brain/
sourcestock-market

Autoresearch: 800VDC rack chain independent 2026 confirmation

Independent 2026 check of the SemiAnalysis-only 800VDC chain: physics, 4-phase timeline, TAM, sidecar suppliers, SST UL, NEC 2029.

Source

Autoresearch: 800VDC rack chain independent 2026 confirmation

Generated by /autoresearch on 2026-09-19. Synthesized across 3 rounds from 14 successful web pages (plus 3 failed fetches), anchored by no encyclopedia (explicitly excluded for 2026 claims). See Provenance. Treat as raw material — review before promoting. Tight gap: the live dc-rack-density-to-800vdc-white-space-tam chain is still single-sourced to the 4 Jun 2026 paywalled SemiAnalysis-style piece. Context: vault/projects/stock-market

Summary

Independent official sources in 2025–2026 confirm the physics direction and a phased hybrid→central path, but they do not independently replicate SemiAnalysis’s dollar TAM, $400–500k sidecar ASP, or $3.6–4.8M electrical-content/MW band. NVIDIA, Schneider, Vertiv, Delta, and ABB all treat 54 V / low-voltage in-rack distribution as physically impractical at a few hundred kilowatts and treat 800 VDC as the architecture for Kyber-class / ~1 MW racks. NVIDIA’s own 2026 roadmap (MGX-compatible power rack H2 2026, row power center 2027, facility DC power block “decade ahead”) and Vertiv’s sidecar→pod→centralized 2027 / 2028–29 path rhyme with SemiAnalysis’s four phases without using that exact 2026/27–2027/28–2028/29–post-NEC-2029 labeling. Citi independently forecasts 800 VDC taking 16.0% of new global DC capacity in 2027 and 79.1% in 2030, and SST demand rising from 901 MVA (2027) to 37,152 MVA (2030) — adoption and MVA, not $11B / $13B. No independent source prices the sidecar at $400–500k. Commercial white-space sidecars are being productized by Vertiv (PowerDirect 5000, 400–900 kW, H2 2026 commercialization), Delta (800 VDC In-Row 660 kW, shown GTC 2026), and Schneider (800 VDC sidecar up to 1.2 MW, claimed in market before Rubin Ultra 2027); Eaton/ABB are in the NVIDIA ecosystem without a fetched named 2026 sidecar SKU. DG Matrix’s own 17 Jun 2026 MGX press still says “pre-certification units… UL certification targeted for the end of Q2 2026” — as of this 19 Sep 2026 pass, no fetched page announces completed SST UL listing. NEC 2029 remains a real standards-work target (Current/OS + ODCA with NFPA; 2026 NEC is the current edition and a staging year for a larger 2029 restructure), not a hard published “800 VDC illegal until 2029” article. Honesty: stay partial on all three chain steps.

Findings

Physics: 54 V becomes unmanageable; 800 VDC is the industry’s mandated architecture at Kyber / 400 kW–1 MW

NVIDIA’s official product and engineering pages treat legacy 54 V as a bottleneck for next-generation AI factories. “The legacy 54 V standard has become a bottleneck” and “Future generations of AI servers will need more power than today’s 54 VDC architectures can support” (NVIDIA 800 VDC Architecture). The May 2025 NVIDIA technical blog is the load-bearing official physics text: “Traditional 54 V in-rack power distribution, designed for kilowatt (KW)-scale racks, isn’t designed to support the megawatt (MW)-scale racks coming soon”; above 200 kW, 54 VDC “begins to hit physical limits”; using the same 54 VDC approach “would mean power shelves would consume up to 64 U of rack space for Kyber at MW scale, leaving no room for compute”; “The physics of using 54 VDC in a single 1 MW rack requires up to 200 kg of copper busbar”; and “At GTC 2025, NVIDIA exhibited an 800 V sidecar to power 576 of the Rubin Ultra GPUs in a single Kyber rack” (NVIDIA 800 VDC Architecture Will Power the Next Generation of AI Factories). The same page claims 800 VDC vs 415 VAC enables “85% more power… through the same conductor size,” “reducing copper requirements by 45%,” and “Improves end-to-end efficiency by up to 5% compared to current 54 V systems” (same NVIDIA May 2025 blog).

