Load-Following Computing Cables: How Fluctuating Green Power Changes Cable Sizing and Joint Life

01What “compute follows power” means for a cable

Conventional data centers buy flat blocks of power from a utility, and their feeders run near a constant load for years. A green direct-connect campus inverts that: the generation fleet sets the supply ceiling, storage reshapes it, and the scheduler moves compute jobs into the valleys. At the green direct-connect cable chain described in our campus guide, the Zhongjin Data base in Ulanqab pairs 300 MW of wind and solar (200 MW wind, 100 MW PV) with 45 MW of storage behind a 16 km private line, and the city reports the country’s first integrated compute-energy dispatch platform there — one that matches computing load to renewable generation swings autonomously (Inner Mongolia Commerce Department, 2026). The operator’s own materials describe weather-model-driven optimization of generation, storage, load and compute demand in near real time (Energy Storage Industry Network, 2026). Under China’s NDRC Order No. 650, a direct-connect project must self-consume at least 60% of the generation and cover at least 30% of total consumption from that physically traceable supply — ratios that make load-following a compliance mechanism, not a marketing phrase (NDRC, 2025).

Now look at that duty from the cable’s side of the meter. Heat generated in a conductor is I²R: double the current and you quadruple the heating. A load that idles overnight at 20%, ramps with the morning wind ramp, dips when clouds cross the array and peaks when the scheduler packs training jobs into a windy evening produces a heating waveform with sharp square-law edges. Three consequences follow, and they structure the rest of this article:

• The ampacity question changes shape. Steady-state ratings in IEC 60287 assume a 100% load factor. A daily cycle is a different calculation, with its own standard: IEC 60853.
• Insulation ageing follows temperature and time through an Arrhenius relationship — a swinging temperature history is a different ageing dose than a flat one at the same peak.
• Joints and terminations age mechanically as conductors expand and contract through every swing — the failure mode that cyclic testing of connection hardware exists to catch.

02Thermal cycling vs a steady load: what the standards actually assume

IEC 60287 is a 100% load factor statement

The ampacity numbers most engineers quote — the table values, the derating chains, the soil-resistivity corrections covered in our cross-section and ampacity physics guide — come from the IEC 60287 series, which is explicitly a steady-state method for a 100% load factor. It answers one question: what constant current holds the conductor exactly at its permissible temperature, forever. A load-following circuit never sees that duty, in either direction: it exceeds the average for hours and sits below it for hours.

IEC 60853 is the cyclic counterpart

The IEC 60853 series calculates cyclic and emergency current ratings. Part 1 covers cables up to and including 18/30(36) kV where the cable’s internal thermal capacitance can be neglected; Part 2 handles cables above 18/30(36) kV whose internal capacitance matters and also gives the emergency-rating method for all voltages; Part 3 extends the cyclic factor to cases where the surrounding soil partially dries out (Cableizer; GRØFT verification paper). The core output is a cyclic rating factor M: multiply the steady-state rating by M to get the permitted peak of a daily load cycle that reaches the same maximum conductor temperature. Because soil and cable mass store heat, a buried cable responds to the hours-long average of I²R rather than to each peak — so M is typically greater than 1, and the spikier the cycle, the larger the credit.

A practical German-language expression of the same idea appears in Prysmian’s handbook data for buried XLPE circuits, which state that the tabulated current-carrying capacities already assume cyclic operation at a utilization factor of 0.7 (Prysmian PGCEE manual). The buried tables many buyers use “as steady state” are, in that practice, already cyclic tables — worth checking which convention your datasheet uses before doubling margins.

The exception that decides most rooftop and tray runs: cables in air

None of this credit exists for cables installed in air. The conductor temperature of an air-installed cable tracks load changes with a time constant of minutes, not hours, so over a daily cycle it follows the load closely enough that the transient method earns nothing — no M above 1 is available (Cableizer; Jicable paper on emergency ratings in air). That single sentence settles the sizing philosophy for PV string cables on trays, BESS rack links inside containers, and most short feeders in this chain: size them as if the peak were flat, which is exactly what the installation-method comparisons in our tray, conduit, direct-burial and open-air ampacity guide are for.

03What actually ages: insulation chemistry, then joints

Insulation: temperature × time, with a documented ceiling

Polymeric insulation ages through thermo-chemical reactions whose rate rises exponentially with temperature — the Arrhenius relationship that the IEC 60216 series turns into a standardized thermal-endurance test and extrapolation framework (BASEC technical guide). Manufacturers express the result as a design temperature history, not a single number. For XLPE distribution constructions the working ceiling is a continuous 90 °C conductor temperature with a 250 °C short-circuit excursion (Prysmian handbook). For the solar constructions in this chain the published figure is even more specific: EN 50618 qualifies H1Z2Z2-K for a normal continuous conductor temperature of 90 °C, with 120 °C permitted for a cumulative 20,000 hours; cable makers extrapolate the same Arrhenius framework to a 30–40 year service life under normal usage (standard text; Nexans product documentation).

Read that 20,000-hour allowance as what it is: a budget for overload excursions, not a second operating point. A load-following campus that lets 120 °C become routine is drawing down the insulation’s designed excursion budget on a continuous basis — the arithmetic of how fast that budget depletes belongs in the operating plan, and it is one of the few genuinely new engineering disciplines this load model creates.

