How AI Data Centers Deliver Power — and Where Taiwan Fits
Why 800 VDC moves power conversion upstream, and where Delta, Lite-On, BizLink, and Chroma fit in the supply chain.
A data center can have enough electricity at the building entrance and still struggle to deliver it to the next generation of AI racks. The constraint is the equipment in between. Power has to pass through switchgear, converters, cables, and connectors before it reaches the accelerators. As more computing is packed into each rack, that equipment has to carry more power through much the same amount of space.
That is the problem behind the push toward 800 volts direct current, or 800 VDC. The shift changes where electricity is converted, how it travels across the data hall, and which parts of the power system suppliers are asked to design and validate. NVIDIA’s overview of 800 VDC power architecture
Why rack density changes the power problem
For a given amount of DC power, raising the voltage reduces the current required. Take an illustrative 100-kilowatt load. Delivering that power at 50 volts takes 2,000 amps. At 800 volts, it takes 125 amps. Actual racks divide power among multiple paths and use several voltage levels, but the arithmetic shows why voltage matters. Higher current requires more conductor capacity, and resistive losses rise with the square of current if resistance stays constant.
At lower rack densities, placing AC-to-DC power supplies close to the servers is a practical choice. At higher densities, those power supplies, busbars, and cables compete with compute hardware for space. Moving some conversion upstream lets designers use higher voltage over a longer part of the power path and reserve the final voltage reductions for the equipment that needs them. NVIDIA’s technical explanation of the 800 VDC architecture
What changes with 800 VDC
In a conventional design, the facility distributes AC power to the racks or nearby equipment. Power supplies convert it to lower-voltage DC, and converters closer to the chips step it down again. A higher-voltage design moves the main AC-to-DC conversion to a dedicated power rack, a row-level power center, or a facility-level power block, depending on the design. The 800 VDC distribution then feeds equipment closer to the compute racks, where downstream converters supply the lower voltages the electronics use. NVIDIA’s hybrid and facility-level designs
| Simplified architecture | Power path |
|---|---|
| Conventional AC distribution | Facility AC → rack-level AC-to-DC conversion → low-voltage DC distribution → chip-level regulation |
| 800 VDC distribution | Facility AC → upstream AC-to-DC conversion → 800 VDC distribution → downstream DC-to-DC conversion → chip-level regulation |
The hybrid design matters because an existing data center can keep its building-level AC infrastructure and add a dedicated power rack for a row of higher-density compute. A new facility has more freedom to distribute DC from farther upstream. These approaches can coexist, so the equipment that wins a place in a retrofit may differ from what a new build needs. NVIDIA’s deployment roadmap
Higher voltage also changes the engineering work. Connectors, insulation, fault protection, and servicing procedures have to suit the new voltage. Backup systems and power controls still have to keep the rack stable when compute demand changes quickly. That helps explain why the supplier opportunity extends from conversion hardware to interconnects and test equipment. LITEON’s engineering overview · Chroma’s high-voltage test applications
Where four Taiwanese suppliers fit
Delta Electronics (2308.TW) helps convert power close to the compute load. Its annual report describes power distribution boards designed to step 800 VDC down to 50 VDC or 12 VDC. Delta also reports shipments of liquid-cooling products across data center applications. The two capabilities illustrate how power conversion and heat removal meet in the same rack design. The disclosures identify the equipment Delta can supply, while the economics depend on what customers buy and at what scale. Delta Electronics 2025 annual report
Lite-On Technology (2301.TW) integrates power at the rack level. Its disclosed 800 VDC power rack combines power shelves with battery backup and liquid cooling. A power shelf houses multiple power modules, and a battery backup unit can bridge a short interruption. Putting those functions in a dedicated rack changes what a customer can buy from the supplier, from individual power supplies to a coordinated power system. Lite-On’s public showcase documents the design and the functions it integrates. LITEON COMPUTEX 2026 product disclosure
BizLink (3665.TW) provides the physical connections that carry power between pieces of equipment. The company has shown busbars, power shelf connectors, and high-voltage DC cables for AI infrastructure. A busbar is a rigid conductor that distributes power where conventional wiring would be cumbersome. Even if higher voltage lowers current for a given load, a supplier’s opportunity depends on the interfaces the customer chooses, the engineering required, and how many connections each deployment uses. BizLink COMPUTEX 2026 interconnect disclosure
Chroma ATE (2360.TW) supplies equipment used to validate power hardware. Its bidirectional test equipment can act as an 800 VDC source to test how a compute rack responds to power input. It can also act as an electronic load to test how a power rack responds when demand changes. That gives Chroma a different source of demand. Its opportunity depends on how much validation and production testing equipment manufacturers need as new designs move toward volume. Chroma’s 800 VDC validation system
From a design opportunity to durable earnings
The four companies illustrate different ways to participate in the same infrastructure shift. Delta and Lite-On can supply conversion hardware and integrated systems. BizLink supplies the connections between equipment. Chroma helps manufacturers verify that the equipment performs as designed. These roles can overlap, and customers can source them separately.
To see who benefits financially, follow the evidence in order. A product demonstration establishes what a supplier can build. Customer qualification or prototype shipments show progress toward deployment. Volume shipments and identifiable revenue help establish commercial scale. Margins, working capital, and cash flow then show how much of that growth the supplier keeps.
That sequence matters because an integrated rack may carry a higher selling price while requiring more inventory, engineering, and field support. A component supplier may sell less per rack while serving more designs. A test equipment maker may see demand arrive in waves as customers qualify new platforms or add production capacity. Each business can benefit from the same transition on a different timetable.
Taiwan Tech Intelligence is preparing a follow-up study on how these four positions translate into operating margins, normalized earnings, cash flow, and valuation. The research is in progress. Explore the upcoming report, Who Captures the Profits From AI Power Infrastructure?