The October 2025 NVIDIA ecosystem blog restates the mandate in current language: “Delivering this level of power at traditional low voltages, like 54 VDC, is physically and economically impractical. The immense current required would lead to high resistive losses and necessitate an unsustainable volume of copper cabling,” with “a megawatt per rack now on the horizon” (Building the 800 VDC Ecosystem). Kyber is designed as an 800 VDC rack: “The NVIDIA MGX architecture will evolve with the upcoming NVIDIA Kyber rack architecture, which is designed to use this new 800 VDC architecture” and a “late-stage, high-ratio 64:1 LLC converter” steps 800 VDC to 12 VDC at the GPU (same).

Independent OEM voices match the direction, not SemiAnalysis’s exact 11,111 A / 750 A / ~220× table. Schneider: “Both of these approaches [400 V 3-phase AC and 48 VDC] become difficult at 200 kW per rack and impossible at 400 kW per rack, which correlate with the NVIDIA Kyber and NVIDIA Rubin Ultra platforms”; “Due to the laws of physics, 800 VDC is necessary for single IT racks 400kW and up to 1 MW!” (Schneider blog, 16 Oct 2025). Schneider’s press release: “The traditional 54 volt in-rack power distribution… is designed for kilowatt-scale racks and can’t support the megawatt-scale racks coming soon” (Schneider press, 13 Oct 2025). Vertiv: “Somewhere between 350 and 400 kW per rack, the traditional approach begins to break down”; operators are preparing for “600 kW racks and megawatt-class scenarios”; “Higher-voltage DC reduces the current required to deliver a given amount of power” (Vertiv, From rack to data hall). ABB: megawatt-range racks are “pushing AC infrastructure to its physical limits”; 800 VDC (or ±400 VDC) is “a more suitable and efficient choice” within the LVDC envelope (ABB, 13 Oct 2025).

What this does not independently confirm: SemiAnalysis’s specific working example (600 kW → ~11,111 A at 54 V vs 750 A at 800 V; I²R fall ~220×) remains SA arithmetic, even though I = P/V is elementary. NVIDIA’s own 1 MW / 54 V copper figure (200 kg) is the closest official analogue (NVIDIA May 2025 blog). SemiAnalysis itself, in the paywalled source this pass is checking, says Phase 1 is “voluntary future-proofing” because Vera Rubin NVL72 “top[s] out at rack densities of 180-220kW” and “Three-phase AC can still deliver that” (SA Part 1, 26 May 2026) — so “physics-mandated in 2026” is overstated relative to both NVIDIA’s 2027 Kyber timing and SA’s own Phase 1 caveat.

Falsifier check (alt architecture solves 54 V): No fetched official source in this pass presents a competing 54 V / liquid-bus architecture that keeps Kyber-class racks on low-voltage distribution. Liquid cooling is treated as parallel thermal work, not a substitute for voltage step-up (Vertiv; Delta GTC 2026).

Timeline: a 3-on-ramp NVIDIA/Vertiv path, not a cited 4-phase 2026–2030 NEC-gated schedule

NVIDIA’s 11 Aug 2026 corporate blog — the most current official roadmap fetched — names three on-ramps, not four numbered phases: (1) “The NVIDIA MGX-compatible 800 VDC power rack, arriving in the second half of 2026… slot into existing AC infrastructure and deliver 800 VDC to compute racks within the row — no changes to the building’s electrical system required”; (2) “the row power center… overhead 800 VDC busway… up to 2 megawatts per row, with availability expected in 2027”; (3) “the DC power block — a facility-scale unit that converts grid power directly to 800 VDC in a single step… the architecture for AI infrastructure being planned for the decade ahead.” NVIDIA, Google, and Microsoft “published a joint white paper March 2026 and the LVDC Solid-State Transformer Specification v0.3 July 2026. More than 80 equipment manufacturers… are already building products to this specification” (Why Scaling AI Compute Performance Requires a New Power Architecture). The May 2025 NVIDIA blog had already said “Full-scale production of 800 VDC data centers will coincide with NVIDIA Kyber rack-scale systems in 2027” (NVIDIA May 2025 blog).

Vertiv’s official path maps even more closely onto SemiAnalysis’s phases, still without the SA labels: sidecar deployments “beginning in 2027”; “Portfolio commercialization begins in the second half of 2026, followed by customer pilots and an expanding deployment ramp through 2027”; “Beyond the pod sits the longer-term horizon: the 2028 to 2029 timeframe and beyond, when 800 VDC-native rack architectures become the norm”; two parallel facility paths — “MV DC UPS: a traditional transformer paired to a rectifier” (higher TRL / first mover) and SST (“Supply chain maturity for solid-state technology still has ground to cover”) (Vertiv). Schneider: “The Rubin Ultra GPUs in a single Kyber rack will begin shipping in 2027. Schneider Electric will have its sidecar in the market well before” (Schneider blog).

