Geostrategic Media Political Commentary, Analysis, Security, Defense
Raditio Ghifiardi
In early June 2026, aboard Air Force One, President Donald Trump confirmed to reporters that the White House was discussing the possibility of taking an equity stake in OpenAI. A few days earlier, Senator Bernie Sanders had introduced a bill proposing the federal government acquire 50 percent of leading AI companies — with voting rights and board seats. Economists pushed back, opinion columns multiplied. What few noticed: Indonesia is quietly running an experiment far larger than owning shares in a single AI firm.
Through Danantara — a state-owned-enterprise super-holding managing assets worth nearly US$1 trillion, roughly 71 percent of Indonesian GDP — the state is openly steering capital into the backbone of AI: data centres, nickel and copper down streaming for semiconductors, and AI models themselves. Add Sahabat-AI, the national language model Indosat developed by fine-tuning Llama 3 for 277 million Bahasa Indonesia speakers. The result is a concrete portrait: a country building AI sovereignty not through speeches but through its balance sheet.
For the developing world this sounds like the long-awaited answer — no longer just users, but owners. Before we applaud, however, there is an old story worth remembering. Ironically, it was written by countries like ours.
The Irony: America Is Learning What We Already Knew
Look at who is criticising the U.S. proposal most loudly. Not radical free-marketeers. Adam Posen of the Peterson Institute called it “a step toward state capitalism.” Nat Purser of Public Knowledge asks the simple question: what happens when a government becomes reluctant to enforce safety rules because doing so could erode the value of its own investment?
This question is hardly new. The OECD and World Bank long ago reached the same conclusion: state-ownership policies work only when the roles of policy maker, regulator, and owner are clearly separated. And where did they draw that lesson? From us.
PDVSA, Venezuela’s oil giant, collapsed long before U.S. sanctions — wrecked by political meddling, mismanagement, and debt. Mexico’s Pemex spent decades trapped in chronic politicisation. Both were living laboratories in what happens when the state becomes owner and referee at once.
Now consider the irony. America is rediscovering a lesson the developing world paid for in cash. And Indonesia, in its drive to build AI sovereignty, risks repeating the same mistake — this time at the most consequential technology layer of the century.
Danantara: Owner and Referee Under One Roof
Picture a football match where the owner of one club also serves as the referee. You need not accuse anyone of cheating — just ask whether you trust the call. That is Danantara’s structural position today. Formed by fusing the Indonesia Investment Authority with parts of the Ministry of State-Owned Enterprises, it holds SOE assets while exercising functions that once belonged to the sectoral policy maker — exactly the blending of roles the OECD has warned against for years.
Drop AI into the equation. When Danantara becomes the dominant shareholder in national data centres and the semiconductor supply chain, who regulates the sector independently? When a citizen complains about an AI model’s safety or a privacy breach, to whom does she turn — the same entity whose dividends depend on the firm she is accusing? Analysts already describe Danantara as a powerful political actor in a state-directed capitalism model, master shareholder for the SOE estate and, through subsidiaries, gatekeeper of strategic export channels.
The lesson from PDVSA and Pemex is unambiguous: without a wall between ownership and oversight, sovereignty slowly becomes another name for state capture by an entity grown too large to regulate.
Indonesia is not alone in this design choice. The UAE created MGX in 2024 — a US$100 billion AI investment vehicle anchored by sovereign wealth fund Mubadala and AI firm G42 — while the same state shapes domestic AI rules through the Artificial Intelligence and Advanced Technology Council. Saudi Arabia followed in May 2025 with HUMAIN, a PIF-owned company chaired by the Crown Prince that operates the entire AI stack from data centres to Arabic LLMs. India has so far chosen a different architecture: the India AI Mission distributes compute subsidies and grants rather than taking equity, leaving the Ministry of Electronics and IT to set policy at arm’s length from any state-owned operator. The contrast is the point — sovereign AI need not mean sovereign ownership, and where Global South states draw the line determines whether the referee can still call a fair game.
The Forgotten Layer: Those Who Clean Our Data
While we argue about shares and wealth funds, there is another story rarely told. The AI we now use — the one that writes poems, translates contracts, and finishes high-school homework — was not really born inside a machine’s head. It was trained by millions of people around the world who typed, classified, and labelled data, item by item, day after day.
