort to develop product: 700,000 engineering hours - Inam says this was needed to make the multiport architecture work. Product footprint at 1 MW: 4 feet by 4 feet - Size of DG Matrix product versus legacy skid-based equipment. Legacy footprint at 1 MW: 30 feet by 20 feet at best - Comparison to two large transformers and multiple skids. Efficiency/feature comparison: 10x - Inam claims DG Matrix offers roughly 10 times the feature set of legacy systems. Low-voltage definition: Under 1,000 volts AC - Utility-language definition used in the discussion. Typical industrial low voltage: 480 V AC / 600 V AC - Examples of low-voltage classes referenced. DC output range: 200 V to 920,000 V - Inam states DG Matrix DC ports cover this range, with extension to 1,500 V soon. NVIDIA rack power: 600 kW to 1 MW per rack - Used to illustrate AI load density compared with traditional racks. Typical older rack power: 6 kW to 20 kW per rack - Historical server rack consumption mentioned for contrast. Average U.S. household consumption: 1,500 watts - Used to help contextualize the power density of AI racks. Solar land requirement: 100,000 square feet per MW - Approximate area cited for solar generation. Solar land requirement alternative: 2 to 2.5 acres per MW - Same estimate restated in acres. Gigawatt solar land estimate: 2,000 to 2,500 acres - Estimated land needed for 1 GW of solar generation. Offtake requests in next 3 years: $10 billion to $20 billion - Demand DG Matrix says it is seeing for its low-voltage architecture. Revenue per watt per year: $10 to $12 - Inam cites data center economics when discussing stranded power. Stranded power example: 20 MW stranded in a 100 MW data center - Illustrates the value of eliminating unused capacity. Manufacturing work cell output: 400 MW per year - A 5,000 square foot work cell in double shift with 20 people. Work cell size: 5,000 square feet - Standard modular manufacturing cell described by Inam. Factory scaling example: 4 GW in about 3 months - Potential output if a 100,000 square foot facility is fully deployed. Current engineering team: 400+ engineers - Used to support product, reliability, and scale-up efforts. PhD count: 24 - Number of PhDs on the SST team. Product shipping status: this year - Interport 360 and all low-voltage SSTs are shipping in the current year.
Pivotal Quotes: "the fastest way to add a gigawatt of power for your AI data center is through distributed microgrids or cellular power as we call it" — Harun Inam: His closing vision for how AI power buildouts should scale. "you got 100 megawatts coming in, but only 60 megawatts is getting to a load, or you have a failure in one of the branches. Now you've got another 10, 20 megawatts stranded" — Harun Inam: Explaining the economic penalty of stranded power in data centers. "we're going to go from AC to DC in one go and deliver 800 volts straight to the rack" — Harun Inam: Describing the design logic behind the NVIDIA-aligned architecture.
Implications: DG Matrix is positioning itself as core infrastructure for the AI buildout: faster-to-deploy, more efficient, and more modular than legacy power trains. If its architecture scales, it could reshape data center power and broader electrification around distributed, software-defined microgrids.