SORIVO Engineering · Codes & Compliance · AI Factories

800 VDC Data Center Cable: What's Settled, What Isn't (Yet)

The voltage move to megawatt racks is real and it is funded. The specification layer underneath it has started to arrive — a certification standard for the cable itself was published in July 2026 — but connector qualification, DC arc-flash assessment and conductor identification are still open. Here is where each piece actually stands, and what to write in an RFQ while the rest catches up.

Published 31 Aug 2026 · Updated 3 Sep 2026 · Reviewed by Luo Qiang

Sheet05
BasisIEC 61439 · IEEE 1584
Issued / Rev.2026-09-03 · B
Checked ByL. QIANG
Direct Answer

800 VDC is a low-voltage DC power architecture for AI data centers that replaces 54 V rack distribution so one rack can reach 1 MW. It is not HVDC: 800 V DC sits inside the IEC low-voltage band, which tops out at 1500 V DC. No ratified 800 VDC data center cable standard exists yet from IEC, ISO, UL, NFPA or ANSI.

800 VDC is a power-delivery architecture for AI factories, pushed hardest by NVIDIA with Google, Microsoft and OCP named alongside it, that replaces rack-level 54 V distribution so a rack can climb toward 1 MW. It is low-voltage DC, not HVDC — inside the IEC band that tops out around 1500 V DC. What is settled: the architecture, the partner ecosystem, the fact that switchgear sold to the IEC 61439 family already covers DC to 1500 V, and — as of July 2026 — a certification standard for the cable itself, published jointly by China Quality Certification Centre, TÜV Rheinland and Shanghai Guolan Testing. What is not settled: no IEC, ISO, UL, NFPA or ANSI document names 800 VDC for data centres, IEEE 1584 arc-flash calculation is scoped to AC, and DC fault current has no natural zero crossing for a breaker to interrupt into. Buyers should name the standard they are testing to and demand DC test evidence; "800 VDC compliant" on its own still means nothing, because there is no 800 VDC system standard to be compliant to.

Most coverage of 800 VDC is written from the silicon side — who wins the power-switch sockets. Almost none of it is written from the copper side, which is where our readers live. That gap is the whole reason for this article, and it is now narrower than it was: the cable half of the specification layer started to close in July 2026.

Scope of this guideWe are a cable manufacturer, not a standards body. NVIDIA's whitepapers sit on a gated CDN and could not be opened for this article — what we could read were its public blog posts and product pages, an independent analyst recap, a Schneider Electric safety study and its accompanying white paper, and the scope titles published by IEEE, IEC and ISO. Where a figure is NVIDIA's, we say so. Where we came up empty, the gap stays empty.

00Before Anything Else: There Is No 800 VDC System Standard

NVIDIA's product page says it, with Google, Microsoft and OCP, "sets the 800 VDC standard." Read plainly, that sentence has caused more confusion in the last twelve months than the architecture itself. The word standard is doing two different jobs, and buyers keep getting charged for the wrong one.

Here is what we could and could not establish. As of early September 2026, we found no IEC, ISO, UL, NFPA or ANSI document that names 800 VDC for data centres. Dell'Oro's recap of the OCP Global Summit — an independent analyst firm, not a vendor press release — describes the industry as pushing toward an open standard, with "much remains in flux" and "codes and standards will need to evolve accordingly." That remains the accurate reading of the system side: 800 VDC is an industry alignment, a reference design and a contribution pipeline. It is not a document you can certify a whole installation against.

The product side is a different story, and it changed while we were writing. On 9 July 2026, at the Open Compute Tech Summit in Beijing, China Quality Certification Centre (CQC), TÜV Rheinland and Shanghai Guolan Testing published a High-Voltage DC Cable for Data Centres standard (CQC announcement), developed from a first review meeting held in Wuxi on 23 April 2026 with nine cable and materials manufacturers signed on as drafters. CQC's own wording is that it "precisely fills the gap in standards supporting 800 V power supply," and it focuses on three things a cable buyer actually cares about: electrical safety, mechanical performance, and long-term thermal endurance.

We have not seen the text of the CQC/TÜV standard, and CQC's two public announcements do not give a standard number, an implementation date, or the voltage and conductor-size range it covers. Everything above comes from those announcements. Get the number and the scope from CQC before you cite it in a submittal — and note that it is a certification-body standard, not an IEC or UL document, so it does not automatically satisfy a specification that names one.