Citi, via a 20 Jul 2026 recap of a Super-Sector report, is the only independent sell-side timeline fetched: 800 VDC “16.0 percent in 2027 to 79.1 percent in 2030”; rack density “projected to exceed 1MW by 2028 with the upcoming Rubin Ultra and Kyber platforms, rendering legacy AC distribution unviable” (The Standard / Citi). A Citi Research podcast transcript (PDF, not HTML-fetched; cited only as a search-extracted official-domain document, not used for dollar TAM) describes sidecar retrofits in 2027–2028 and “purpose-built solid-state transformer or SST native data centers from 2029 onward,” with “Sidecars probably could peak in 2027 to 2028; SST may take over from 2029 onward.” That directionally agrees with SA’s sidecar-then-SST sequence and 2029 SST inflection; it is not the four-phase SA calendar and was not treated as a fetched HTML source.

What remains SA-only: the exact “Phase 1 2026/27 / Phase 2 Kyber-Rubin 2027/28 / Phase 3 centralized 2028/29 / Phase 4 SST post-NEC-2029” schema, Google/Meta as the named Phase 1 deployers, and NEC 2029 as the phase-gate for SST commercialization.

OCP’s Diablo 400 / SST spec pages were not successfully fetched (Cloudflare on the OCP blog). Search extraction of the official PDFs (Google/Meta/Microsoft Diablo 400 v0.7, effective 1 Mar 2026; Google/Microsoft/NVIDIA SST spec v0.3, effective 22 Jun 2026) is not treated as a 2026 citation here — those PDFs are off the *.gov whitelist.

TAM: $11B / $13B / $3.6–4.8M/MW remain single-source (and SA internally disagrees on SST dollars)

Not independently confirmed. Every $11B power-rack-peak-2028, $13B SST-peak-2030, and $3.6–4.8M electrical-content/MW figure traces to SemiAnalysis’s Industrials Model (SA Part 1) or to recaps that cite it. SA’s own fetched text says sidecar TAM “peak at ~$11B in 2028” at “$0.5M/MW” and “Total electrical content per MW stays in a $3.6-4.8M band” (same). The same SA page says “By 2030, we expect SST TAM to reach ~$32B” at “$1.25M/MW” (same) — so the vault’s $13B SST figure is not even stable inside the originating piece. DG Matrix’s 17 Jun 2026 press repeats the $13B number while citing that same SA study (DG Matrix MGX) — that is amplification, not a second model.

Citi’s independent numbers are adoption % and SST MVA, not dollar TAM and not content/MW: 79.1% of new global DC capacity on 800 VDC by 2030; SST demand 901 MVA (2027) → 37,152 MVA (2030); BESS+BBU “80 percent-plus” CAGR; displacement of central UPS and LV transformers (The Standard / Citi). Converting Citi MVA into dollars would be invention; this pass does not do it. NVIDIA cites Wood Mackenzie “$9 trillion in global AI and data infrastructure investment through 2040” — campus-scale, not power-rack or SST TAM (NVIDIA Aug 2026 blog).

$400–500k sidecar ASP: SA estimate only — “We estimate the ASP for the Power Rack to reach $400-500k per unit, roughly 10x the ~$40k ASP of standard AC power-rack equipment. On a deployed-MW basis, that lands near $500k/MW” (SA Part 1). No OEM IR page fetched in this pass publishes a list price or ASP in that band.

Falsifier check (one incumbent takes all power-rack TAM): fetched supplier set is multi-vendor (Vertiv named SKU; Delta named SKU; Schneider named sidecar to 1.2 MW; NVIDIA “80+” builders). That does not support a single-incumbent sweep as of Sep 2026. It also does not size shares.

Who commercially supplies white-space sidecar racks

Named, official, 2026-visible products (not an exhaustive BOM; no buy/sell):

