Wich inverters are used for the new H200 datacenters? How much percent of total cost?

Inverters in H200 Datacenters:

1. Inverter Type and Vendors:
   The NVIDIA H200 GPU is typically deployed in hyperscale datacenters designed for high-density AI workloads. These facilities generally use three-phase, high-efficiency uninterruptible power supply (UPS) systems with integrated inverters. The most common inverters are:

   • IGBT-based (Insulated Gate Bipolar Transistor) inverters for large-scale, high-reliability AC-DC-AC conversion.
   • Leading vendors: Eaton, Schneider Electric, Vertiv (Emerson), APC by Schneider, and Huawei.
   • Model examples: Eaton 93PM, Schneider Galaxy VX, Vertiv Liebert EXL S1.

2. Technical Specifications:

   • Conversion efficiency: 96–98% (double conversion online UPS)
   • Capacity: 500 kVA to 2+ MVA per module, scalable in parallel
   • Redundancy: N+1 or 2N, depending on Tier III/IV requirements

3. Role in H200 Datacenters:

   • Inverters are not used per rack or per server, but as part of the central UPS system, ensuring clean, uninterrupted AC power to all critical infrastructure, including the H200 GPU clusters.
   • Some facilities may use modular UPS with hot-swappable inverter modules for serviceability and scalability.

---

Cost Breakdown:

• UPS (including inverters) as % of Total Datacenter Build Cost:

  • Industry-standard for hyperscale datacenters: UPS systems (including inverters) account for 5–8% of the total capital expenditure (CAPEX) for the datacenter build.
  • For a facility purpose-built for H200-class density (40–80 kW/rack), UPS/inverter costs are at the higher end of this range due to increased power and redundancy requirements.

• Example Calculation:

  • If total datacenter build cost (excluding land) is $200M, UPS/inverter systems typically cost **$10M–$16M**.
  • This includes all inverter hardware, batteries, switchgear, and installation.

---

Summary Table:

Note:

• Actual inverter models and cost percentages may vary based on datacenter tier, redundancy, and regional procurement.
• For exact models used in a specific H200 deployment, refer to the facility’s electrical design documentation or procurement records.
• The inverter is always part of the larger UPS system; standalone inverter use is rare in hyperscale datacenters.