BTM economics — PV + BESS at a 100 MW DC anchor

v3 update (2026-05-08): Fixed a partial-year averaging bug in btm_economics.evaluate that biased every NPV downward by ~52 M RM uniformly (MYT 2025 partial year of 7 hours / 1 MWh was being included in the avg_annual_savings_rm denominator alongside 5 full years). Net effect: PV-only NPV @ 60 MW jumps from +30.5 → +82.7 M; IRR 9.9% → 13.0%; payback 9 → 8 years. PV+BESS uniformly +52 M but still net negative. Tornado recalibrated against new LP base (pv_calibration 0.9322 → 1.1186, bess_md_capture 0.5045 → 0.6611). Aggregator and trigger constants updated (PV per-MW: 432 → 532 kRM/yr; BESS-incremental per-MWh: 81 → 100 kRM/yr). Strategic conclusions unchanged. See § “v3 update — partial-year fix” below.

v2 update (2026-05-07): Added cooling-CDH temperature sensitivity to the DC load (PUE rises 0.008/°C above 25°C). Net effect: +6.6 to +8.1 M NPV across scenarios; PV-only IRR moves 9.5% → 9.9%. Modest model-fidelity improvement; strategic conclusions unchanged. See § “v2 update — cooling CDH” below.

v1 update (2026-05-07): BESS dispatch upgraded from heuristic to LP-optimal (CBC, monthly LP, MD-aware). The structural conclusion holds: BESS at 5M RM/MWh is NPV-negative even with optimal dispatch. See § “v1 update — LP results” below for the head-to-head and the cleaner explanation.

---

Generated: 2026-05-08 (v3 partial-year corrected). Inputs: 5y full-year silver.weather_hourly (2020-2024 calendar years; partial year 2025 dropped), Tariff E3 (HV ETOU industrial proxy), TNB ICPT historical surcharge, placeholder capex curves.

Site assumptions

Field	Value
Announced IT capacity	100 MW
Utilization	80%
PUE	1.4
Gross load (constant, 7×24)	112 MW
Tariff	E3 (HV ETOU industrial proxy)
Energy peak rate	0.337 RM/kWh + ICPT
Energy off-peak rate	0.202 RM/kWh + ICPT
MD charge	35.50 RM/kW/month (peak-window only)
Discount rate	8%
Horizon	20 years
PV capex	3.5M RM/MW (~ USD 700/kW)
BESS capex	5.0M RM/MWh (~ USD 1,000/kWh)
BESS RTE	88%
BESS sizing	4-h duration; power = PV/4 (i.e., 30 MW PV → 7.5 MW × 4 h = 30 MWh)

Sensitivity sweep (mean savings over 2020-2024 weather, RM millions)

ICPT period-aligned (mix of historical -2 to +20 sen)

PV (MW)	BESS (MWh)	CapEx (RMm)	Annual save (RMm)	NPV (RMm)	IRR	Payback
0	0	0	0	0	n/a	—
30	0	105	13.33	+15.6	9.9%	9 yr
30	30	255	16.47	−168	−3.3%	>20
60	0	210	25.87	+23.4	9.5%	9 yr
60	60	510	32.03	−344	−3.7%	>20
90	0	315	38.41	+31.2	9.3%	9 yr
90	90	765	47.56	−522	−3.9%	>20

ICPT held at +16 sen/kWh (current 2024-H2 level)

PV (MW)	BESS (MWh)	CapEx (RMm)	Annual save (RMm)	NPV (RMm)	IRR	Payback
30	0	105	16.35	+45.2	13.4%	7 yr
30	30	255	19.41	−139	−0.9%	>20
60	0	210	31.91	+82.7	13.0%	8 yr
60	60	510	37.92	−287	−1.2%	>20
90	0	315	47.47	+120.1	12.8%	8 yr
90	90	765	56.40	−435	−1.3%	>20

Five read-throughs

1. PV alone clears the IRR hurdle once ICPT > ~10 sen/kWh

Without ICPT or with the historical 2020-2021 rebate baked in, PV NPV is modestly positive at 8% WACC (period-aligned IRR ~10%). Once ICPT settles at the post-2022 industrial level (16+ sen/kWh), PV alone returns 13.0% IRR with an 8-year payback (60 MW base case) — comfortably bankable for project-finance equity returns.

