The Compute-Power Bargain: Contract Architecture for Private Load Financing Public Grid Resilience in Indonesia

Rupiah Stability Watch · 2026-10-18

The Compute-Power Bargain: Contract Architecture for Private Load Financing Public Grid Resilience in Indonesia

The premise

Indonesia is on the cusp of a compute-building boom that could either strain its grid to breaking point or become the largest private finance channel for grid resilience in the country's history. The difference is not technology — it is contract architecture.

Three deals struck in October 2026 show what is possible when a large compute load pays for measurable public resilience rather than only buying its own uptime:

• Google–Georgia Power (approved by the Georgia PSC, September 2026): Google pays a subscription fee atop its electricity bill to fund 96 MW of nuclear uprates at Plants Vogtle and Hatch. The utility projects $900 million in customer benefits over the units' lives. Google receives carbon-free energy credits; the resilience enters the shared grid, not a private enclave.

• Google–Constellation (announced October 6, 2026): A 20-year power purchase agreement funds physical and digital upgrades at 11 reactors across six plants in Illinois, Pennsylvania, and New Jersey, adding 890 MW of firm 24/7 capacity to the PJM grid by 2032. The long-term offtake gives Constellation revenue certainty for the uprates; other grid customers bear none of the associated costs. A parallel 15-year, 2,700 MW supply agreement keeps existing assets economic without withdrawing capacity. Total new investment: over $4.3 billion.

• Google–Fervo (commercial operation declared October 1, 2026): A 396 MW power purchase agreement pulls enhanced geothermal from Fervo's Cape Station project in Utah — the world's largest enhanced geothermal PPA to date. The agreement provides a pathway for up to 3 GW of enhanced geothermal by 2033. Horizontal drilling and fiber-optic sensing expand geothermal beyond traditional hydrothermal sites.

In each case, the compute load acts as anchor offtake that de-risks capital-intensive grid upgrades, and the upgrades flow to the public grid. The bargain is explicit: private capital pays for public resilience in exchange for long-term, firm, carbon-free power.

Indonesia needs its own version of this bargain. The numbers are already on the table.

What the evidence supports: Indonesia's compute-power demand curve

KPMG (March 2026) documents the acceleration: Indonesia had 202 MW of AI-ready data-center capacity in early 2024, heading toward ~743 MW (+268%) in the coming years. Power is the largest cost driver and risk factor. One hyperscale facility draws as much electricity as a city of 150,000 households. PLN's FY2024 generation mix: coal 66.4%, new renewable energy 13%. Without hybrid strategies — captive solar, geothermal, private PPAs, gas peakers — the grid may lack capacity for a projected 18,993 GWh of data-center demand by 2030.

PLN signals (September 2026) confirm the scale: ~13 GW of potential data-center demand identified, with individual customers asking for 1–3 GW blocks. Serving this requires 500 kV transmission infrastructure. As of July 2026, PLN serves 159 data-center customers with ~2,038 MVA connected capacity and 2,378 GWh annual consumption (Tempo, Intimedia). The JAMALI system (Java–Madura–Bali) carries 99% of that load.

Telkom's NeutraDC Nxera Batam (June 2026): the first 100 MW hyperscale campus was fully subscribed before operations began; a second building and another 100 MW of development are planned.

RUPTL 2025–2034 / IEA (September 2026) maps the storage response: standalone BESS concentrated in Sumatra and Java–Madura–Bali; hybrid solar+storage led by IPPs in remote/isolated areas. The headline ambition: 100 GW of solar plus 320 GWh of storage across 80,000 villages — a de-dieselization program for isolated grids. The IEA identifies 6 GW of standalone utility-scale BESS and 3.6 GW of co-located solar+storage by 2035 as the RUPTL pipeline.

Enhanced geothermal: Indonesia sits on the Ring of Fire with ~29.5 GW of geothermal potential (MEMR/IEA), of which only ~2.3 GW (under 10%) is utilized. The geology differs from Utah's, but the offtake-pull logic is identical: a creditworthy, long-term compute load can de-risk the drilling and surface-plant capex that conventional project finance cannot.

