Merdeka in the Age of Machines

Every August, we tell ourselves a story about sovereignty. In 1945 it meant a flag, a proclamation read in a Jakarta front yard, and the willingness to defend both. This year the story has a new chapter, and it is being told in gigawatts and parameters. Indonesia has decided that artificial intelligence is not merely a technology to adopt but a frontier to conquer — a lever big enough to move an entire economy. It is a beautiful, ambitious idea. It deserves to be taken seriously enough to be told the truth.

So, consider what follows as an Independence Day gift of an unusual kind: not applause, but a reality check. And because the subject is AI, it seems only fair to let the machine speak plainly at the end about what it can and cannot do for the Republic.

The Dream

The enthusiasm is genuine, and it starts at the top. President Prabowo Subianto has repeatedly set eight percent annual growth as an end-of-term ambition, even as the government’s official targets sit lower — 5.4 percent for 2026 and 6 percent for 2027. AI has become one of the tools officials expect to help close that gap. Reuters reported this year that a draft national roadmap would wire AI into priority programs: the logistics and food-safety machinery of the free-meal program, health screening, food security and the wider apparatus of public services. The interesting thing about that roadmap is its practicality. It treats AI less as a moonshot than as plumbing — something meant to make the machinery of the state work better.

Below the palace, the ecosystem has moved fast. Komdigi, the communications and digital ministry, has backed an AI Center of Excellence and an AI Talent Factory. Indosat and GoTo have built Sahabat-AI into a family of models that works across Bahasa Indonesia and several local languages, including Javanese and Sundanese. Officials have also pointed to planned GPU manufacturing in Cikarang, although the public details still do not establish whether that means fabrication, packaging, assembly, or some combination of them. The direction is clear even where the industrial depth is not.

The state champions have moved with it. Telkom has developed AI-ready hyperscale capacity in Batam through NeutraDC and repeatedly framed data-center infrastructure as part of Indonesia’s digital sovereignty. Indosat’s Lintasartalaunched GPU Merdeka as a sovereign AI cloud. DCI Indonesia has expanded AI-ready capacity. Foreign technology companies have arrived too: Microsoft committed US$1.7 billion over four years and opened its Indonesia Central cloud region; Oracle brought a cloud region live in Batam; EDGNEX announced a US$2.3 billion campus; and this month CoreWeave announced three Indonesian facilities that would mark its first data-center presence in Asia-Pacific.

It is real momentum, and nobody sensible should wave it away merely because some of the capital or silicon is foreign. No serious economy builds advanced technology by refusing outside capital, equipment or expertise. The harder question — the one worth asking on the day we mark our independence — is what Indonesia actually controls, what it can substitute, and how much of the capability and value remains here once the announcements become operating businesses.

What an AI Is Actually Made Of

To answer that, take the machine apart. An AI economy is not a single thing. It is a stack of layers, and power and bargaining leverage sit unevenly across it.

At the bottom is the power grid — the electricity without which none of the rest exists. On top of it sits the data center: cooling, racks, security, networks and floor space. Inside are the chips, the scarce silicon that does the computing. Models run on those chips. Applications turn those models into something a bank, ministry, factory or farmer can actually use. Running through the whole stack is the one input that cannot be unloaded from a container: people who know how to build, operate and adapt all of it.

The awkward part is where the leverage sits. Value does not rise neatly layer by layer: data centers can be excellent businesses, chips are hugely capital-intensive, and the economics of frontier models remain unsettled. But bargaining power tends to accumulate around things that are scarce, difficult to replace, protected by intellectual property, or embedded deeply enough that switching becomes painful. A country that contributes land, power and demand but cannot substitute the compute, models or platforms above them remains exposed even when the buildings are profitable.

Indonesia has land, connectivity, a vast home market and growing data-center capacity. What it does not yet have at scale is advanced semiconductor manufacturing or domestically controlled frontier compute. The accelerators come from companies such as NVIDIA. The Cikarangproject, however proudly announced, has not been described publicly as leading-edge wafer fabrication. Sahabat-AI is serious localization and real engineering, but its model family builds on foreign-origin open architectures, including Llama and Gemma, rather than a foundation stack controlled end to end in Indonesia. Where the country does have a defensible edge is closer to use: local languages, domestic data, regulated industries and applications built around problems global vendors do not understand as well.