  • Vertiv — Vertiv PowerDirect 5000. “an 800 VDC sidecar, delivering from 400 kW up to 900 kW directly into the compute over a busbar tied to the GPU rack.” Commercialization “second half of 2026,” pilots/ramp “through 2027.” Scott Armul (May 2026 Investor Conference): “This is a story of DC power in the rack moving to DC power in the pod.” Most of what Vertiv presented is “tied to specific customer engagements” (Vertiv).
  • Delta — 800 VDC In-Row 660 kW Power Rack. Six 110 kW shelves, 80 kW BBU each (480 kW BBU total), AC/DC PSU efficiency “as high as 98%”; shown at NVIDIA GTC 2026 (March) for MGX. Delta also showed an SST converting MVAC to 800 VDC at “98.5%” (Delta Americas, 28 May 2026; PR Newswire, 16 Mar 2026). Earlier OCP 2025 Delta materials (search-only, not fetched this pass) described a 1.1 MW in-row configuration — not cited as fetched.
  • Schneider Electric — 800 VDC sidecar to 1.2 MW. “collaborating with NVIDIA to develop an 800 VDC sidecar capable of powering racks of up to 1.2 MW”; modular conversion shelves, modular energy storage, Live Swap (Schneider press, 13 Oct 2025). Blog: sidecar “in the market well before” Rubin Ultra 2027; supports “Google, Meta, and others” (Schneider blog). No fetched 2026 shipment or ASP.
  • NVIDIA reference. “NVIDIA MGX-compatible 800 VDC power rack, arriving in the second half of 2026” (NVIDIA Aug 2026 blog); GTC 2025 exhibited sidecar for Kyber 576 (NVIDIA May 2025 blog).
  • Eaton / ABB. NVIDIA lists Eaton, Schneider, Vertiv, ABB, GE Vernova, Heron Power, Hitachi Energy, Mitsubishi Electric, Siemens as data-center power-system collaborators (NVIDIA Oct 2025 blog). ABB’s fetched article is architecture + Infinitus solid-state breaker, not a named sidecar SKU (ABB). Eaton’s 2025 800 VDC reference-architecture and 2026 Beam Rubin DSX releases appeared in search; both official Eaton HTML fetches timed out — do not treat those pages as fetched.

$400–500k “who supplies that ASP”: still unanswered by OEM price cards. The commercial suppliers above are the white-space sidecar set; the dollar figure stays SA.

SST UL certification as of 19 Sep 2026

No fetched announcement that any vendor completed UL listing for a datacenter SST after the Q2 2026 DG Matrix target.

DG Matrix, 17 Jun 2026 (post-GTC, still inside or just after Q2): “Pre-certification units are shipping now, with UL certification targeted for the end of Q2 2026.” The same release cites SemiAnalysis’s four phases and $13B SST figure (DG Matrix MGX). Later DG Matrix index/press pages still carry that target language in search extracts; this pass did not find a “UL listed / UL 1741 certified Interport SST” statement. The Feb 2026 Ampace collaboration is “UL-certified, grid-active BESS integrated with” an SST architecture — BESS listing, not a standalone SST UL grant (search/recap only; not used as a 2026 UL-complete claim).

NVIDIA is “actively studying the CapEx and, OpEx, and safety implications of both traditional transformer-based and solid-state transformer (SST) approaches” (NVIDIA May 2025 blog). Vertiv is developing SST in parallel with a higher-TRL MV DC UPS because SST supply-chain maturity “still has ground to cover” (Vertiv). OCP SST spec v0.3 (July 2026 per NVIDIA) is a design spec, not a UL certificate (NVIDIA Aug 2026 blog).

SA as of May 2026: “no vendor has completed UL certification for datacenter SST deployment as of May 2026” (SA Part 1). This Sep 2026 pass does not upgrade that status.

NEC 2029: still a standards-work gate, not a published 800 VDC prohibition

Partially confirmed, with an important voltage-class caveat.

Facts from NFPA’s own 2026 key-changes blog (29 Jan 2026): the current edition is NEC 2026; 2026 changes “make room for the more significant structural shift coming for the 2029 edition”; new 2026 articles cover systems over 1000 V ac / 1500 V dc; “More modernization is planned for the 2029 edition” (What Changed in the 2026 NEC?). 800 VDC sits under the 1500 V dc line, so it is not waiting on those new over-voltage articles to exist as a voltage class. NFPA’s code-development page (fetched but CSS-heavy) states the correlating committee is completing reorganization in the 2029 edition.

Current/OS and ODCA “are working with the National Fire Protection Association (NFPA) toward updates in the 2029 National Electrical Code revision cycle, which would then be adopted state by state” after publication. Current/OS president Yannick Neyret: electrical codes are “ruled by national rules or national standards … very, very difficult to change.” ODCA’s Stammberger: a formal standard is helpful but not always a legal prerequisite (German “state of the art” path). First DC-native data centers “planned for completion around the end of 2027” (Data Center Knowledge, 31 Mar 2026).

