Solid State Transformers

Silicon replaces steel in AI power infrastructure.

Lyrah builds silicon carbide solid state transformers that convert grid medium voltage straight to 800V DC — one stage instead of five. 98.5% efficient, 70% smaller, and on site in 16 weeks instead of two years.

35kV → 800VDirect AC to DC conversion in a single stage
98.5%End-to-end efficiency, versus 93.6% legacy
16 weeksLead time, versus 80–100 weeks for iron-core
Fast protectionFault isolation far quicker than mechanical breakers
The problem

AI is hitting an energy wall.

Rack density jumped from 5–15kW to 120kW–1MW in a single hardware generation, while the equipment that feeds those racks is still built the way it was in 1960. Data centers are being finished and sitting dark, waiting on electricity.

01

Density outran the grid

NVIDIA Blackwell-class racks draw 120kW to 1MW each. The distribution gear in front of them was specified for a tenth of that.

02

Two-year lead times

Conventional transformers ship in 80–100 weeks. AI compute demand grows 10× every two to three years. The math does not work.

03

A $50B+ gap

US data center power infrastructure alone is short tens of billions of dollars of deliverable capacity.

04

Built but powerless

Finished facilities sit idle because interconnection and equipment can't arrive on the same timeline as the servers.

And the architecture itself is broken

Five conversion stages, 5–8% lost as heat Every step from medium-voltage AC down to rack DC burns energy you already paid for.
Mechanical breakers can't protect modern silicon At >10ms trip time, a breaker is far too slow to save a rack of $100K GPUs from a fault.
Electrical gear eats 30–50% of the floor Every square foot of switchgear is a square foot not earning revenue as compute.
The solution

One stage. 35kV AC straight to 800V DC.

Our solid state transformer collapses the entire legacy conversion chain into a single modular, liquid-cooled cabinet — and adds protection, control and grid services that a passive iron-core transformer can never provide.

35kV ACMedium-voltage utility feed
Lyrah SSTOne conversion stage, in place of five
800V DCDirect to the rack busbar
MetricTraditionalLyrah SST
Deployment lead time80–100 weeks16 weeks
Footprint29 m² per MW70% smaller
Efficiency93.6%98.5%+
Fault protectionMechanical, > 10 msFast protection
ScalabilityMonolithicModular, seamless expansion

Plug & play

Factory-integrated cabinet arrives ready to energize. No multi-trade site build-out.

More GPU per acre

Reclaim the 30–50% of floor area currently consumed by electrical rooms.

Lower OPEX

Roughly five points of efficiency straight off the utility bill, every hour, for 25 years.

Fail-safe by design

Modular redundancy and fast isolation keep the load up through faults.

Technology

Silicon replaces steel.

Silicon carbide switching in the tens of kilohertz shrinks the magnetics by 90%. What used to be a 60Hz iron core the size of a shipping container becomes multiple high-frequency transformers inside a cabinet — with three defensible technologies wrapped around it.

Partial discharge free isolation

An isolation barrier with no partial discharge pathway, giving a service life measured against the building rather than the warranty.

25-year lifespanZero partial discharge

Grid-forming control

The unit is an active grid asset, not a passive load. Black start, islanding and voltage support turn your power infrastructure into an ancillary-services revenue line.

Black startIslandingAncillary revenue

Liquid cooling

Liquid cooling pulls heat out five times faster than air, so each SiC die does twice the work it otherwise could.

5× heat transfer2× SiC utilizationAutomotive-grade supply chain

Automotive-grade components

Built on the high-volume EV traction supply chain — proven reliability, real cost curves, and capacity that already exists.

Integrated 5–7MW cabinet

Prototype hardware complete: power modules, protection and controls integrated into a single 35kV-to-800V data center power supply unit.

Certification path

Engineering toward UL 3400 and UL 1741 for commercial deployment in North America.

Customer value

$2.4M per year, per 5MW.

The efficiency gain pays for itself. The space you get back is what actually changes the business case — 70% less footprint is roughly 250 square feet returned to the floor, which is about ten more GPU racks earning revenue.

Annual benefit — 5MW system
$2.4M
  • Reclaimed space → ~10 additional racks$2.2M / yr
  • 4.9 points of efficiency improvement$170K / yr
  • Up-front CAPEX savings42% lower
What changes on site
70%

less electrical footprint per megawatt

  • Time to energized16 weeks
  • Fault isolationFast
  • Expansion modelAdd modules
  • Design life25 years
Market & deployment

800V DC is becoming the standard.

AI power demand roughly doubles by 2030, from about 11GW to 21GW, while the industry converges on 800V DC distribution and efficiency mandates tighten. The solid state transformer market across data centers, EV charging and grid applications reaches $15–25B by the end of the decade.

DRIVER

AI power demand

Nearly doubling by 2030 as training and inference build-outs compound.

DRIVER

800V DC adoption

Rack-level DC distribution is becoming the reference architecture for high-density AI.

DRIVER

Lead-time crisis

80–100 week transformer queues make speed of delivery a purchasing decision by itself.

DRIVER

Efficiency mandates

ESG commitments and utility interconnection limits both reward every point of efficiency.

Market

Hyperscalers

Hyperscale operators and AI cloud providers building high-density compute, where footprint, efficiency and time to energization all decide the economics of a site.

Market

Energy storage

Grid-scale storage, where medium-voltage conversion and grid-forming control let an installation provide ancillary services rather than only shifting energy.

Market sizing and performance figures reflect Lyrah's internal modeling and prototype testing. Production specifications are confirmed per project.

Get in touch

Power your next megawatt in 16 weeks.

We work with data center developers, colocation providers and AI infrastructure teams on 5–7MW deployments. Tell us about your site and load profile.