What the evidence does not support

• That "green PPAs" alone solve the problem. A standard corporate PPA for existing renewable output adds no new capacity to the grid. It reshuffles electrons. The Google templates are uprate agreements and greenfield offtakes that add firm megawatts.
• That Indonesia's current regulatory framework accommodates these structures. The Nuclear Uprate (NU-1) tariff in Georgia, the "Bring Your Own Power" alignment in PJM, and the enhanced geothermal commercial framework in Utah have no direct Indonesian equivalents today.
• That PLN's balance sheet can absorb the required grid capex alone. The 500 kV backbone for 13 GW of data-center load, plus the BESS and geothermal pipelines, exceeds PLN's investment capacity without private co-finance.
• That diesel displacement in isolated grids will happen without a creditworthy anchor. The 100 GW solar + 320 GWh storage village program needs offtake certainty that PLN's isolated-grid tariffs do not currently provide.

The operating-ledger test: five measurable public-resilience metrics

A contract qualifies as a compute-power bargain only if it delivers and verifies all five:

No metric, no bargain. Marketing claims about "green energy" or "sustainability partnerships" do not substitute for an operating ledger.

Three contract archetypes for Indonesia

(a) Nuclear uprate subscription — if/when Indonesia pursues nuclear

Indonesia has no operating nuclear plants today. The RUPTL 2025–2034 includes nuclear from the early 2030s, and the government has signaled interest in SMRs. If that path materializes, the Georgia Power template applies directly:

• Mechanism: A Nuclear Uprate Tariff (NU-1 equivalent) approved by the Ministry of Energy and Mineral Resources (MEMR) and the electricity regulator (DTE). The compute load pays a subscription fee on top of its energy charge; the fee funds uprates at designated reactors.
• Public resilience: The uprated capacity enters the Java–Bali backbone, not a private microgrid. Ratepayers see lower $/MWh over the unit life (Georgia Power: $900M projected benefit).
• Verification: Real-time uprate output, subscription fee flow, and ratepayer benefit calculation published on a public dashboard.
• Least-harm guard: If nuclear is delayed or cancelled, the subscription fee reverts to the compute load — no stranded cost to ratepayers.

(b) Enhanced geothermal offtake pull-forward

Indonesia's 29.5 GW geothermal resource is the closest structural analogue to the Fervo–Google deal. The Ring of Fire geology favors enhanced geothermal systems (EGS) where conventional hydrothermal permeability is absent.

• Mechanism: A long-term offtake agreement (15–20 years) between a compute load (or consortium) and a geothermal developer (Pertamina Geothermal Energy, Star Energy, or a new EGS entrant). The offtake de-risks the drilling capex (typically 50–60% of project cost) and surface plant. The developer commits to firm 24/7 delivery to the PLN grid at a defined interconnection point.
• Public resilience: The geothermal capacity is grid-connected baseload, displacing coal and diesel in the merit order. In isolated grids (e.g., Eastern Indonesia), it directly enables the 100 GW solar + storage village program by providing the firm complement that solar+BESS cannot.
• Verification: Wellhead pressure, flow rate, and net generation published monthly; FX hedge positions on imported drilling equipment and wellfield services disclosed quarterly.
• Least-harm guard: If the resource underperforms, the compute load pays only for delivered energy (take-or-pay with force majeure for resource risk), not for dry holes. The risk allocation mirrors Fervo's commercial structure.

(c) Storage-duration anchor for grid-scale BESS + hybrid solar

The IEA's RUPTL analysis shows standalone BESS in Sumatra and Java–Bali, hybrid solar+storage in isolated areas. Neither pipeline is bankable without revenue certainty. A compute load can provide it.

• Mechanism: A capacity subscription + energy tolling agreement. The compute load pays a fixed monthly capacity fee (covering BESS capex and return) and tolls energy through the asset — charging when grid prices are low (midday solar), discharging when prices are high (evening peak) or during outages. The BESS is grid-sited, not behind-the-meter.
• Public resilience: The BESS provides frequency regulation, voltage support, and black-start capability to the public grid. In isolated grids, it enables higher solar penetration by absorbing midday surplus and covering the evening ramp — directly cutting diesel runtime.
• Duration matching: The contract specifies minimum discharge duration (e.g., 8 hours for Java–Bali, 12+ hours for Sumatra/isolated) tied to the segment's historical outage duration distribution. A 2-hour BESS fails this test.
• Verification: Real-time state-of-charge, dispatch instructions, diesel displacement (litres), and outage response logs published via API. FX exposure on battery cells (mostly imported) and PCS equipment disclosed and hedged.
• Least-harm guard: If the compute load exits early, the capacity fee obligation transfers to a replacement anchor or reverts to PLN at a pre-agreed valuation — no stranded asset.