Most of them sit in the Global South.
In Kenya, data annotators reportedly earn under US$2 per hour. In the United States, similar work pays over US$20 per hour. In the Philippines, some annotators are not even paid the regional minimum wage of around US$6 a day. Indonesia — among the world’s largest internet user bases
— is one of the major data and market suppliers in this chain, and that role will deepen as AI penetration grows.
Jaron Lanier, one of Silicon Valley’s earliest insiders and most piercing critics, has a name for this problem: data dignity. The idea is simple — data and annotation labour are forms of work, and work should be paid. Not treated as “exhaust” free for the taking.
But caution is in order. The Centre for International Governance Innovation warns that hastily designed pay-for-data schemes can produce new digital divides and still fail to address citizens’ deeper anxieties about privacy, security, and misuse.
For Indonesia, the implication is clear: “pay annotators fairly” and “make the Global South an owner of AI infrastructure” are two distinct agendas. The first is a labour-standards question. The second is a structural claim about who captures the rent of AI. Indonesia needs both — but must know when to use the hammer and when to use the screwdriver.
Sovereignty Without Separation Equals Capture
So what to do? Not stop Danantara or Sahabat-AI. In principle, both are correct answers to this century’s question. What needs reinforcing is the architecture — on three levels.
First, separate owner from regulator. Norway did this with Statoil. Chile runs Codelco on a similar model. Closer to home, Brazil’s R$23 billion Plano Brasileiro de Inteligência Artificial (PBIA) routes capital through the BNDES development bank while the regulatory track runs separately — through the ANPD data-protection authority and the AI Bill (PL 2338) advancing in the Senate. South Africa’s 2024 National AI Policy Framework goes further on paper, proposing a layered architecture with policy lead at the DCDT, enforcement at the Information Regulator, and infrastructure security at the Cybersecurity Hub — the wall drawn before the concrete is poured. The regulator of data centres and any future AI authority must not sit under the same institutional umbrella as Danantara. This is not bureaucratic excess — it is the precondition for public trust that AI-safety decisions will not be compromised to protect a dividend.
Second, insulate the AI dividend from the political cycle. Alaska’s experience is instructive. It has a statutory formula four decades old — and still fragile. Since 2016, billions the formula would have distributed have been withheld in annual budget tugs-of-war. Mongolia is more dramatic: it promised coal-dividend payouts before elections and ended up delivering only around US$92 per citizen — far below the pledge. If Indonesia wants its AI sovereign fund to be credible, the rules must be locked at statutory level, not left to the mercy of next year’s budget meeting.
Third, pair ownership with targeted transfers. Indonesia has been here before. The original 2005 BLT disappointed — targeting missed, no conditionality to encourage productive behaviour. Then came an innovation many forget Indonesia pioneered: community-based targeting. Village
meetings define local poverty and rank households, and pandemic-era BLT-Dana Desa distributed IDR 600,000 per month to eight million families with lists finalised through village deliberation. If an AI-driven automation wave really hits labour markets, a flat Alaska-style dividend spread evenly across the population will be too thin to matter. What is needed is a Bolsa Família-style conditional model — rooted in the community-targeting infrastructure we already built.
Position First, Distribution Later
The thread is simple. America can debate how to slice the AI pie because it already owns the pie. The more basic question for the Global South: will we own any pie at all?
Indonesia is choosing one right answer — build position first, before the distribution table is set. Brazil is choosing another: route the capital through BNDES, keep the regulator at arm’s length. South Africa is sketching the institutional wall before the building goes up. Each is a different wager on where to draw the line between owner and referee. But any of these right answers can still end badly if the institutions are brittle. Without a wall between owner and regulator, without statutory locks on the political cycle, and without targeting that reaches those genuinely displaced, AI sovereignty risks becoming a slogan as hollow as an AI dividend with no real funding behind it.
The lessons from PDVSA, Pemex, Alaska, Mongolia, and Bolsa Família were not written in Washington. They were written in our part of the world, by countries like ours. It would be a bitter irony if Indonesia — and Brazil, South Africa, and every state now reaching for AI sovereignty — were to forget the very lessons our part of the world wrote.