Why does this distinction matter to a buyer? Because "compliant to the 800 VDC standard" is a sentence a supplier can write on a quotation and nobody can check. If you are the one signing the purchase order, the useful question is not "is it 800 VDC compliant" but "which published standard does this product test to, and which questions about DC duty are you answering outside any standard."

Timeline of the public record on 800 VDC for AI data centres, marked by whose claim each entry is Eight dated entries from May 2025 to the 2026-2027 roadmap. Most are NVIDIA's own publications. Two, the Dell'Oro OCP Summit recap of October 2025 and the CQC standard of July 2026, are independent. Two other entries are citations made by NVIDIA that could not be opened at source, and are drawn as unverified. FIG. 1 · THE PAPER TRAIL — AND HOW MUCH OF IT IS INDEPENDENT OF ONE VENDOR 20 May 2025 first whitepaper + "AI factories" blog 13 Oct 2025 ecosystem blog "157% more power" vs 415 VAC 20 Oct 2025 Dell'Oro recap: "much remains in flux" 9 Jul 2026 CQC / TÜV / Guolan cable standard 11 Aug 2026 corporate blog (H. Petty) 12 Aug 2026 whitepaper V2.0 "Industry Alignment & Execution" Mar / Jul 2026 OCP joint white paper; SST spec — not opened 2H26 → 2027 MGX power rack; Kyber rows — vendor roadmap NVIDIA's own material independent of NVIDIA cited but not opened / forecast Two entries are now independent of NVIDIA: an analyst recap, and a published cable standard. Nothing on this line is a code or standards deadline. Adoption dates are a vendor roadmap, not a procurement obligation.
Fig. 1 — 800 VDC data center cable: the public record, and how much of it is independent of one vendor. The paper trail is dense and mostly one-sided. The July 2026 entry is the first published standard aimed at the cable itself.

01Why 54 V Runs Out of Rack: 18,500 A vs 1,250 A

Because at megawatt-class racks, the low-voltage answer stops being an engineering trade-off and starts being a physics wall.

Do the arithmetic on one 1 MW rack. At the 54 V busbar that has sat inside server racks for years, 1 MW is roughly 18,500 A. At 800 V it is about 1,250 A. Those are not connector part numbers, they are busbar cross-sections, torque, thermal mass and the weight of the copper you have to hang overhead. NVIDIA frames it the same way in its own words: traditional 54 V in-rack distribution "isn't designed to support the megawatt-scale racks." The company also puts the busbar mass behind a 1 MW rack at around 200 kg, and extrapolates that to 200 tonnes of copper for a gigawatt facility — its own figure, presented as an illustration rather than a measured build.

We ran the copper comparison once already, in our AI data center power cable design guide, which carries the 208 V / 415 V / 480 V / 800 VDC table and the thermal side of it. What that table cannot tell you is which of the circulating efficiency claims are actually independent — so here they are side by side instead.

NVIDIA's own published 800 VDC benefit figures, shown side by side Four cards: up to 5 percent end-to-end efficiency improvement, 45 percent less copper, the same wire gauge carrying 85 percent more power, and 157 percent more power than 415 VAC. All four are NVIDIA's own claims. The last two use different baselines and are bracketed as a pair that should not be added together. FIG. 2 · THE FOUR NUMBERS YOU WILL BE QUOTED — ALL FROM ONE COMPANY, TWO OF THEM MUTUALLY AWKWARD up to 5% end-to-end power delivery efficiency vs 54 VDC NVIDIA · May 2025 45% less copper ("requirements", vs today) NVIDIA · May 2025 85% "the same wire gauge carries 85% more" NVIDIA · May 2025 157% "more power than 415 VAC" NVIDIA · Oct 2025 same kind of claim, two numbers — different baselines Cards 1-3 come from one May 2025 post; card 4 from October 2025. None of the four is independently confirmed. A 45% copper saving and an 85% power-per-gauge gain are different statements about different things. Do not add them. Ask for the whitepaper PDF rather than accepting a blog screenshot — start from NVIDIA's own technical blog.
Fig. 2 — The four 800 VDC benefit figures you will be quoted, all from one company. We're not disputing these. We're labelling them. A buyer who quotes "5% efficiency gain" to a client as an industry fact is exposed the moment the client asks who measured it.