The IRR doesn’t change with PV size (linear scaling — no economies or diseconomies in this model). Real projects would see modest economies via fixed dev costs and diseconomies via interconnection complexity at >50 MW.

2. BESS is severely value-negative under the v0 dispatch heuristic — and the failure mode is structural

Adding 30/60/90 MWh BESS only adds 0.5–1.6 RM-million of annual savings, nowhere near the 75–135 RM-million capex. The IRR drops from +6/9% (PV alone) to −6/−9%. This is not a BESS economics problem — it’s a dispatch problem.

The heuristic discharges BESS during the early ETOU peak hours (14–18 MYT) when PV is also active. By 18:00 MYT BESS is empty, but PV has dropped to zero. The 18:00–22:00 ETOU peak hours then run at full grid import (~112 MW gross), so MD ends up unchanged from the no-BESS case. BESS captures only the energy spread (~1.2× peak/off-peak with RTE 0.88 → ~6.9k RM per cycle × ~365 cycles ≈ 2.5M RM/yr), not the MD avoided cost (which would be worth ~6.4M RM/yr per 15 MW shaved at 35.50 RM/kW/mo).

Fix in v1: smarter dispatch — reserve a fraction of BESS energy for the last-peak hours (18–22 MYT). Could be rule-based (“never let SoC drop below 50% before 18:00”) or LP-optimised. Estimated lift: 4–5 RM-million per year per 15 MW BESS power.

3. The “no surplus to charge from PV” trap is real for under-sized PV

A 100 MW DC has 112 MW gross load. Even 90 MW PV at peak generation produces ~70 MW (after PR), still less than load. PV surplus = 0 always. Hence BESS charge has to come from off-peak grid (which the v0 dispatch now does). For PV ≥ load (i.e., > 130–150 MW for a 112 MW gross site, which doesn’t fit on typical DC rooftop/land), the surplus-to-BESS path opens up — but at that scale we hit grid export PPA constraints and curtailment instead.

Implication for design: sizing PV at 30–60% of DC gross load is the sweet spot. >70% just adds curtailment without changing economics.

4. The avoided-cost margin is large enough that base PV economics don’t need ENEGEM

A 60 MW PV system at 16 sen/kWh ICPT saves 25.9M RM/yr against 210M RM capex — i.e., the BTM avoided-cost case alone delivers 9.5% IRR before any cross-border export, green-attribute resale, or VPP service. Adding ENEGEM as upside (even at conservative 0.5–1 RM/kWh assumed export net margin) only swings the curve up, not the sign.

This validates the cross-comparison report’s recommendation: lead the commercial case with BTM avoided cost; treat ENEGEM and VPP service revenue as upside, not the foundation.

5. ICPT trajectory is the single biggest external risk — and the post-2022 step is sticky

The +16 sen/kWh case is +35.8M NPV at 60 MW; the period-aligned case (which weights heavily by 2020–2021 rebate years) is −25.5M NPV. A 10 sen move in ICPT changes annual savings by ~9.8M RM at 60 MW PV, about a 20% NPV swing.

Government subsidy framework keeps domestic ICPT at zero, but industrial ICPT has been positive every period since 2022-H2. The risk to this case is policy reversion (e.g., subsidy expansion to cover industrial), not fuel prices. Recommend explicit policy-risk hedge in the commercial structure — e.g., a tariff-floor clause in the BTM PPA.

v3 update — partial-year fix (2026-05-08)

What broke

btm_economics.evaluate averaged savings_rm across the per-year annual list. Because the 5-year weather window (UTC 2020-01-01 .. 2024-12-31) leaks ~7 hours into MYT 2025 (UTC+8 offset), the LP wrote 6 annual rows for every scenario: 5 full years + a “year 2025” with 1 MWh PV and 470 RM savings. Averaging over 6 rows instead of 5 dragged every per-year mean down by ~17%.

The bias was uniform across scenarios:

Scenario @ ICPT 16 sen	All-rows mean save	Full-year mean save	NPV bias
60 MW PV alone	26.59	31.91	−52.2
60 MW PV + 60 MWh BESS	31.60	37.92	−52.1
90 MW PV alone	39.56	47.47	−78.0

The bug existed since stage 4 v0; the v1 (LP) and v2 (cooling-CDH) refactors preserved it. Caught when the PPA pricing auto-derive infrastructure (derive_ppa_inputs_from_btm) used full-year-only denominators and surfaced a 17% delta vs the report’s stated tenant ceiling.

