Two-part tariff for data centres: captive versus third-party under Vietnam's July 2026 pilot

Decree 146/2025/NĐ-CP defines pilot eligibility through three conditions, all of which are structural rather than tariff-class based:

1. Consumption threshold — average monthly consumption of at least 200,000 kWh over the prior 12 months
2. Voltage threshold — grid connection at 22 kV or higher
3. Customer type — originally framed around "manufacturing customers" in the Decree text and EVN's pilot documentation, with hospitality and commercial customers excluded

A captive data centre — one operating as a shared IT facility inside a larger enterprise or industrial group, with no external tenants paying for rack space — sits on the manufacturing tariff under Circular 60/2025's December 2025 reclassification. It meets all three conditions and is clearly in the pilot.

A third-party data centre — one selling colocation, managed hosting, or hyperscale-lease capacity to external customers — was moved onto the commercial (kinh doanh) tariff by Circular 60/2025. The commercial tariff is formally excluded from the pilot. But the data-centre sector's ≥200,000 kWh/month consumption and ≥22 kV voltage characteristics mean these operators sit uncomfortably between "pilot cohort by size" and "excluded by classification." MOIT has not publicly clarified which test controls.

In practice: three different readings are circulating in the Vietnamese C&I market as of April 2026. The first: third-party data centres are out, because commercial-tariff customers are excluded. The second: they are in, because the pilot was designed around the pre-Circular 60 universe where commercial-scale data centres were on the manufacturing tariff, and the scope text has not been amended. The third: scope will be decided case-by-case at the EVN regional corporation level when pilot onboarding begins. Any data-centre operator making commercial decisions against a July 2026 trigger needs a written position from their EVN regional corporation before 1 July 2026, not an assumption.

4.1 · Energy-band compression hits arbitrage revenue

Under Decision 1279/QĐ-BCT, the current peak-to-off-peak spread is VND 2,208/kWh for 22 kV manufacturing and VND 3,416/kWh for 22 kV business. Under the two-part tariff pilot, EVN's published energy rate ranges (off-peak VND 843–904, standard VND 1,253–1,332, peak VND 2,162–2,251) compress this to roughly VND 1,300–1,400/kWh — a reduction of about 35–40% in the peak-to-off-peak spread.

Battery storage sized for energy arbitrage alone — charging at off-peak, discharging at peak — earns roughly one-third less revenue per cycle under the pilot than under Decision 1279. This hits both captive and third-party data centres equally once either enters the pilot. The compensating mechanism is the capacity-charge saving from peak-shaving, which data centres struggle to capture (see §4.2 below).

In practice: Arcus's factory models rely on arbitrage as a secondary revenue stream layered onto capacity-charge savings. For data centres, arbitrage is usually the primary revenue stream, so the compression matters more. A BESS business case at a data centre sized against the current VND 3,416 spread will not survive re-sizing against a VND 1,350 spread without material increase in dispatch depth or cycles per year.

4.2 · Data-centre load profiles do not reward peak-shaving

The two-part tariff rewards customers who can flatten their demand profile. The capacity charge is billed on registered maximum demand (Pmax), so every kilowatt shaved off Pmax saves VND 235,414 per month at 22 kV, in perpetuity. At a factory with a peaky shift pattern, this is the main economic driver of a BESS investment — 15–20% of Pmax is a realistic peak-shaving target.

Data-centre load works differently. IT load is near-flat by design — that is the defining product characteristic of a data centre. The variability that exists comes from two sources that BESS cannot economically reshape:

• Cooling overhead — HVAC and chiller load scales with ambient temperature and adds a predictable afternoon peak of roughly 15–25% above the IT baseload. A 4-hour BESS can flatten part of this but is sized for it, not sized to eliminate it
• Redundancy overbuild — N+1 cooling and N+1 power infrastructure force the operator to declare a Pmax that reflects fully-redundant capacity, not the lower figure that actual load hits. The Pmax declaration is set by what the site could draw, not what it typically does draw

Both mean the capacity-charge saving from BESS at a data centre is structurally smaller than at a factory of comparable consumption. A 5 MW BESS at a 20 MW factory might shave 3 MW off Pmax; the same BESS at a 20 MW data centre might shave 0.5–1 MW if the operator is conservative about redundancy reservations, which they almost always are.

In practice: when pricing BESS for data-centre customers, lead the economics with arbitrage and backup-value displacement. Capacity-charge saving is a real third revenue stream but typically smaller than the factory-case reader would expect. The anchor page note that "BESS economics rely more on outage-backup and site-resilience value than on peak-shaving" for flat-load data centres is directionally correct; this page quantifies why.

4.3 · Captive advantage narrows but does not reverse

Circular 60/2025 moved third-party data centres onto the commercial tariff effective 2 December 2025, raising their blended electricity cost by 45–55%. Captive data centres stayed on the manufacturing tariff and were unaffected. That gap is a direct cost advantage for a single-tenant enterprise data centre versus a colocation competitor, and it has visibly reshaped the Vietnamese colocation pricing conversation since December 2025.

The two-part tariff narrows this gap through two mechanisms. First, if the ambiguity on third-party pilot scope resolves toward "in," both cohorts face the same arbitrage-spread compression, but third-party DCs benefit proportionally more because their starting bill is higher — every 1% reduction in variable-rate exposure saves them more VND than the equivalent reduction saves a captive. Second, the capacity charge falls proportionally more heavily on captive data centres because captive load is flatter and harder to reduce through peak-shaving, removing an offset the third-party cohort can at least partially capture.

A reasonable directional estimate, using Arcus's working assumptions on typical Pmax declarations and utilisation: the 45–55% Circular 60 gap narrows by 5–10 percentage points once the two-part tariff activates and both cohorts are in the pilot. Captive still pays materially less overall — the advantage does not reverse — but the competitive case for the captive architecture becomes less absolute than Circular 60 on its own made it look.

In practice: data-centre operators making 2026–2028 build-versus-lease decisions should not overweight the Circular 60 captive advantage as it stood in December 2025. That advantage narrows when the two-part tariff activates, and will narrow further if the 2028–2030 rollout extends to all commercial customers under Politburo Resolution 70. A decision that only pays back under the full Circular 60 gap is a fragile decision; one that pays back under a narrower gap is robust.