Notice something else in that line-up. NVIDIA's product page compares 800 VDC against a 480 V AC facility baseline; its technical blog compares against 415 V AC. Both are real three-phase facility voltages, and the vendor's own materials disagree about which one is the reference. When you see a spec sheet or a sales deck quoting a percentage saving, the first question worth asking is against what.

02800 VDC vs ±400 VDC: Two DC Routes, Two Cable Answers

This is the part most cable buyers haven't seen, because it doesn't fit the headline. The data-centre DC conversation is not "AC versus DC." It's a fight between two DC topologies, and they don't put the same demands on your conductors.

NVIDIA's is a single unipolar 800 V bus — its own blog describes two-conductor 800 V feeds running down the aisle. The competing approach is ±400 V DC: a bipolar system with a mid-point, so three conductors carry two poles at 400 V to mid. Same delivered power, different cable count, different voltage-to-earth on every insulation surface, different fault loop. The ±400 V work is associated with Open Compute Project documentation under the name "Mt Diablo," and OCP's own announcement describes a published Mt Diablo ±400 V sidecar specification with a later revision to support native 800 V.

Unipolar 800 VDC bus compared with a bipolar plus-or-minus 400 VDC bus Left: a two-conductor 800 V bus with the full voltage across the pair. Right: a three-conductor bipolar bus at plus 400, mid and minus 400 volts, so each pole sits at 400 volts to the midpoint. The difference matters for conductor count and for insulation stress to earth. FIG. 3 · THE TOPOLOGY SPLIT — WHY "DC" ISN'T ONE CABLE QUESTION 800 VDC — UNIPOLAR BUS +800 V return rack load two conductors per circuit full 800 V appears across the pair ±400 VDC — BIPOLAR BUS +400 V mid -400 V load load three conductors; each pole 400 V to mid insulation sees half the pole-to-pole stress The camp split — including where Meta sits — still rests on one source, from a supplier selling the switches either outcome consumes. NVIDIA describes its own feeds as two-conductor 800 V. The conductor-count contrast above is our reading, not a standard.
Fig. 3 — 800 VDC vs ±400 VDC: two DC bus topologies and what each does to the cable. A vendor that hedges across both camps is telling you something about procurement risk: a cable line qualified for one bus isn't automatically qualified for the other.
Table · 800 VDC vs ±400 VDC for data center cabling (Rev. B · 2026-09-03)
Item800 VDC (unipolar)±400 VDC (bipolar)
Conductors per circuit23
Voltage to midpointFull 800 V across the pair400 V per pole
Pole-to-pole fault800 V event800 V event
Pole-to-mid faultNot applicable400 V event
Tray fill and copperLower conductor count per circuitHigher conductor count per circuit
Standard statusCertification standard for the cable exists (CQC/TÜV, Jul 2026); no system standardOCP "Mt Diablo" documentation; no published DC bus standard

Why should a cable buyer care about a topology fight they aren't part of? Three concrete reasons. Conductor count changes the copper per circuit and the tray fill. Voltage-to-earth changes insulation stress, creepage and clearance, and therefore what you test to — the same principle we set out in our comparison of why DC and AC cable are not interchangeable. And a fault between one pole and mid on a ±400 system is a 400 V event, while a pole-to-pole fault is 800 V — the same nominal "800 V class" bus can present two very different fault duties to your protective device.

On who backs what, we'll be careful. NVIDIA names Google, Microsoft and OCP. Trade coverage puts Meta behind the ±400 V camp instead, and describes Google and Microsoft as working both sides — but that attribution traces to a single supplier analysis, from a company selling the switching semiconductors that either outcome would consume. So we won't state it as fact. What we can say is that two architectures are live, and nobody outside the hyperscalers knows which one their next project will be.

03The Standards Map: What Governs an 800 VDC Data Center Cable

This is the section we wrote the article for. Take the things you'd normally pin a data-centre power cable to, and ask what covers them at 800 V DC in a computing facility.