The fix

Filter years to those with ≥4000 hours of weather data before LP simulation. Year 2025 (7 hours) is dropped; years 2020–2024 retained.

year_hours = {y: weather.filter(pl.col("year_myt") == y).height for y in years}
years = [y for y in years if year_hours[y] >= MIN_HOURS_PER_FULL_YEAR]

Side-by-side: v2 (biased) vs v3 (corrected)

Scenario @ ICPT 16 sen	v2 NPV	v3 NPV	Δ NPV	v2 IRR	v3 IRR
30 MW PV alone	+18.4	+45.2	+26.8	10.3%	13.4%
60 MW PV alone	+30.5	+82.7	+52.2	9.9%	13.0%
90 MW PV alone	+42.4	+120.1	+77.7	9.7%	12.8%
60 MW PV + 60 MWh BESS	−349	−287	+62	n/a	n/a
90 MW PV + 90 MWh BESS	−527	−435	+92	n/a	n/a

Strategic implications

Quantitative magnitudes are larger; qualitative conclusions unchanged. PV alone is now comfortably above any reasonable hurdle rate (13% IRR vs typical 10% project-finance bar). BESS still NPV-negative by ~290 M at the 60+60 MWh size; trigger curve for BESS has shifted earlier by 1 year for the “Strong VPP stack” and “24/7 CFE premium” scenarios.

The v3 commercial framing strengthens: PV-anchor PPA + carbon-attribute bundle is comfortably investable; BESS remains conditional on VPP service contract.

Capital-stack implications (added 2026-05-08): The PPA-tariff IRR sweep in ppa_pricing.json shows the developer-floor crosses the tenant ceiling between 12% and 15% target IRR. At 15% IRR the floor (454 RM/MWh) exceeds the LP-derived tenant avoided cost (406 RM/MWh) — vanilla kWh-only PPAs become structurally infeasible. Bundled-carbon PPAs (ceiling 696 RM/MWh under hyperscaler internal carbon at USD 100/tCO2) accommodate up to 18% IRR comfortably. At 15% IRR the developer floor exceeds the LP-derived tenant ceiling — vanilla kWh-only PPAs become structurally infeasible; equity-heavy structures must bundle carbon to close. Debt-heavy structures at 8% effective discount have full headroom on either path.

Known modeling simplifications (2026-05-08 audit)

The dispatch_lp module was audited end-to-end and is free of boundary or partial-year bugs (UTC↔MYT month-edge handling correct; per-year LP runs isolated; SoC reset between years is a deliberate simplification with negligible NPV impact). Two modeling choices remain noteworthy:

1. PV degradation extrapolation past year 5. The LP simulation runs over the 5-year weather window (2020-2024) with intra-window degradation applied. For the 20-year NPV horizon, years 6-20 use the 5-year-mean savings flat, rather than continuing to degrade at 0.5%/yr. Replacing with proper continuing-degradation reduces PV-only NPV by ~7.7 M RM (-9%, from +82.7 to ~+75.0 M). Conclusion: PV-alone is still bankable under the more conservative model (still > 8% project-finance hurdle), but the bargaining-range cliff at 13-15% IRR shifts down by ~1-2pp. Carbon-stacked structures unaffected (their +211 M margin absorbs the shift trivially).

2. Year-end BESS SoC discarded between years. The LP solves each year independently, starting BESS at SoC=0. End-of-year SoC is not carried forward. Order of magnitude: at 60 MWh BESS × 4 boundary years × 0.5 typical end-of-year SoC × peak rate = ~16 kRM/MWh missed value across the 20-yr horizon. Negligible (<0.01% of NPV).

These are documented for transparency rather than fixed because (a) neither changes the strategic conclusions and (b) the cascade-update work across reports outweighs the analytical gain.

Defense-in-depth follow-ups

To prevent re-occurrence, post-v3 the codebase has:

1. Runtime auto-derive in CLI commands: aggregator, trigger, ppa-pricing, risk all read btm_economics_dc100.json at runtime instead of using hardcoded constants.
2. Static module-constant guards: pytest asserts the per-MW PV and per-MWh BESS-incremental fallback constants in aggregator.py and trigger_curve.py stay within 5% of LP-derived values.
3. Drift checker (jb-vpp model check-summary --strict): 15 narrative citations (in EXECUTIVE_SUMMARY.md and other reports) verified against live JSON every CI run.