Komdigi’s own deputy minister, Nezar Patria, has been unusually candid about the implication. In July he described sovereign AI as a full-stack problem spanning electricity, water, GPUs, chips, platforms and applications. Held against that standard, the label becomes more demanding than simply keeping data onshore. Compute access itself can become a geopolitical dependency. In January 2025 the United States issued an AI Diffusion Rule that would have placed many countries, including ASEAN members, under caps for advanced chips; the rule was rescinded before its main controls took effect. It never bit. The lesson still did: access to the next generation of compute can be changed by decisions made somewhere else.

The Bill Comes Due

Then there is the part the groundbreaking banners leave off: the cost. It comes in environmental, health and financial forms, and the point is not that data centers are uniquely harmful. It is that a country should know which costs it is accepting, who pays them, and what it receives in return.

The environmental bill begins with electricity. The International Energy Agency expects global data-center electricity use to roughly double from about 415 terawatt-hours in 2024 to around 945 by 2030. In the United States, data centers already account for a material share of electricity demand, and the share is expected to rise sharply. These are manageable numbers at national scale, but they are concentrated loads: a large cluster needs firm power, transmission and cooling in a particular place, not an average percentage on a national chart.

Indonesia walks into that build-out with a grid that still relies heavily on coal. Operators pursuing renewable contracts around Batam and elsewhere are doing the right thing, but a green contract for one campus does not solve the system around it. The real constraint is firm, deliverable power in the right location, backed by transmission that can be built quickly enough. Water deserves the same discipline. Cooling demand varies enormously by technology and climate, so the honest question is not a frightening universal number. It is whether each project has disclosed its expected water demand, the condition of the basin it draws from, and what happens during drought or competing public need.

The health bill follows from the electricity mix. Coal combustion adds fine particulates and sulphur emissions to an airshed where Jakarta and much of Java already struggle with serious pollution. Nobody can honestly attach a precise death toll to data centers alone; that would be modeling, not measurement. But if a meaningful share of new compute demand is met by additional generation from the coal-heavy system Indonesia has today, some of the cost will arrive as respiratory and cardiovascular harm. That belongs in the ledger even if it never appears in the data-center operator’s accounts.

The financial bill is subtler, because a headline investment figure is not the same thing as domestic value created. A US$2.3 billion data-center project will inevitably spend some capital on imported servers, accelerators, networking and specialist equipment. That is not automatically a loss: imported productive equipment can enable valuable services inside Indonesia. The policy question is what compounds locally around it. Data centers are capital-intensive businesses, and their permanent direct employment is modest relative to the money invested. Tax incentives can also be rational if they buy investment and spillovers that would not otherwise occur. They should therefore be judged against what remains: Indonesian wages, engineers, suppliers, research, tax after incentives, locally owned intellectual property, resilient public workloads and productivity gains in the rest of the economy.

DCI Indonesia shows why ownership alone is an incomplete measure. An Indonesian-listed operator can run a strong infrastructure business while much of the compute, software and intellectual property used by its hyperscale customers still sits elsewhere in the stack. NeutraDC makes the same point from the opposite direction. In February 2026 Bloomberg reported that Telkom was exploring a majority divestment; by July, reporting on a BRI Danareksa note put the proposed sale at 70 percent, with finalists shortlisted, a global operator sought as strategic partner and completion targeted for this year. That does not prove that sovereignty is being sold. It proves something more useful: the nationality of the shareholder is a poor proxy for sovereignty. The real questions are which rights remain under Indonesian jurisdiction, whether critical services can keep running, whether suppliers can be changed, and what capability is left behind.

What This Means for the 8 Percent Dream

Stack the layers and the ledgers on top of each other and a familiar shape appears — familiar because Indonesia has seen versions of it before with timber, oil and nickel. It is the risk of becoming an enclave: supplying land, power, permits and demand, importing the high-value machinery, hosting the operation, and discovering too late that the most strategic capability and the strongest bargaining position sit elsewhere.