Standards coverage map for an 800 VDC data-centre power circuit Seven rows. Switchgear assemblies and fire performance are covered by existing standards. Cable voltage rating and protective device interruption are partly covered. Arc-flash hazard assessment, connector qualification and conductor identification have no ratified 800 VDC data-centre document that we could find. FIG. 4 · WHAT ACTUALLY GOVERNS EACH LINE OF AN 800 VDC SPECIFICATION WHAT YOU MUST SPECIFY STATUS WHERE IT STANDS Cable voltage rating and insulation PARTLY certification std for the cable; no SDO 800 VDC doc Switchgear and busway assembly YES IEC 61439 reaches 1500 V DC and covers DC duty Protective device interruption PARTLY SSCBs certify to 1500 Vdc; system coordination thin Arc-flash hazard assessment NO IEEE 1584 line-up is scoped to AC; DC methods are AC-derived Connector qualification NO no published 800 VDC data-centre connector standard Conductor identification NO HD 308 excludes DC applications, per Nexans' guidance Fire performance (reaction to fire) YES EN 50575 / IEC 60332 / 61034 apply — orthogonal to voltage ⚑ "NO" = no ratified IEC, ISO, UL, NFPA or ANSI document. A certification standard for the cable now exists — see Table 1.
Fig. 4 — What governs each line of an 800 VDC data center cable specification. Two clean gaps and one that will surprise safety people: the arc-flash math everyone runs is an AC math, applied to a bus with no current zero.

Unpack the two rows that carry the most weight.

Switchgear is further along than people assume. The IEC 61439 family covers low-voltage switchgear and controlgear assemblies, and its scope reaches 1000 V AC or 1500 V DC. The 2020 edition of the general rules adds explicit requirements in respect of DC — precisely because DC gives you continuous arcing during a fault with no natural current zero to interrupt into. So 800 VDC is inside that envelope. Confirm the clause reference against the edition you are actually buying to before it goes on a submittal; the voltage limits in particular are worth reading in the primary text rather than in anyone's summary, including ours.

Arc-flash assessment is the real hole. IEEE's own scope title for 1584.2-2025 puts it in the world of "Three-Phase 50/60 Hz AC 1000 V and Below." Schneider Electric's August 2026 study and its accompanying white paper WP219 say the same thing in plainer language: the assessment methods in use were developed for AC systems, and the analysis is "preliminary, direction-setting."

An 800 VDC bus isn't automatically more dangerous than a 415 V AC one. It's differently dangerous.

That distinction is the one worth carrying into any DC design review: system topology affects arc-flash exposure more than the voltage level itself. The difference is converter-driven, time-dependent, and full of capacitor discharge in the first milliseconds.

Table 1 · Standards register for an 800 VDC data-centre circuit (Rev. B · 2026-09-03)
DocumentWhat it actually coversStatus for 800 VDCBuyer action
IEC 61439 familyLV switchgear and controlgear assemblies; the 2020 general rules add DC requirements; scope quoted to 1000 V AC / 1500 V DCCovers the dutyAsk which part and which edition the assembly is type-tested to, and for the DC clause
IEEE Std 1584 / 1584.2-2025Arc-flash hazard calculation guidesAC-scoped by titleDon't accept an AC arc-flash study as DC evidence; ask what method was used
High-Voltage DC Cable for Data Centres — CQC / TÜV Rheinland / Shanghai Guolan TestingCable electrical safety, mechanical performance and long-term thermal endurance; published at OCTS 2026 after a first review meeting on 23 Apr 2026New in this revisionGet the standard number and scope from CQC before citing it. It is a certification-body standard, not an IEC or UL document
ISO/IEC 22237-1 (2021) and -3 (2021); ISO/IEC TS 22237-5 (2018, under revision)Data centre facilities and infrastructures — general concepts, power distribution, telecom cablingCurrent; predates 800 VDCUseful for facility-level intent, not for DC product qualification
IEC 61140 / 60364 seriesShock-protection thresholds and LV installation rules; the LV band is quoted to 1500 V DC800 VDC is inside LV, not HVConfirm the band figures against the standard you're designing to
EN 50575 / IEC 60332 / IEC 61034Reaction to fire, flame propagation, smoke production of installed cableApplies unchangedFire class is still a separate question from voltage class — specify both
ANSI/UL 489I Ed.1 (2025)Solid-state and solid-state hybrid circuit breakers rated to 1000 Vac / 1500 Vdc; an Outline of Investigation, not yet a consensus standardDevice-level path onlyA device certification path exists, but it does not solve system-level coordination at 800 VDC
An 800 VDC data-centre system, connector or protection standard from IEC, ISO, UL, NFPA or ANSI—None found as of Sep 2026Treat "800 VDC compliant" as an unsupported phrase; ask which real standard is behind it
UL 891 / 1977 / 1699 / 489 / 248 DC ratingsSwitchboards, busways, arc-fault devices, branch protection, fusesNot verified this sessionDC ratings vary by certification scope — check the specific UL file number rather than the standard title

That last row is unusual for an article like this, and deliberately so. We looked at the DC voltage ratings quoted for the North American product standards and found they differ by manufacturer and by certification scope, so we haven't printed numbers for them. If you're selling into a US project that needs them, go get them from UL rather than from a blog — including ours.