---

v2 update — cooling CDH

DC load model upgraded from flat 7×24 to temperature-dependent:

$$\text{load}_t = \text{IT}_{\text{MW}} \times (\text{PUE}_{\text{base}} + s_{cool} \cdot \max(0, T_t - T_{\text{cool}}))$$

with $s_{cool} = 0.008$ /°C, $T_{\text{cool}} = 25$ °C, base PUE = 1.4. At JB’s 5y mean ambient (26.94 °C), effective PUE is 1.415 (+1.1%); at 32 °C hot-month peak, PUE = 1.456 (+4%).

Side-by-side: flat-load vs cooling-CDH (LP dispatch, 5y avg)

Scenario @ ICPT 16 sen	Flat save	Cooling save	Δ save	Flat NPV	Cooling NPV	Δ NPV
30 MW PV alone	12.96	13.62	+0.66	+11.9	+18.4	+6.5
30 MW PV + 30 MWh BESS	15.39	16.18	+0.79	−178	−171	+7.8
60 MW PV alone	25.92	26.59	+0.67	+23.9	+30.5	+6.6
60 MW PV + 60 MWh BESS	30.77	31.60	+0.83	−357	−349	+8.1
90 MW PV alone	38.88	39.56	+0.67	+35.8	+42.4	+6.6
90 MW PV + 90 MWh BESS	46.16	47.00	+0.84	−535	−527	+8.2

(All RM millions/yr unless noted. NPV improvements concentrate on high-CDH peak hours where PV and BESS displace expensive grid kWh.)

What the cooling lift actually adds

• Per-MW PV: +11 kRM/yr (+2.6% over flat baseline). Cooling load is biased toward afternoon hours, which overlap the ETOU peak window (14–22 MYT) — those incremental kWh are displaced at peak rates.
• Per-MWh BESS: +14 kRM/yr (+17% over flat baseline incremental). Higher peak demand → more MD to shave → BESS captures more.
• PV-only IRR: 9.5% → 9.9%. Crosses the 10% hurdle that some corporate sponsors apply.

Strategic implications

The cooling refinement is model fidelity, not a strategic shift. The dominant conclusions (PV bankable solo, BESS gap requires capex fall + service contract) are unchanged. The main commercial use: show ~+10 M NPV uplift in the v2 deck vs v1 to demonstrate model maturation and quantify how the project benefits from operating in hot markets. Useful framing for hyperscaler-anchor commercial discussions because it’s exactly the math the tenant will run for their own BTM case.

---

v1 update — LP results

The v0 heuristic discharged BESS in early peak hours and ran out by 18–22 MYT, delivering near-zero MD value. v1 replaces the heuristic with an LP-optimal monthly dispatch (pulp + CBC) that knows about the MD term in the objective. Result: same scenarios, better dispatch.

Scenario @ ICPT 16 sen	Heuristic save	LP save	Lift	LP NPV	LP IRR
30 MW PV alone	12.96	12.96	—	+11.9	9.5%
30 MW PV + 30 MWh BESS	13.48	15.39	+14%	−178	−4.3%
60 MW PV alone	25.92	25.92	—	+23.9	9.5%
60 MW PV + 60 MWh BESS	26.95	30.77	+14%	−357	−4.3%
90 MW PV alone	38.88	38.88	—	+35.8	9.5%
90 MW PV + 90 MWh BESS	40.43	46.16	+14%	−535	−4.3%

(All savings in RM millions/yr; NPV in RM millions; LP results from CBC solver with MD constraint enforced over the ETOU peak window.)

What the LP actually does that the heuristic didn’t

1. Spreads discharge across the entire 8-hour peak window instead of front-loading. With a 30 MWh / 7.5 MW BESS (60 MW PV scenario sized to PV/4 power), heuristic discharges 4 hours at 7.5 MW then sits empty. LP discharges 3.5 MW continuously across all 8 peak hours, shaving 3.5 MW of MD instead of 0 MW.
2. Charges from off-peak grid when PV surplus is unavailable. v0 had already added this rule, but LP optimises the timing — it charges in the cheapest off-peak hours, not the first available ones.
3. Curtails PV when needed (rare at this PV/load ratio, but happens).

