brain/
← all mechanisms
medium convictionactive · updated 2026-07-09T00:00:00.000Z

AI rack density → 800VDC architecture mandate → sequential white-space TAM ($11B–$13B) → ETN/VRT as Phase 4 incumbents

As GPU cluster rack density approaches 600kW, resistive losses at 54V AC are physically unmanageable (cuts current 15× at 800V); 800VDC becomes mandatory architecture across a 4-phase transition (2026–2030+), creating a sequential $11B power-rack TAM (peak 2028) and $13B SST TAM (peak 2030) — while total electrical content per MW stays constant, content shifts from grey to white space, favoring Phase 1–3 power-rack suppliers now and Phase 4 SST incumbents (ABB, Eaton, Schneider, VRT) later.

The chain
1
AI GPU clusters approaching 600kW per rack create an unmanageable resistive-loss physics problem at 54V AC (current = 11,111A vs. 750A at 800V; losses ∝ I²R, fall ~220×) — 800VDC architecture is therefore mandated by physics, not preference.
From 2026-06-04-inside-the-800vdc-revolution-part-1: "At 600kW rack power: current drops from ~11,111A at 54V to 750A at 800V (15× reduction). Resistive losses (∝ I²R) fall by ~220×."
From 2026-06-04-inside-the-800vdc-revolution-part-1: "Moving to 800VDC eliminates conversion stages and cuts facility power consumption ~5% at 1GW IT load (saves 50MW continuously)."
2
The transition unfolds in 4 phases across 2026–2030+: Phase 1 (2026/27) = white-space sidecar power racks retrofit existing AC DCs; Phase 2 (2027/28) = 800VDC-native compute blades (Kyber/Rubin); Phase 3 (late 2028/29) = centralized DC-bus distribution; Phase 4 (>2029) = Solid State Transformers (SST) replace conventional LV transformers.
From 2026-06-04-inside-the-800vdc-revolution-part-1: "Phase 1 (2026/2027) — White Space Retrofit: Google/Meta add HVDC power racks to existing AC infrastructure. Power rack converts 415V AC to 800VDC at row level. Cost: ~$400–500K per unit."
From 2026-06-04-inside-the-800vdc-revolution-part-1: "Phase 2 (2027/2028) — Native 800VDC Compute: Kyber and Vera Rubin racks with 800VDC-native systems." "Phase 3 (late 2028/2029) — Centralized Distribution: Facility-level redesign." "Phase 4 (>2029) — Solid State Transformers: SST replaces conventional LV transformers + rectifiers."
From 2026-06-04-inside-the-800vdc-revolution-part-1: "Full 800VDC code support targets NEC 2029. Pre-2029 deployments require custom approvals." — regulatory gate explains Phase 1–3 before SST commercialization.
3
Sequential equipment TAM: power-rack (sidecar) segment peaks ~$11B in 2028; SST peaks ~$13B in 2030. Total electrical content per MW holds at $3.6–4.8M band — content shifts from grey space (traditional PDUs, switchgear) to white space, benefiting power-rack and SST suppliers. Phase 4 SST vendors include startup Heron Power/DG Matrix/Amperesand alongside incumbents ABB, Eaton, Schneider Electric, VRT.
From 2026-06-04-inside-the-800vdc-revolution-part-1: TAM table: Sidecar/Power Rack ~$11B peak 2028; SST ~$13B peak 2030.
From 2026-06-04-inside-the-800vdc-revolution-part-1: "Total electrical content per MW: $3.6–4.8M band — constant across phases, content shifts from grey to white space."
From 2026-06-04-inside-the-800vdc-revolution-part-1: "Incumbents entering [SST]: ABB, Eaton, Schneider Electric." "Phase 4 efficiency: 87.4% vs 82% baseline → ~69MW savings at 1GW scale."
What would falsify this
  • Step 1: An alternative cooling/distribution architecture resolves the resistive-loss problem at 54V (e.g., liquid cooling at the bus level) — would reduce the 800VDC imperative.
  • Step 2: NEC code cycle delays SST certification past 2029 — shifts Phase 4 TAM timeline right, reducing investor visibility.
  • Step 3: Power-rack TAM claimed by a single dominant incumbent (ABB / Eaton) who wins the entire white-space retrofit — reduces breadth of tradeable names.
Contradictions / tensions
  • Phase 4 (SST) timing is speculative: UL cert target was Q2 2026 for DG Matrix but no vendor had completed it as of May 2026. Phase 4 TAM is a 4-year+ out bet.
  • Grey-to-white-space content shift may favor vendors not yet publicly-traded or visible. Paywalled "winners/losers" section likely names specific beneficiaries not available in free excerpt.
  • NEC 2029 code adoption is a regulatory gate — could slip to 2031 cycle, delaying Phase 4 SST TAM.
Implications
  • **Phase 1–2 (near-term, 2026–2028):** Power-rack sidecar is the active TAM segment. Google and Meta are already deploying (OCP Mt. Diablo spec, Oct 2024/May 2025). Suppliers of 800V power conversion at the rack level benefit first. Content is ~$400–500K/unit for white-space power racks.
  • **Phase 3–4 (medium-term, late 2028–2030+):** SST is the largest TAM segment but requires NEC 2029 code adoption and UL certification (no vendor has completed UL cert as of May 2026). VRT, ETN (Eaton), ABB, Schneider as incumbent entrants.
  • **Total content per MW stays $3.6–4.8M** — the 800VDC transition doesn't shrink the electrical content market, it reshuffles it from grey (OEMS of traditional switchgear/PDUs) to white (power-rack + SST vendors).
  • **Partial source caveat:** Source is paywalled; winner/loser section at "The Winners and Losers of the 800VDC Revolution" not extracted. Tradeable names in the Phase 1–3 window need further research (specifically: who makes the $400–500K power rack units at scale?).
  • **ETN (Eaton) is specifically named** as an incumbent entering SST (Phase 4) — extending ferc-large-load-to-dc-gridscale-construction's ETN identification for a different product category.
Companies
none
Concepts
Data Center Physical Construction → Electrical/HVAC Infrastructure Picks-and-ShovelsS-Curve Evaluation Lens
Open questions
Data Center Physical Construction → Electrical/HVAC Infrastructure Picks-and-Shovels