04Where 800 VDC Sits: Low Voltage, Not HVDC — and the Rulebook Changes

Terminology drift is doing quiet damage here. Several trade articles call 800 VDC "HVDC," and one even used the phrase "±400 V HVDC." It isn't.

Voltage tiers from 54 V DC to 13.8 kV AC with the standards regime that applies at each A stepped ladder: 54 V DC in-rack busbar in the extra-low band; 415 or 480 V AC facility distribution; 800 V DC, the subject of this article, with no purpose-written SDO standard; 1500 V DC as the low-voltage ceiling and medium-voltage floor; and 13.8 kV AC utility supply governed by the NEC medium-voltage articles. FIG. 5 · THE LADDER, TAGGED BY WHAT GOVERNS — NOT BY MARKETING NAMES 54 V DC in-rack busbar today ELVDC band 415 / 480 V AC facility distribution LV · IEC 60364 / NEC ch.2 800 V DC the AI-factory bus no SDO doc names it 1500 V DC IEC LV ceiling NEC MV floor, too 13.8 kV AC utility MV to the site NEC 245 / 265-268 this article's subject 800 VDC sits inside the IEC low-voltage band — secondary summaries give the DC ceiling as 1500 V. It is not HVDC. ⚑ Band figures come from summaries of IEC 61140 / 60364-1, not the paywalled originals. 13.8 kV is typical, not code-cited. Verify the numbers against the standard your project is actually designed to before they appear on a submittal.
Fig. 5 — Where 800 VDC sits on the voltage ladder, and why it is not HVDC. Different from the code-boundary ruler in our NEC 2026 data-centre piece: that one asked where the law changes. This one asks what you can buy a product tested against.

Back to procurement: because 800 VDC is low voltage, it inherits the LV rulebook rather than the medium-voltage one. No MV switchgear discipline, no MV cable class, no utility-style protection scheme. That's good news for cost and bad news for rigour, because the LV rulebook was written for a world where nobody ran hundreds of amperes of DC through a computer room.

On the US side, the number to hold on to is the one we verified for our NEC 2026 data-center power guide: the code's medium-voltage articles begin above 1000 V AC / 1500 V DC nominal. 800 V DC is comfortably under that line, so it stays in the general provisions. The 1500 VDC BESS cable whitepaper covers the voltage-class logic that carries across, and it's a better template for questioning a DC cable supplier than most data-centre material currently published.

05What Changes on an 800 VDC Data Center Cable Datasheet

Strip out the marketing and an 800 VDC bus asks a cable to answer six questions an AC feeder datasheet never had to.

Comparison of what a 480 volt AC feeder datasheet answers against what an 800 volt DC one must answer Six paired rows. On the left, the familiar AC answers: rated voltage, AC ampacity, short-circuit withstand, IEEE 1584 arc-flash energy, HD 308 core identification and the IEC 60332-3 bunched flame test. On the right, the DC equivalents, several of which have no agreed method or standard behind them yet. FIG. 6 · THE DATASHEET DELTA — SIX LINES THAT STOP BEING ANSWERABLE THE WAY THEY WERE A 480 V AC FEEDER DATASHEET ANSWERS AN 800 V DC ONE HAS TO ANSWER Rated voltage, 0.6/1 kV AC Rated DC voltage + a DC withstand test report Ampacity at 30 °C ambient, AC DC ampacity including conductor self-heating Short-circuit withstand, 1 s DC fault make and break — no current zero Arc-flash energy, IEEE 1584 (AC) DC arc energy: no agreed method yet Core identification, HD 308 DC polarity ID: outside HD 308's scope Bunched flame test, IEC 60332-3 Unchanged — fire ignores your voltage class Row 1 now has somewhere to point. Rows 3 and 4 still do not. That is the honest state of it.
Fig. 6 — The 800 VDC cable datasheet delta: six lines that stop being answerable. Row 6 is the useful reassurance. Reaction to fire doesn't care whether the circuit is AC or DC, so your CPR or UL fire classification work carries straight over.