Why BESS still loses money at any size

The annual incremental savings BESS delivers per MWh capacity is ~80 kRM/MWh/yr under LP dispatch (60 MW PV scenario: 4.85 M RM extra ÷ 60 MWh = 81 kRM/MWh/yr). Decomposing:

• Energy arbitrage: ~28 MWh delivered per cycle × 250 weekday cycles × spread (peak − off-peak − ICPT cancels out at first order) × RTE = ~830 kRM/MWh/yr per cycled MWh. Already capped — can’t cycle more than once a day.
• MD reduction: 3.5 MW shaved × 35.50 RM/kW/mo × 12 = 5.0 M RM/yr per 3.5 MW of constant-rate MD shave = ~167 kRM/MWh/yr if BESS is sized exactly 4-h relative to MD rate.

Total LP capture: ~80 kRM/MWh/yr (lower than my back-of-envelope because weekend hours don’t contribute — only ~250 cycle-weekdays/yr useful).

Capex 5,000 kRM/MWh ÷ 80 kRM/MWh/yr = 62-year simple payback. With 8% WACC and 20-year horizon, NPV breakeven needs ~510 kRM/MWh/yr in savings, or 6.4× the current capture. Three paths to closing the gap:

1. BESS capex 5M → 1.6M RM/MWh (USD 320/kWh, possible by 2028–2030 per BNEF curves)
2. VPP service revenue stack on top — frequency response (DRC), capacity payments under future markets, ENEGEM dispatch on the MY → SG leg. Conservatively need ~430 kRM/MWh/yr from non-BTM sources to close the gap at current capex. That’s a steep ask but not impossible if multiple small revenue streams stack.
3. Increase peak/off-peak spread — would require ICPT > ~40 sen/kWh sustained, which is the high-fuel-cost stress scenario, not the base case.

Verdict

LP dispatch is a 14% lift but does not change the structural bottom line: at 5 M RM/MWh installed cost and 100 MW DC scale, BESS is value-destroying on a pure BTM avoided-cost basis. The path to BESS-positive is one of:

• Wait for capex to fall (timing-dependent, beyond our control)
• Stack additional revenue streams via VPP service contracts
• Find a customer with a much larger peak/off-peak spread or a tariff with capacity payments

For the JB anchor commercial case today, the recommendation stays: fund PV alone, structure BESS as a future option contract that triggers when capex falls or a VPP service contract anchors it.

---

What v1 needs to fix

Deficiency	Impact	Fix
BESS dispatch under-utilises MD reduction	−5 to −6 M RM/yr per scenario	LP-optimised or “save SoC for late peak” heuristic
Flat DC load ignores cooling auxiliary CDH variation	~3% diurnal underestimate	Multiply by (1 + 0.012 × CDH_t) from POWER T
Tariff E3 rate values are placeholder	±20% absolute number error	Verify against current TNB PDF
ICPT 2025-H1 onwards is placeholder	Scenario-only matters for forward NPV	Update yaml each 6-month cycle
PV degradation flat 0.5%/yr	1–2% NPV underestimate	Match warranty curve from EPC quote
No grid-export revenue line	Misses occasional ENEGEM upside	Add bess_can_export flag and a USEP price feed (blocked)
No real DC anchor data — using stub	Can’t compare vs actual sites	Pull from dc_tracker/jb_data_centers.yaml once populated

Bottom-line VPP design takeaway (v3)

For a 100 MW JB data-center anchor under current ICPT levels (post-v3 numbers):

1. A 60 MW BTM PV system is the core bankable position. ~210M RM capex, 31.9M RM/yr savings, 8-year payback, 13.0% IRR, +83 M RM NPV at 8% WACC. Independent of ENEGEM / VPP service / green-attribute monetisation.

2. BESS as configured today is value-destroying. Don’t add BESS unless either (a) the dispatch is LP-optimised to capture MD reduction, or (b) BESS capex falls 30–40% (USD 600/kWh range), or (c) a VPP service revenue stream (e.g., ENEGEM dispatch) is contracted.

3. PV size sweet spot is 50–70% of gross load. Above this, curtailment eats marginal returns; below this, the absolute savings are too small to justify development cost.

4. The commercial structure should isolate the BTM PV case from the BESS / ENEGEM / VPP case — bank the PV, fund the rest as upside-tracking options. This matches the actual economics produced by the model.