Indonesia SMR Deployment: The First Plant Is Not Supposed to Maximize Returns
- May 7
- 10 min read
There is a sentence in our Indonesia session that every deep-tech founder building physical infrastructure should read twice. It came from Matt Wilkinson, CEO of ThorCon International, at the end of a long answer about how you finance something no bank will lend against:
"The strategic principle is simple: the first plant's goal is not to maximize returns. It is to get built, get operating, and establish a track record that makes every subsequent plant commercially financeable."
Most founders in capital-intensive hardware treat the first unit as a commercial deal that happens to be early. Wilkinson treats it as a different asset class with a different purpose. That distinction is the most portable idea to come out of this webinar series, and Indonesia SMR deployment is where it gets tested.
The supply wall is real, and it just got steeper
Indonesia is ASEAN's third-largest data centre market. Greater Jakarta supports roughly 300 MW of operational capacity. Power generation is growing at under 7% annually while data centre demand accelerates at roughly 30%.
We opened the session with a projection of nearly 800 MW by 2031. Kengo Murakami of Cushman & Wakefield, who advises hyperscalers on site selection in Indonesia, corrected us on air:
"First, let me correct one point. I think the projection by 2031 is closer to 1.5 GW, which is double 800 MW. That is mostly coming from AI."
We are publishing his correction rather than our original figure, because his is the number worth planning against. He then made it concrete: this is the first year an AI campus is being built in Jakarta, going live around 2027 or early 2028, at roughly 300 MW from a single hyperscaler. One company is adding capacity equal to everything currently live in Jakarta.
That is what a supply wall looks like from the demand side. Not a forecast — a construction schedule.
Murakami also noted a nuance in what that demand actually wants. Many of the AI projects currently under construction in Indonesia are being built by non-US hyperscalers with less stringent carbon-free requirements. For Google or AWS, "certainty around carbon-free energy is very important." Today the available answer is renewable energy certificates: Equinix, which globally runs almost entirely on renewables, operates on conventional power in Indonesia and buys RECs from PLN. His view is that this changes once a firm carbon-free option exists.
Policy recognition is not investment-grade certainty
Indonesia's roadmap has moved substantially. Wilkinson catalogued it precisely: the 2025–2034 electricity supply plan sets a 500 MW SMR target for 2032 — "the first concrete nuclear figure in an official electricity planning document." Nuclear was formally included in the National Energy Policy with the longstanding "last resort" condition removed. PLN received explicit authorisation to procure and develop nuclear. BAPETEN, the regulator, is updating its licensing framework for modern designs.
Then he drew the line that matters:
"Taken together, the direction is right. But policy recognition is not equivalent to investment-grade certainty."
What Indonesia SMR deployment still needs from policy
What closes that gap, in his account, is three specific instruments: a standardised licensing pathway with clear and enforceable timelines, a credible offtake and tariff framework that makes revenue foreseeable over a 30-year horizon, and a well-defined, proportionate nuclear liability regime. He pointed to the United States as a jurisdiction that lacked these for years and then built them deliberately — the ADVANCE Act, production tax credits, DOE loan guarantees, and the NRC's shift toward risk-informed, technology-neutral licensing. His framing: "These were not symbolic gestures. They were deliberate mechanisms to reduce first-of-a-kind risk."
Wilkinson also raised governance durability, and was candid about why long-horizon investors watch it. In his words, investors have long memories, and the Freeport-McMoRan experience at Grasberg — decades of investment followed by midstream renegotiation of a legally established agreement — "remains a reference point for anyone committing long-term capital in Indonesia." He credited Indonesia's real progress on transparency and anti-corruption, while arguing that nuclear-scale capital additionally requires independent regulatory oversight insulated from political interference, contract sanctity with access to international arbitration, and transparent procurement. His conclusion was not pessimistic: "Indonesia has the demand, the political will, and the policy foundation. What it needs next is the institutional architecture that turns a roadmap into a bankable project."
Rully Hidayatullah of the ASEAN Centre for Energy described that architecture being assembled. Indonesia is preparing a NEPIO — a Nuclear Energy Program Implementing Organization, following the IAEA Nuclear Infrastructure Milestones framework, as the Philippines has done. Given deployment urgency in the 2030s and 2040s, he expects it to include not only technical stakeholders but government agencies, financial institutions, environmental stakeholders, and consumers. He also identified a specific unfinished step: Indonesia's National Strategic Projects framework, which confers priority status across major infrastructure, does not yet include nuclear.