The interruption problem, in one picture

Rows 3 and 4 deserve the space, because they're where "it's just a higher DC voltage" thinking falls apart.

Why alternating current fault current is easier to interrupt than direct current Left: an AC fault current waveform crossing zero repeatedly, with the arc self-extinguishing at each crossing. Right: a DC fault current that rises and stays, with no zero crossing, so the arc must be stretched and the energy forcibly removed by the device. FIG. 7 · NO CURRENT ZERO — THE ONE PHYSICS FACT BEHIND MOST OF THE STANDARDS GAP AC FAULT CURRENT arc self-extinguishes at every zero DC FAULT CURRENT no zero crossing the arc holds until energy is forced out What it changes: · AC devices lean on the wave · DC devices must create a zero · UL 489I covers SSCBs to 1500 Vdc — devices, not systems ⚑ MCCB-now / SSCB-later is a vendor reading The physics is not in dispute. What is missing is an agreed way to assess the hazard at DC — see Fig. 4, row 4.
Fig. 7 — No current zero: the physics behind the 800 VDC arc-flash standards gap. The cable is the half that now has a standard to be tested against. Protection is the half that doesn't.

Then there's the one nobody raises until the cable is on site: conductor identification. The colour schedule most of Europe sells AC cable against, HD 308, doesn't reach DC at all — Nexans' own guidance lists cables for DC applications as outside its scope. So "brown for positive, grey for negative" tables you'll find circulating online aren't HD 308 answers, and we could not find a single authoritative schedule for a DC bus in a computing facility that we'd be comfortable printing as fact. If a customer asks you for polarity-marked DC conductors, treat the identification method as an open item to close with the designer of record, not a line to fill from a datasheet. Our HD 308, BS 7671 and NEC colour-code guide covers where the AC rules do agree — and where they don't.

Two items in this section rest on a single source each: the HD 308 DC-exclusion reading (Nexans' guidance) and the MCCB-to-SSCB protection roadmap (a vendor analysis of a whitepaper we could not open). Get both confirmed at source before a submittal.
Spec impact — the buyer's angleA voltage change is the easy half of an 800 VDC conversion: bigger clearances, a DC rating, a different test. The hard half is evidence. Ask which question each document in the supplier's file answers, and which one nobody has written a method for yet.

06How to Specify an 800 VDC Data Center Cable Before 2027

Two years of grey zone. Here's how we'd work it, and none of it involves guessing which hyperscaler wins the topology fight.

How to specify an 800 VDC data center cable today

  1. Write standard names, not "800 VDC compliant." There is no 800 VDC system standard to comply with yet. A defensible clause reads: cable and assembly tested to the named standard, at the stated DC voltage, with test reports referencing that standard.
  2. Ask for DC-specific test evidence (not AC evidence relabelled). DC withstand, DC ampacity basis, and — if the supplier claims it — how a DC short-circuit or fault-make figure was established. If the answer is an AC arc-flash study, you've learned something useful about the rest of the file.
  3. Treat protection as the open question. Cable is the easier half. If a design depends on DC breakers at 800 V, ask who modelled the fault, with what method, and whether that method was built for AC. That single question separates a serious package from a repainted one.
  4. Keep fire performance in its own column. Reaction-to-fire classification (EN 50575 in the EU, UL listings in the US) is unaffected by the DC question and still has to be specified for the building. Our building cable fire compliance guide and our EN 50575 CPR importer's checklist both still apply, unchanged, to a cable destined for an 800 VDC aisle.
  5. Price the uncertainty, don't absorb it. If you're quoting a project that names 800 VDC for a 2027 energisation, the qualification work isn't in your cost model yet. Put a clause in — the same discipline we set out in our cable supply contract clauses guide — so a mid-project standard change doesn't land on the supplier.
  6. Watch the standards bodies, not the keynotes. The signal that this becomes a procurement category rather than a reference design is a published document: an OCP specification that reaches draft-ratified status, an IEC or IEEE work item with a number, or a UL category. Until then, treat every "compliant" claim as an architecture statement.
Suggested RFQ wording — replace before use────────────────────────────────────────────── Cable and cable assembly shall be type-tested to [ named standard, edition + date ] at a rated DC voltage of [ ___ ] V. Test reports shall state: 1. DC withstand — value applied, duration, clause referenced, sample count 2. DC ampacity — calculation method, ambient basis, installation method, grouping factor 3. Reaction to fire — [ EN 50575 class / UL listing ] 4. Polarity identification — method to be agreed with the designer of record "800 VDC compliant" is not an acceptable substitute for any of the above. ────────────────────────────────────────────── ⚑ No standard currently specifies items 1-2 for 800 VDC data-centre duty from IEC, ISO, UL, NFPA or ANSI. Whatever the supplier fills in, ask which clause it came from.