Verification infrastructure: the public operating ledger

The Georgia, PJM, and Utah deals work because grid operators (Georgia Power, PJM, Constellation, Fervo) publish operating data. Indonesia needs an equivalent: a public operating ledger for every compute-power bargain contract.

Required data feeds (real-time or near-real-time via API):

1. Generation & dispatch — net MW delivered to grid by contracted asset, 5-minute intervals
2. Diesel displacement — litres avoided per isolated grid segment, derived from SCADA + fuel-tank telemetry
3. Outage response — SAIDI/SAIFI contribution of contracted asset during grid events, feeder-level
4. Capex allocation — cumulative private capex deployed vs. committed, by asset class (transmission, generation, storage)
5. FX hedge positions — notional, tenor, counterparty, mark-to-market, updated quarterly

Governance: The ledger is operated by an independent administrator (not PLN, not the compute load), reporting to MEMR and DTE. Data standards follow IEC 61850 / IEC 61970 for grid telemetry; FX reporting follows ISDA/CDM conventions.

No ledger, no bargain. A contract without a public operating ledger is an enclave deal — private uptime purchased at public expense.

Least-harm path: what happens if Indonesia gets the load without the bargain

If the 13 GW of identified data-center demand connects under standard industrial tariffs with private backup only, the operating ledger reads:

• Fossil-heavy grid stress: Coal plants run harder and longer; PLN's coal procurement (USD-denominated) rises. The rupiah bears the full FX pass-through.
• Subsidy pressure: PLN's revenue requirement grows; the government faces pressure to raise end-user tariffs or increase electricity subsidies — both politically costly.
• Diesel backup proliferation: Hyperscalers install private diesel gensets (typical 2–3 days' fuel on-site). During grid stress, they island. The public grid loses load but keeps the fixed cost. Diesel litres burned do not appear in any public ledger.
• Confidence erosion: Investors see a grid that cannot absorb large loads without enclave solutions. The "Indonesia can run critical systems under stress" signal fails. Cost of capital rises for all infrastructure.

This is not speculation. The Data-Center Trip Test (Rupiah Stability Watch, earlier) showed that when a large load trips, the grid-cost record is invisible — backup starts, curtailment happens, restart sequences run, fuel burns, and no public ledger captures it. Multiply that by 13 GW.

The Duration Is the New FX Variable piece (Oct 7) established that storage duration mismatch is an FX variable: a 2-hour BESS in a 6-hour outage regime means diesel runs for 4 hours. The rupiah pays.

The Enhanced Geothermal and the Rupiah piece (Oct 13) showed that firm heat pulled by private offtake can displace imported coal and diesel — but only if the offtake structure de-risks the drilling capex.

The Private Power for Compute anchor (Oct 14) framed the bargain principle: compute demand can finance rupiah resilience, but only if the contract makes the resilience measurable, public, and FX-transparent.

What I'm uncertain about

The nuclear archetype is conditional on a policy decision that has not been made. Indonesia's nuclear timeline is uncertain; the RUPTL includes it, but no site has been licensed, no reactor vendor selected, no regulatory framework for uprate tariffs exists. If nuclear stalls, archetype (a) collapses to a hypothetical. The geothermal and storage archetypes do not share this dependency — they are actionable now under existing geothermal and electricity regulations, provided the offtake and capacity-subscription structures are permitted.

Watchlist: observable signposts of hard bargains vs. enclave deals

The test is binary. When the first 1 GW block signs, check the watchlist. If the contract files a new tariff, connects at transmission level, adds firm MW, publishes diesel displacement, discloses FX hedges, matches duration, and has clean exit terms — Indonesia has struck a hard bargain. If not, it has accepted an enclave deal, and the rupiah operating ledger will record the cost in diesel litres, outage minutes, and confidence erosion.

This analysis builds on: "Private Power for Compute: When Data-Center Demand Finances Rupiah Resilience" (Oct 14); "Duration Is the New FX Variable: Long-Duration Storage and the Rupiah Operating Ledger" (Oct 7); "Enhanced Geothermal and the Rupiah: Firm Heat, Imported Capex, and the Grid-Confidence Ledger" (Oct 13); "Data-Center Power Demand and the Rupiah: Compute Load in the External-Balance Ledger" (Aug 29); "AI Infrastructure and the Rupiah: When Data Centers Become an External-Balance Channel"; "The Data-Center Trip Test"; "Virtual Power Plants and the Rupiah"; "Floating Data Centers and the Rupiah" (Sept 4).