SA’s stronger claim — “Full 800VDC code support targets NEC 2029. Pre-2029 deployments therefore require custom AHJ approvals” and even SA’s own hedge that “NEC 2029 will achieve partial provision, while full code maturity probably lands at NEC 2032 or 2035” (SA Part 1) — is not contradicted by NFPA (2029 is still the next big cycle) and is supported in direction by Current/OS–ODCA–NFPA work. It is not independently confirmed as a single hard “SST cannot ship until NEC 2029” legal gate. NVIDIA/Vertiv/Schneider are commercializing sidecars on existing AC plants in 2026–27, which is the SA Phase 1 workaround, not a wait-for-2029 halt.

Falsifier check (NEC slips past 2029): nothing fetched says the 2029 cycle was cancelled or slipped to 2031/2032 as the edition year. SA itself already allows “full maturity” later than 2029. State-by-state adoption after publication is the slower operational gate (DCK).

Contradictions and open questions

  • Stay partial on all three mechanism steps. Physics and the phased path now have independent official corroboration; dollar TAM, sidecar ASP, and the exact four-phase/NEC-as-SST-gate schema do not.
  • SA $13B vs SA $32B SST TAM in the same originating article. Vault currently carries $13B. Do not “confirm” either number.
  • When is 800 VDC physics-mandatory? Schneider: impossible ≥400 kW. Vertiv: breakdown 350–400 kW. NVIDIA: limits above 200 kW; Kyber MW-scale 2027. SA: 600 kW Kyber Ultra; Rubin NVL72 still AC-feasible at 180–220 kW. The 2026 sidecar wave is hybrid/future-proofing, not a 2026 600 kW forced cutover.
  • Monopolar 800 V vs ±400 V (Diablo 400). NVIDIA Kyber is monopolar 800 V; OCP Diablo 400 (not HTML-fetched) is hyperscaler ±400 V with an 800 V option. Two architectures share a marketing label.
  • SST UL: Q2 2026 DG Matrix target appears missed or unpublished as of 19 Sep 2026. Need a listing number, UL 1741/equivalent grant, or an honest “still pre-cert” IR line.
  • Sidecar ASP / content per MW: still no OEM price card, Eaton/Vertiv/Schneider IR attach, or second research house dollar TAM.
  • OCP official HTML (blog + Diablo/SST spec pages) Cloudflare-failed. Re-fetch via a non-blocked path before citing spec text.
  • Eaton official pages timed out. Beam Rubin DSX / 800 VDC reference architecture remain search-only.
  • Alt 54 V architecture: not found. Single-incumbent TAM sweep: not supported by the named multi-vendor set.

Provenance

Rounds run: 3 of 3 (full)

Sub-questions by round:

Round 1 (broad survey):

  1. Do NVIDIA / OCP / Google / Meta official pages independently state that ~600 kW racks make 54 V unmanageable and 800 VDC physics-mandated?
  2. Who commercially offers white-space 800 VDC sidecar / power racks in 2026?
  3. SST UL status in 2026 (DG Matrix Q2 target and peers)?
  4. Is NEC 2029 still the US code gate for 800 VDC / SST?
  5. Does any independent house publish power-rack / SST dollar TAM or $3.6–4.8M electrical content per MW?

Round 2 (drill-down):

  1. Did DG Matrix (or any SST vendor) complete UL after Q2 2026? — targeting the May 2026 “no vendor UL’d” freeze.
  2. Independent TAM / adoption: Citi and OEM IR vs SA dollars.
  3. Named sidecar SKUs from Delta, Schneider, ABB, Eaton.
  4. Exact NVIDIA current/loss numbers on HTML (not the off-whitelist PDF).

Round 3 (resolve remaining uncertainty):

  1. NEC 2029 vs already-published 2026 NEC — is 800 VDC waiting on 2029 or already in the <1500 V dc book?
  2. Any SST UL listing number as of Sep 2026?
  3. Any non-SA $400–500k sidecar ASP?

Anchor source (Grokipedia, fetched before round 1):

  • Skipped. User instruction: do not use encyclopedia as a source for 2026 claims. Noted as “no Grokipedia anchor (excluded).”

X sources: not requested (--include-x absent).

URLs fetched (14 successful, 3 failed):

Round 1:

Round 2:

Round 3:

Consulted but not fetched (reliability / whitelist): OCP Diablo 400 and SST spec PDFs (off *.gov PDF whitelist); Citi podcast PDF on citigroup.com; NVIDIA whitepaper PDF on dam-cdn.nvd.orangelogic.com.

Tools used: WebSearch, WebFetch. Encyclopedia fetch skipped. No x-fetch. Generated: 2026-09-19 ~21:15 UTC

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