Items 1 and 2 are the two worth dwelling on, because there is a right way and a wrong way to ask. There is no agreed DC withstand value for an 800 VDC data-centre cable — that is the point. So don't ask "does it pass a DC withstand test." Ask what voltage, applied for how long, against which clause, and on how many samples. A supplier who can answer all four has done the work; one who answers "yes" has not.

Want the clause, not the advice?We'll send the 800 VDC cable RFQ wording we'd use ourselves — the named-standard line, the DC test evidence to request, and the three questions that expose a repainted AC datasheet. Request the clause text.

07800 VDC Data Center Cable: Settled vs Open Spec Register

One page, nine lines, current as of 3 September 2026. The middle column is the one to quote in a meeting; the right-hand column is the one that keeps you out of trouble.

Table 2 · 800 VDC data center cable standards: settled vs open (Rev. B · 2026-09-03)
ItemStatus as of 3 Sep 2026What to do about it
The architecture itselfSettled — public whitepapers, named partners, hardware shown at OCP Global Summit 2025Assume it's coming for megawatt racks; don't assume it's coming for yours
800 VDC vs ±400 VDCTwo live topologies; the camp split is reported by a single supplier analysisQualify to a voltage class and topology, not to a brand's roadmap
Voltage classification800 VDC is low-voltage DC under IEC bands, not HVDC (band figures via secondary summaries)Correct the "HVDC" wording when you see it in a specification
Switchgear assembliesIEC 61439 covers DC duty to a quoted 1500 V DC ceilingAsk which part and edition; confirm the DC clause yourself
Arc-flash assessmentIEEE 1584 line-up is AC-scoped; DC methods described as AC-derived and preliminaryRequire the method name; refuse an AC study presented as DC evidence
Cable product standardA certification standard for the cable was published in July 2026 by CQC, TÜV Rheinland and Shanghai Guolan Testing. Number and scope not published in the announcementsGet the number and scope from CQC. Note it is a certification-body standard, not an IEC or UL document
Connector product standardNo published 800 VDC data-centre connector standard foundSpecify named connector standards plus DC test reports; drop "800 VDC compliant"
Conductor identificationHD 308 does not cover DC applications, per one cablemaker's guidanceAgree the polarity marking method with the designer, in writing
Fire performanceUnchanged and independent of voltage classSpecify as normal — EN 50575 / UL listing still required

Every standards statement above was checked against at least two sources where two exist, and marked as single-source where they don't. The IEC band figures came from secondary summaries of paywalled standards, NVIDIA's whitepaper-level detail sits behind a gated CDN we could not open, and the CQC/TÜV standard we have not seen the text of. Confirm against the primary documents before any of this enters a contract or a design.

08Five Questions Specifiers Actually Ask Us

Q1

Is there an 800 VDC standard I can specify cable against?

Not one from IEC, ISO, UL, NFPA or ANSI, as far as we could find. What exists is the general low-voltage DC rulebook — IEC 61439 for assemblies, DC-rated protective devices, installation standards that treat 800 V as low voltage — plus NVIDIA's own reference architecture, which it describes as setting a standard but which is an industry alignment rather than a published normative document. Since July 2026 there is also a cable-specific standard published jointly by CQC, TÜV Rheinland and Shanghai Guolan Testing; we have not seen its text and its number has not been published, so get both before citing it. Specify against named existing standards and ask for DC test evidence.

Q2

Is 800 VDC the same as HVDC?

No. Under the IEC voltage bands quoted in the sources we read, 800 V DC is low-voltage DC — above extra-low voltage, well below the medium-voltage threshold, which the NEC places above 1000 V AC / 1500 V DC. Calling it HVDC in a specification will send your supplier looking in the wrong product catalogue. The practical effect is that 800 VDC inherits the LV rulebook, which was never written with hundreds of amperes of DC inside a computer room in mind.

