Infineon to supply silicon carbide devices for Eaton's 800VDC transformer platform
The two firms say silicon carbide will cut conversion losses in Eaton's MVSST 2.0 platform; no customer, site or delivery date is attached.
Infineon Technologies will supply silicon carbide power devices for Eaton's MVSST 2.0 platform, a medium-voltage solid-state transformer that the power management firm is aiming at 800VDC distribution inside data centers. The two companies frame the collaboration around the efficiency and reliability of the conversion steps that sit between the electrical grid and an AI rack, and it has the shape these pairings usually take: a technical role defined for each side, and no volume, contract value or deployment date attached.
The chain they are working on is longer than the equipment list suggests, because a conventional data center draws alternating current from the utility, runs it through a series of conversion stages and delivers direct current at the server, shedding energy at every step. Eaton says the MVSST 2.0 platform reduces the number of conversion stages against conventional architectures, which it credits with better efficiency, higher power density and more deployment flexibility, and that the Infineon devices improve the transformer further; silicon carbide, a synthetically produced crystalline compound of silicon and carbon, conducts heat better, switches faster and dissipates less than conventional silicon, the combination that makes it useful in high-voltage power devices.
Rack density is the pressure behind the redesign: at 1MW, the coverage says, the shift from AC to DC distribution becomes necessary, and 800VDC has emerged as the class that can carry power to high-density AI racks. The driver is named without much hedging, as Data Center Dynamics attributes the 800VDC roadmap to Nvidia's direction; solid-state transformers enter the picture because they tie together disparate power sources and loads while making the conversion cheaper to run, and Eaton's pitch for the platform is fewer conversion stages, more density, faster deployment.
"As energy demand from AI data centers and industrial applications continues to grow, the demand for efficient, reliable, and sustainable power conversion is increasing," said Andreas Weisl, Infineon's EVP and chief sales officer for industrial and infrastructure; Eaton's David Zheng, VP of R&D for power quality in APAC, said the goal is to answer rising requirements for power efficiency, system reliability and scalability in AI data centers.
A month of upgrades on the same side of the meter
Eaton and Infineon join a September run of announcements about the hardware between the utility service and the rack, where Schneider Electric said it would factory-build 2.5MW power modules, a Siemens Grid Software webinar put AI racks at 230 kilowatts and recommended scenario-based planning, and now a chipmaker and a transformer maker are tuning the medium-voltage step. The direction is consistent: the compressible part of the power train is inside the fence, while the constraint that sets the pace of the buildout sits outside it.
The grid does not compress on a semiconductor's schedule, because interconnection queues, load-class rules and tariff design decide how many megawatts reach a campus, and a transformer that wastes fewer of them changes the arithmetic on the operator's side of the meter without changing the operator's place in line. If the efficiency claims hold, the gain lands in two places on a site's bill, the power it buys and the cooling load that electrical losses would otherwise have created, but neither company put a figure on the gain, and the announcement names no customer, site or shipping date.
Two markets are being described at once, with the component side making silicon carbide content a line item as conversion moves to higher voltages and the semiconductor vendor selling into a platform rather than a single product, while the project side still turns a data center's value on whether the utility will serve it, on what terms and by when. Grid access has become the asset, and a better transformer does not manufacture a queue position, though it does suggest that an interconnection of a given size goes further, since power that is not lost in conversion is power that reaches the rack, a modest multiplier that nobody has yet put a number on.
Where the collaboration could travel further is the generation side, because solid-state transformers are described as a way to integrate diverse power sources and loads, and a medium-voltage device built to accept several inputs is a plausible candidate for sites that add behind-the-meter generation that developers are now building. That is inference, not disclosure; the agreement commits neither company to a project, a capacity figure or a customer.
The power train has become a product category with semiconductor vendors bidding for a place inside the transformer rather than only at the server, and the next useful data point is a named campus running 800VDC distribution off a medium-voltage solid-state transformer, or a second supplier signed to the same platform. Either would say more about how fast this moves than a statement of intent does.
Save this analysis and keep the funds you follow together in My Desk.
Sign in to save articles or follow funds.