Governance before infrastructure, not alongside it
Julius Trajano of RSIS made the sequencing argument, and it cuts against how most urgent infrastructure programmes actually run:
"In nuclear energy preparation, speed cannot come at the expense of credibility. Public trust and investor confidence are not byproducts of nuclear deployment. They are preconditions for it."
His prescription is a "governance before infrastructure" mindset — legislation, licensing systems, liability frameworks, emergency preparedness, and long-term waste management strategy prioritised before project decisions rather than developed in parallel with them. Supporting mechanisms include IAEA alignment, regulatory independence with sustainable human resource development, and regional cooperation through ASEANTOM, the ASEAN Network of Regulatory Bodies on Atomic Energy.
The line worth keeping: "In nuclear governance, credibility is built incrementally, but it can be lost very quickly."
He was also the only panelist to name the political reality of siting reactors near demand centres in a decentralised state. Regulatory frameworks built for large reactors will need risk-informed, site-specific licensing, rethought emergency preparedness zones for dense environments, and cybersecurity designed in from the start. But the binding constraint may be local rather than technical:
"What happens if the governor of Jakarta decides to prohibit construction of any nuclear power plant in the city or province to support data centers? That is a question that needs to be addressed."
On public trust, his view was that it is "built less through messaging and more through visible institutional competence, regulatory independence, and consistent safety performance over time." His suggested starting point is an asset Indonesia already has: decades of nuclear science and research reactor operation through institutions such as BRIN. "That shows the public that Indonesia is not starting from scratch."
Why the grid is not the plan
Wilkinson's description of physical reality explains the whole shift toward private power:
"Indonesia operates five largely disconnected wide-area grids, with 47,000 kilometers of new transmission called for in the current ten-year plan. But the financing and private sector frameworks to build that at scale do not exist yet. For a data center operator that needs reliable, high-quality power today, waiting for centralized grid capacity is not really a strategy. It is a risk."
He placed it in global context with three commitments that are already signed: Microsoft's 20-year deal to restart Three Mile Island, Google's first-ever corporate SMR purchase agreement with Kairos Power, and Amazon's commitment of nearly 2 GW of nuclear capacity through 2042. "These are not exploratory conversations."
The behind-the-meter advantages he listed are unglamorous and decisive: no transmission losses, long-horizon price certainty, control over reliability, and no competition with the grid's constrained capacity. On that last point he made an equity argument rather than an engineering one, citing a $9.3 billion capacity price increase in a single PJM cycle in the United States that added over $15 a month to residential bills in surrounding states. His position: "Indonesian consumers and industry should not have to subsidize hyperscale data centers."
He named the trade-off honestly too. Every plant goes offline for maintenance, and a behind-the-meter system sized exactly to load has no margin when the reactor is down. ThorCon's answer is a floating gas plant that travels between sites to cover scheduled downtime — a solution notable mainly for being unromantic.
The part of his answer least connected to data centres was the most striking. Indonesia is an archipelago of more than 17,000 islands, where roughly 5.8 GW of installed capacity runs on diesel at up to 65 cents per kilowatt-hour, supported by more than $2 billion a year in government subsidies. Those communities cannot be reached by transmission cable at any reasonable cost. "They are not waiting for Java-Bali to extend to them." For ThorCon, he said, "this is not just a data center story. It is an Indonesia story."
Founders should note what just happened there. A company positioned for the AI power market articulated a second, larger, less competitive market that the same product addresses. That is not messaging. That is optionality, and it changes how the first plant gets underwritten.
Murakami confirmed the geography is already shifting on the demand side: hyperscalers are building in suburban locations such as Jababeka and Karawang rather than dense urban connectivity hubs, because land costs are lower and power infrastructure can be built alongside. Rully Hidayatullah added that captive power is not novel in Indonesia — smelters and remote industrial clusters already operate their own plants — and that off-grid, grid-to-grid, and point-to-point arrangements exist in some form. The unresolved question is regulatory: for nuclear, it is "not only about electricity regulations. It is also about nuclear safety infrastructure and safety governance."
The two financing problems in Indonesia SMR deployment, and why founders conflate them
The most useful analytical move in the session was Wilkinson's insistence on separating first-of-a-kind financing from nth-of-a-kind financing. They are not the same problem and they do not have the same solution.