Q3

What's the real difference between 800 VDC and ±400 VDC for cabling?

Conductor count and voltage stress. A unipolar 800 V bus is two conductors with the full 800 V across them. A bipolar ±400 V bus is three conductors, with each pole at 400 V to the midpoint, so a pole-to-mid fault is a 400 V event while a pole-to-pole fault is 800 V. Same nominal class, different insulation stress to earth, different fault duty, different tray fill. Which one wins isn't settled, and a line qualified for one isn't automatically qualified for the other.

Q4

Do I need to change my fire classification work because of 800 VDC?

No. Reaction to fire is a property of the cable in the building, not of the circuit it carries. EN 50575 classes, IEC 60332 flame propagation and IEC 61034 smoke measurement apply the same way whether the bus is 415 V AC or 800 V DC. Keep fire and voltage in two separate columns of your specification — treating them as one question is how cables get approved on the wrong basis.

Q5

Should a normal enterprise data center be planning for 800 VDC now?

Probably not for the next build, and that's worth saying plainly. The public roadmap points at 800 VDC power hardware in the second half of 2026 and full rack or row deployments around 2027, tied to specific NVIDIA generations, and independent analysis expects hybrid AC-and-DC deployments to dominate in the meantime. There's no announced date for when this becomes procurable outside hyperscale-style builds. What a non-hyperscale buyer should take from it is the procurement hygiene: name your standards, demand DC test evidence, and put a standards-change clause in your contracts.

Next step

Specifying a DC bus against a standard that isn't written yet?

Tell us the voltage, the topology and the certification date. SORIVO will quote DC-duty cable and assemblies against named standards, with the DC withstand and ampacity evidence to go behind the claim — and flag the lines where no standard exists yet. Or email sale@sorivocable.com directly.

Request the 800 VDC spec checklist Talk to an application engineer Email Us
Luo Qiang, Senior Cable Application Engineer at Sorivo Cable
Reviewed by Luo Qiang — Senior Cable Application Engineer, Sorivo
15+ years specifying industrial and renewable-energy cable. For this article he checked every standards claim against IEEE, IEC and ISO scope titles, and left the UL DC ratings blank rather than guess them.
Sources
Primary documents we opened
IEC 61439-1:2020 scope and revision notes (IEC Webstore). IEEE Std 1584.2-2025 scope title (IEEE SA). ISO catalogue entries for ISO/IEC 22237-1:2021, -3:2021 and TS 22237-5:2018. IEC TC 99 official scope title. UL 489I Ed.1-2025 scope (ANSI Webstore) and UL Solutions' own note that it is an Outline of Investigation rather than a consensus standard. CQC's two public announcements on the High-Voltage DC Cable for Data Centres standard.
Vendor material
NVIDIA technical blog (20 May 2025), 800 VDC product page, ecosystem blog (13 Oct 2025) and corporate blog (11 Aug 2026) — the origin of every efficiency and copper figure quoted here.
Independent analysis
Dell'Oro Group, "Inside OCP Global Summit 2025" (20 Oct 2025). Schneider Electric, "800 VDC safety in AI data centers" (20 Aug 2026) and white paper WP219, DC Arc Flash Analysis: A Practical Study on 800 VDC AI Data Centers (31 Jul 2026) — self-described as preliminary and direction-setting. Nexans' HD 308 conductor-identification guidance.
Secondary summaries and single sources
Engineering references summarising IEC 61140, IEC 61201 and the IEC 61439 voltage limits; one SiC supplier's architecture comparison. NVIDIA's whitepaper PDFs sit on a gated CDN and were not opened. UL product-standard DC ratings were not verified and vary by certification scope, so no numbers are printed here.
Related reading
AI Data Center Power Cable Design · 2026 AI Data Center Cable Selection Guide · NEC 2026 for Data Center Power · BESS Cable Selection Whitepaper (1500 VDC) · DC Cable vs AC Cable: Why Interchanging Them Is Costly · Conductor Colour Codes: HD 308, BS 7671, NEC · Prefabricated Cable Assemblies vs Field Wiring · Cable Supply Contract Clauses
This article summarizes published information about 800 VDC data-centre power architecture for general guidance. The governing documents are the standards named in your project's specification; verify every clause-level claim against the primary text before contractual or design use.