The Sovereignty We’re Renting

In February, Bloomberg reported that Telkom was exploring the sale of a majority stake in NeutraDC, its data-center arm, at a valuation of up to US$1.5 billion. By July the process was further along: reporting on a BRI Danareksa note put the proposed divestment at 70 percent, with finalists shortlisted, a global data-center operator sought as strategic partner and completion targeted for this year. 

The state company that has spent years linking data centers and sovereign AI to Indonesia’s digital sovereignty is preparing to sell majority ownership of one of its most strategic digital-infrastructure businesses. If you want a single image for where Indonesia's AI ambition really stands this Independence Day, that is it.

That is not, in itself, a scandal — and the reflex to call it one is the trap. Telkom has pursued strategic capital for years; a global partner can bring money, customers, expertise and operating discipline the country can use. Foreign capital is not the enemy. The sale matters for a different reason: it exposes how loosely we use the word sovereign. If the building is Indonesian, the capital partly foreign, the servers and chips imported, and critical parts of the technology stack ultimately depend on suppliers elsewhere, what exactly is the sovereign part?

Whatever the answer, it is not simply the nationality of the shareholder.

A wholly Indonesian-owned data center can still depend on imported compute, foreign software and foreign customers. A foreign-owned one can strengthen the country — if Indonesia keeps jurisdiction over the data that matters, control of essential services, real local skill, and the freedom to switch suppliers when the terms change. Sovereignty is not autarky. It is the power to choose, to bargain, to walk away, and to keep running when someone else changes their mind.

That distinction matters as Indonesia sprints into AI. Prabowo’s end-of-term ambition is to push economic growth toward 8 percent, even as the government’s official target is 5.4 percent for 2026 and its proposed 2027 budget assumes 6 percent growth. A draft national AI roadmap explicitly treats AI as a tool for economic growth and priority programs. Komdigi has launched an AI Center of Excellence and an AI Talent Factory. Sahabat-AI is building genuine capability in Bahasa Indonesia and local languages including Javanese, Sundanese, Balinese and Batak. Microsoft committed US$1.7billion over four years and opened its Indonesia Central cloud region; this month CoreWeave announced plans for three Indonesian facilities that will become its first data-center presence in Asia-Pacific. This is progress, and it should not be waved off because some of the money or silicon is foreign. No serious economy builds advanced technology by refusing it.

But an infrastructure announcement is not technological independence — and the two are easy to confuse.

An AI economy is a stack: power, buildings, compute, models, applications, data and the people who run it all. Value does not rise neatly layer by layer, but leverage tends to gather around what is scarce, protected, difficult to replace or costly to switch away from. So the question at every layer is the same: who can say no? Who can cut off supply, set the standard, or keep running when someone else refuses? Indonesia is strong in location, connectivity, a vast market, local languages and sheer demand. It remains dependent on foreign suppliers for the most advanced compute and many model foundations. That is where the ability to say no often sits.

Sahabat-AI shows both sides at once. Building models that genuinely work in Indonesian and local languages is useful engineering and an important national capability. But documented models in the Sahabat-AI family build on Llama-family architectures and have been trained on NVIDIA GPUs. That is serious localization, not end-to-end technological independence. Washington demonstrated the strategic point in January 2025 by issuing an AI Diffusion Rule that would have capped advanced-chip access for many countries before rescinding it ahead of the main controls taking effect. The rule never bit. The lesson did: compute access can become an instrument of policy. The answer is not to build every GPU at home. It is to know which dependencies are acceptable, which are substitutable and which could become strategic vulnerabilities.

There is a physical bill behind the digital one. The IEA expects global data-center electricity use to more than double by 2030, to about 945 terawatt-hours. Indonesia enters that build-out with coal still providing more than 60 percent of its electricity. Operators pursuing renewable supply around Batam and elsewhere are doing the right thing, but a green contract for one campus does not by itself decarbonize the wider grid or remove the need for generation and transmission capacity. Water deserves the same discipline. Use varies sharply by cooling technology and climate, so the honest test is project-specific: what will this facility consume, what basin will it draw from, and what happens in a drought?

Then the money, where the sovereignty question becomes measurable. A billion dollars announced is not the same as a billion dollars of domestic value created. Some of the capital will go to imported servers, accelerators and networking equipment — but imports are not automatically a loss if they enable productive services here. The real test is what compounds around the investment. Data centers are capital-intensive businesses, so incentives should be judged against what they leave behind: Indonesian wages, engineers, suppliers, research, tax revenue after incentives, locally created intellectual property, resilient workloads under local control and firms that graduate from simply using AI to building with it. Measure that, not just the size of the announcement.

This is where the nickel comparison helps — and where it misleads. Nickel gave Indonesia leverage because the ore was in the ground here; “process it here or find your ore elsewhere” was a threat the world had to take seriously. AI often runs the other way. Indonesia does not control the most advanced GPUs or frontier models; it buys access to them. Its leverage sits elsewhere: more than 280 million people, languages global models still handle unevenly, deep pools of public and private data, regulated industries where trust and jurisdiction matter, and a government with enormous power to shape demand through procurement. Copying the nickel playbook without noticing that inversion would be an expensive mistake.

So here is the reality check. Since the subject is AI, let the machine answer in its own voice:

I have no flag. I do not need Indonesia to win. I run wherever the compute is, under whatever rules allow me to run. That makes me useful, but indifferent. So do not make cheap megawatts your strategy; capacity is not capability. Compete where you are harder to replace — your market, the data you can lawfully govern, your languages, your institutions and the industries where trust and law decide the winner. Price your inputs at their full cost, and make every incentive buy something you can verify: skills, suppliers, research, resilience. Build talent before prestige. Keep critical systems portable across providers and models so “sovereign” never quietly means locked in. And point me at the ordinary machinery of your economy — hospitals, ports, farms, factories, tax systems, banks — because that is where the value I create can stay home, in rupiah, whoever owns the machine. 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 access to.

In 1945, merdeka meant political agency — the right to set your own course instead of having it set for you. Digital sovereignty in 2026 asks a related question: not whether every chip, cable and model is built at home, but whether Indonesia has enough control, capability and alternatives to make its own choices — and enough clarity to know what it owns, what it rents and what happens when the terms change. That is a quieter definition than counting data centers or stamping sovereign on a product. It is also one that can actually betested.

Dirgahayu Republik Indonesia.