For nth-of-a-kind, the logic is conventional: a creditworthy offtaker signs a long-term PPA, and that PPA becomes the backbone of a project finance structure that commercial lenders can underwrite against contracted cash flows over 30 years. Fuel cost is essentially fixed, capacity factor exceeds intermittent renewables, and "for a sophisticated investor, the profile is highly attractive once technology risk is removed."
For first-of-a-kind, none of that applies:
"Commercial lenders do not lend on unproven technology. Cost benchmarks do not exist. The regulatory pathway has not yet been walked. First-of-a-kind capital has to come from a different universe, one motivated by policy, strategic positioning, or national interest rather than pure risk-adjusted return."
He then described two live paths. The first is geopolitical: US foreign policy interest in seeing American rather than Chinese or Russian nuclear technology embedded in Indonesian infrastructure, executed through the US International Development Finance Corporation and the Export-Import Bank, with PLN or another designated state entity as plant owner and lead borrower while ThorCon contributes technology and accepts flexible economics on the first project. The 123 Agreement is signed; a reciprocal trade agreement was recently added.
The second is commercial: an anchor offtaker structure in which a hyperscale data centre operator co-develops the first plant in exchange for a long-term behind-the-meter PPA. His reasoning is worth quoting because it explains why this works where a normal customer relationship would not:
"A tech anchor does two things at once. It provides development capital and creditworthiness as an offtaker. That transforms an otherwise unbankable project into one development lenders can support."
He was equally clear about what ThorCon will not do. Nuclear-as-a-service, where the developer retains ownership and sells power under long-term contracts, is "the right model in theory" and may suit specific plants with the right partners and financial backstop. But "a small technology-focused company carrying full ownership risk on every plant is not feasible." Their near-term model is technology licensing — recurring revenue that scales with the fleet without requiring ThorCon to be the balance sheet behind every project.
Rully Hidayatullah added the institutional dimension, noting that multilateral development banks including the World Bank and ADB are reconsidering long-standing prohibitions on nuclear financing. His point was that their most useful contribution may not be project capital at all: grants and frameworks for capacity building, regulatory advancement, and preparatory work before construction. "If a country has its regulatory framework in place, institutional readiness, and financing readiness, it can attract more investment."
Trajano closed the loop on why strong regulation is a commercial asset rather than an obstacle: "Investors and private offtakers will only commit capital if they trust that nuclear safety oversight is robust, rules are stable and clear, and governance institutions are credible over time. Strong regulation is not a barrier. It is a prerequisite."
The answer nobody wanted
An audience member asked whether the first SMRs could realistically be ready within three years to serve data centres coming online now. Wilkinson's answer was one word long:
"No. We will not be deployed before 2030."
We are including it because that kind of answer is rarer than it should be, and because it is the honest frame for everything above. Trajano was equally direct on what fills the interim: for the next decade, Indonesia will rely on gas, coal, and some renewables — "bad for energy diversification and also bad for Indonesia's greenhouse gas emissions commitments." Wilkinson agreed there is no alternative bridge other than fossil fuels, predominantly coal, and said he did not have much hope that intermittent renewables without storage could close the gap.
Asked whether Indonesia risks losing digital sovereignty to faster neighbours, Rully Hidayatullah reframed the question rather than answering it competitively: "I do not see this only as a question of who runs faster, but how we can move together."
What this means if you are building
Three transferable lessons, none of them about nuclear.
Separate your first unit from your business model. The first unit exists to create a reference. Price it, structure it, and fund it accordingly — from capital motivated by policy, strategy, or national interest rather than risk-adjusted return. Wilkinson's phrase for it is that the reference plant "is the most valuable asset ThorCon can acquire."
Find the counterparty who is both customer and financier. An anchor offtaker with a balance sheet solves the credit problem and the capital problem in one signature. Founders spend enormous effort separating fundraising from sales when the highest-leverage move is often to merge them.
Indonesia will not have a reactor operating before 2030. The country is nonetheless where the demand, the policy foundation, and the political will already exist. What remains is the institutional architecture — and institutions, unlike reactors, can be built in parallel.
This article draws on Earth VC's webinar "Powering Indonesia's AI Boom: Scaling SMRs to Close the AI Infrastructure Gap," part of the Nuclear-AI Nexus series following our March 2026 report, Nuclear Energy Fuels AI Boom in Southeast Asia Data Centers.
Comments