Joints and terminations: the cycle itself is the enemy

Conductor metal expands through every temperature swing; the connector, the oxide layers, the crimp barrel and the insulation around it do not all expand equally. Over years of daily cycling this thermo-mechanical working fatigues the interfaces, raises contact resistance, and lets the joint run hotter than the conductor it joins — a feedback loop that ends in localized failure long before the cable body is exhausted. The industry already encodes this knowledge in hardware testing: when aluminum branch wiring developed a field-failure record in North America, UL’s response was a critical heating test of 500 four-hour cycles with high overload currents — a 20 A device cycled at 53 A (265% of rating) with its binding screw deliberately tightened to a low torque of 6 lb·in, and solid aluminum 12 AWG conductors tested at 40 A (267%) — and only hardware that survived carried the CO/ALR marking (American Electricians’ Handbook). The test’s premise is the point: termination quality under cyclic overload, including marginal workmanship, is a distinct failure mode that steady-state testing does not capture.

For a load-following campus the procurement translation is concrete:

• Specify compression (crimped) connectors with published cyclic-test pedigree rather than set-screw terminations on any circuit that will swing daily.
• Hold installation torque and crimp-tool calibration to the hardware’s documented values — the UL test was designed with deliberately bad torque to prove why it matters.
• Treat dissimilar-metal interfaces (aluminum collector pigtails, copper buswork) with the compound and hardware systems the manufacturer specifies, because cycling accelerates every interface weakness.

04Where the duty lands differently: array, battery, buried feeder, hall door

The same campus weather produces four quite different cable duties. Table 1 sorts them by the standards lens that applies.

Two asymmetries in that table deserve emphasis. First, the flattest duty sits at the hall door: scheduling compute into generation valleys is precisely what makes IT feeders conventional. The exotic thermal duty concentrates upstream, on the generation and storage side. Second, the buried run is the only place the cycle buys anything back: a swinging feeder can legitimately carry a higher peak than the same circuit under flat load, provided the soil does not dry around it — which is why the moisture assumptions behind direct-buried ratings, covered in our soil thermal resistivity guide, are load-following questions and not just installation questions.

05Sizing margins: how much headroom is enough

Margins multiply silently in this discipline, and each layer looks defensible alone. The discipline is to take each layer from its own source and stop there:

1. The continuous-load rule. North American practice sizes the conductor for 125% of the continuous portion of the load (American Electricians’ Handbook, summarizing NEC sizing rules). This is a load-characterization rule, not a thermal-cycling rule — it exists so protection and conductors are not sized to the exact continuous draw.
2. The cyclic credit — only where the physics allows. For buried circuits, an IEC 60853 cyclic calculation against the actual daily profile can legitimately show a permitted peak above the steady-state rating. For circuits in air, take nothing: the answer to “can I exceed the table because the load swings?” is no.
3. The emergency dimension. IEC 60853-2 also gives the short-time emergency method — how much overload a cable already at steady temperature can carry for a defined duration without exceeding the permissible temperature; the standard’s worked cases cover emergency loads up to 2.5 times rated current, which is a boundary of those cases rather than a ceiling on the physics. In the standard’s own worked example, a cable rated at 1,550 A (I_R), already carrying 1,195 A under normal steady state, can tolerate 2,131 A for a 6-hour emergency (IEC 60853-2 example F6, values per the standard’s worked example). For a load-following campus, that same mathematics is what the dispatcher needs when a wind lull coincides with a training deadline — it is a rating, not slack to be absorbed into normal margins.

The failure pattern to avoid is stacking: a 125% continuous factor applied on top of a cyclic credit on top of a “just this once” emergency assumption. Each layer answers a different question — load characterization, daily cycling, rare contingencies — and none of them is a substitute for another. Where the real margin lives is mostly not in multipliers at all but in installation data: actual soil resistivity, real burial depth, grouping distances — the correction-factor chain in our derating factors guide, and the measured-versus-assumed gap documented in our rated-vs-actual current analysis.

06Spec checklist for load-following circuits

Before a load-following campus cable package goes to tender, the following items convert everything above into line items a supplier can price and a reviewer can verify.

Item 1 is the one most tenders still miss. A cable schedule that says “500 A, buried, 35 kV” cannot be checked against cyclic physics because the shape of the load is missing; attaching the curve costs one page and makes every downstream number auditable.

07The campus that made this the normal case

Nothing in this article is exotic engineering — IEC 60853 has existed for decades, UL has been cyclically testing terminations since the aluminum era, and every ampacity handbook warns against stacking margins. What changed is that load-following stopped being a special case reserved for island grids. A dispatch platform that matches computing load to renewable output is now a reported, operating fact at the first large-scale wind-solar-storage direct-supply computing campus (Inner Mongolia Commerce Department; Energy Storage Industry Network), and Order No. 650 makes that operating mode the compliance path for the model itself. Every campus built on that pattern will specify cables against swinging duty — and the tenders that state the load curve, the rating basis per circuit class and the joint hardware will be the ones that neither overspend on copper nor under-spec the joints that the cycle will quietly test for twenty years.

We supply both ends of that specification: TÜV 2PfG 2693 energy storage cable for the battery links and EN 50618 / UL 4703 constructions for the array side, with the full range on the renewable energy solutions page.

Reviewed by Luo Qiang — Senior Cable Application Engineer, Sorivo
15+ years in industrial and renewable energy cable specification. Experienced in cable specification aligned with IEC standards.