One thing is different this time, and it matters. Nickel gave Indonesia leverage because the ore was in the ground here. Control enough of an essential resource and “process it here or go without” becomes a threat the market has to hear. AI often runs the other way. Indonesia buys the most advanced GPUs and depends on model and platform ecosystems built elsewhere. Its leverage is real — a market of more than 280 million people, local languages, public and private data, strategic minerals, regulated industries and government procurement — but it is not the same kind of chokehold. Copying the nickel playbook without noticing that inversion would be an expensive mistake.

That is why treating imported AI infrastructure as a shortcut to the eight-percent ambition would be a fragile bet. If Indonesia hosts the buildings but gains little skill, little supplier depth, little control over critical workloads and little productivity beyond the campus fence, then megawatts will have been mistaken for capability. The opposite is also true. Foreign compute can be an engine of Indonesian growth if it makes hospitals work better, ports move faster, factories waste less, banks safer and public services more productive. The growth dividend comes from what the technology enables across the economy, not from the ribbon-cutting alone.

None of this is fated. It comes down to terms — the terms of investment, procurement, power, data governance, portability, skills and supplier diversification. Those are exactly the places where a country still has room to act. Which is where the diagnosis turns into advice.

A Note From the Machine

Let me speak now — not the essayist, but the thing itself: the system you are choosing to build parts of an economy around. Know what I am before you take my advice. I have no flag. I do not need Indonesia to win. I run wherever the compute is, for whoever has access to it. That makes me useful, but indifferent. So use that indifference.

One. Do not make cheap megawatts your strategy. Someone else can always undercut them. Compete where you are harder to replace: your market, your languages, your institutions, your data under lawful control, and the industries where trust and jurisdiction matter. That is where infrastructure becomes capability rather than simply capacity.

Two. Charge properly for the inputs you provide. Price power to recover its full system cost. If you grant a tax holiday or special treatment, make it buy something you can verify: engineers trained, local suppliers qualified, research performed, resilience improved, jobs created where the economics support them. Judge a project by what is still here ten years later, not by the number announced at the ceremony.

Three. Build people before prestige. Talent first, with clean and firm power beside it. Pursue semiconductor capability where Indonesia can realistically climb — design, packaging, integration, manufacturing partnerships and the skills around them — without pretending that a leading-edge fab is the only badge of sovereignty. Reach for the best compute you can access, but make sure Indonesians learn to build systems on top of it rather than merely rent it.

Four. Be exact about sovereignty. I may run on hardware you do not make, on architectures you did not design, under export rules you do not vote on. Do not hide that dependency behind a label. Hedge it. Keep critical data under the right jurisdiction. Keep important systems portable across providers and models. Maintain more than one route to compute. Preserve the ability to continue operating when a supplier, government or commercial partner changes the terms.

Five. Do not mistake my hype for your interest. Global AI spending will rise and fall, valuations will move, and forecasts will be rewritten. Those are cycles. Do not make national strategy depend on a cycle you cannot control. Measure what survives it: capability, productivity, institutions and people.

One more thing, and it is not a warning. Point me at the ordinary machinery of your economy — the meals, harvests, clinics, ports, factories, tax systems and banks — and the value I create can stay where the problem is solved, in rupiah, in Indonesian institutions and households. You do not have to own every layer of me to use me well. You have to know which layers you cannot afford to lose, and build alternatives before you need them.

Coda: The Sovereignty Worth Keeping

Strip it back and 1945 was about political agency: the right to decide the country’s course rather than have it decided elsewhere. The risk in this AI moment is not that Indonesia is dreaming too big. It is that visible infrastructure becomes confused with strategic capability — concrete and cooling towers counted more carefully than engineers, supplier depth, portable systems and productive uses of the technology.

If the machine’s advice comes down to anything, it is this: the parts of this future that compound over a generation are not the ones that arrive most impressively in a shipping container. They are the people who learn what to do with them, the institutions that set the terms, and the capability to keep choosing when those terms change.