SolidUltra Raises €3 Million for Solid-State Transformers for AI Data Centers

French startup SolidUltra has secured €3 million to build modular solid-state transformers (SSTs) — power equipment, not AI models — that convert medium-voltage grid electricity into the 800-volt direct current future AI data centers are preparing to distribute. The company raised the money within three months of its June 2026 incorporation. Its founding partner is French power-electronics group Watt & Well.

The bet addresses a bottleneck that now sits at the center of AI expansion. Adding accelerators requires more than securing chips: operators also need equipment that delivers power reliably without consuming excessive space or losing too much energy along the way. Rising compute density is making electrical design an increasingly hard constraint on AI buildouts, which is why power conversion has become a funding category of its own.

What is SolidUltra actually building?

Conventional data center power systems chain a low-frequency transformer together with additional conversion stages before electricity reaches computing hardware. Each stage adds electrical losses, cooling requirements, and physical footprint — trade-offs that become harder to manage as operators plan megawatt-scale racks.

SolidUltra wants to replace that chain with a single integrated grid-to-800VDC conversion system. According to the company's website, the design uses 100-kilowatt cells connected in series on the medium-voltage input and in parallel on the DC output, built on automotive-grade silicon-carbide components. The cells are redundant and field-replaceable, so operators can assemble larger systems from repeated power blocks rather than designing new equipment for every capacity increase.

The funding announcement outlines a scaling range from hundreds of kilowatts to 10 megawatts and beyond. For a growing AI facility, the practical appeal is expanding power-conversion capacity in step with computing requirements. In plain words: SolidUltra proposes one repeatable power block instead of a bespoke electrical redesign at every expansion.

Why does 800VDC matter to AI infrastructure?

SolidUltra is building into a wider industry transition. NVIDIA's 800VDC roadmap covers both existing AC-powered facilities and new DC-native AI factories, with alignment work running through the Open Compute Project to standardize equipment requirements across the ecosystem.

The electrical logic is straightforward. At a given power level, a higher distribution voltage requires less current; lower current reduces conductor demands and resistive losses, making it easier to move substantial power through a constrained physical space. One caveat the source is careful to draw: 800VDC is an intermediate distribution voltage, and computing hardware still needs further conversion before electricity reaches the chips.

There is also a distinction between retrofitting local conversion equipment into an existing facility and designing a new power chain around direct medium-voltage conversion. SolidUltra is pursuing the latter. And its alignment with the 800VDC transition does not, by itself, establish a commercial partnership with NVIDIA — a point worth holding onto when reading any 800VDC startup announcement.

What performance numbers is the company claiming?

SolidUltra's targets are aggressive, and the company itself frames them as development goals rather than demonstrated results. The claimed figures:

• 98.5% conversion efficiency
• Power losses three times lower than conventional architectures
• Equipment footprint reduced by a factor of four
• Approximately 30% lower total cost of ownership
• Up to a tenfold reduction in the carbon footprint of power-conversion infrastructure

Every one of these claims carries conditions. The carbon figure concerns only the conversion infrastructure, not the data center. The cost and loss comparisons depend on which conventional system serves as the baseline. Conversion efficiency is not the same as whole-facility efficiency. None of these are independently demonstrated operating results from commercial installations — they are targets for a planned product.

Can the hardware be maintained and deployed quickly?

Modularity is also the reliability plan. SolidUltra says its field-replaceable power modules are designed to be swapped in under 15 minutes, with an availability target of 99.999%. The objective is granular maintenance: when a single module needs attention, less total capacity goes offline.

The company also aims to shorten equipment deployment from years to weeks through a factory-built design. Faster installation of conversion hardware could help operators bring prepared sites online sooner — but it does not remove the separate requirements for utility capacity, grid connections, permits, or the rest of a facility's electrical system. Faster transformers do not conjure a grid connection; the bottleneck outside the equipment rack remains.

Who is behind the company?

SolidUltra was co-founded by CEO Raul Iglesias, Chief Business Officer Romain Beaume, Alain Rouault, and Watt & Well. The model pairs a dedicated AI-infrastructure company with an established partner's power-electronics engineering and manufacturing capability.

Watt & Well's existing portfolio gives that partnership substance. The group develops modular power equipment for electric-vehicle charging, including AC/DC and DC/DC applications, plus bidirectional conversion products for battery energy storage. It also serves demanding energy and aerospace customers. That experience is relevant to designing and producing power hardware, though SolidUltra's proposed data center system still faces its own development and qualification work.

"AI is not only changing computing. It is fundamentally changing the electrical infrastructure required to power it," Iglesias said in the funding announcement. He described the investment as a way to move toward industrial deployment with Watt & Well's backing.

What happens after 2028?

The €3 million funds engineering, product development, industrialization, and early customer engagement ahead of a targeted commercial launch in 2028. Following that launch, SolidUltra aims to scale production to 10,000 modular systems by 2032 — approximately one gigawatt of installed power. That figure is a production ambition, not a disclosed order backlog.

The company's opportunity rests on a growing mismatch: AI racks are becoming denser while the electrical infrastructure serving them takes considerable time and space to build. A compact, modular conversion system could address part of that mismatch — if SolidUltra can deliver its performance, maintenance, and manufacturing targets at commercial scale. The €3 million buys the engineering work between a proposed architecture and deployable equipment; the verdict arrives in 2028.