Artificial Intelligence in Indonesia

Indonesia’s AI agenda is no longer just a forward-looking strategy document. By March 18, 2026 it includes an existing 2045 frame, a draft 2026-2029 roadmap, an ethics circular already in force, a still-pending ethics Presidential Regulation, sovereign-infrastructure announcements, and a flagship local-language model project.^{[1][8][15][16][23]}

The 2020-2045 strategy still matters because it means Indonesia did not arrive late to the question of why AI matters nationally. What changed in 2025 and early 2026 is the move from ambition to operating design: which instruments are binding, which ones are still draft, which institutions own the work, and where public money and capacity are supposed to sit.^[1][15][23]

That is why the AI Policy Dialogue Country Report with the UK is useful. It did not treat AI as one giant generic opportunity. It broke the issue down into sectors such as e-commerce, banking and finance, the creative economy, health, education, and sustainability, which is a more realistic way to think about Indonesian deployment.^[22]

Editorially, the best way to read Indonesia now is as a state trying to become an AI-producing jurisdiction without pretending every layer is finished. The ambition is real, but so is the amount of design work still happening in public.^{[5][8][15][23]}

The main governance mistake readers can make is to assume every frequently cited document is already operative law. In reality, Indonesia’s AI policy stack now mixes a live strategy, a live ethics circular, consultation-stage documents, and Presidential Regulations that were still being discussed in February 2026.^{[1][7][15][16][23]}

UNESCO’s RAM work is especially useful here because it does not just celebrate momentum. It records the structural issues: underfunded research, uneven digital access, a thin advanced-skill base, and the need for a stronger institutional framework to coordinate AI nationally.^[2][14]

The JDIH record then shows how Indonesia is responding. On February 11, 2026, officials said the draft Presidential Regulation on the National AI Roadmap 2026-2029 was being discussed across more than 30 ministries and agencies. One week later, JDIH said the draft Presidential Regulation on AI ethics was also still under cross-ministry discussion.^[15][23]

That means the right reading is neither “Indonesia has no rules” nor “everything is settled.” The country has a live ethics circular and a growing governance architecture, but the higher-order legal framework was still visibly under construction at the time of review.^[15][16][23]

• UNESCO’s observatory page recommends updating standards, investing in ethical AI, strengthening sandboxes, and creating a national AI agency.^[14]
• The government’s own legal pipeline now openly includes both a roadmap regulation and an ethics regulation, which is more concrete than broad statements about “responsible AI.”^[15][23]

The infrastructure story is now one of the clearest pieces of Indonesia’s AI agenda. Oracle’s planned public-cloud region, the AI Center of Excellence, and the sovereign Security Operations Center described by Cisco all point to a state that wants more of the stack inside Indonesian institutional control.^[3][5][9]

Komdigi’s May 2025 Oracle announcement was unusually direct. It framed the planned GPU-heavy public-cloud region as a route to national AI competitiveness and as evidence that Indonesia wanted to be more than an end market for foreign services.^[3]

The July 2025 AI Center of Excellence launch then made the implementation logic clearer. Komdigi described the AI CoE as the execution arm for the roadmap, while Cisco said the initiative would combine sovereign AI capabilities, a sovereign SOC, local data controls, and large-scale workforce development.^[5][9]

This matters because infrastructure changes who captures value. If public agencies, enterprises, and developers can run more of their sensitive workloads locally, Indonesia has a better chance of building an ecosystem that compounds domestically instead of importing most of the higher-value layers.^[3][5][9]

Sahabat-AI is currently Indonesia’s clearest native model story because it ties together local language, local hosting, open access, public-sector suitability, and wide consumer distribution.^[4][8]

The June 2, 2025 release answered the practical questions readers usually ask. The 70-billion-parameter model is available through a public chat service on sahabat-ai.com and through the GoPay app. It runs across Bahasa Indonesia, Javanese, Sundanese, Balinese, and Bataknese, and it is hosted on locally operated infrastructure designed around Indonesian data-residency expectations.^[8]

The partnership structure also became more legible. GoTo said the model is already helping inside its own ecosystem by lowering costs, improving service quality, and deepening user engagement, while Indosat framed GPU Merdeka and Lintasarta’s AI Factory as the technical backbone for training, inference, and scale.^[8]

The ecosystem angle is important too. The project is connected to universities including UI, UGM, ITB, IPB, Udayana, and the University of Sumatera Utara, plus media groups such as Kompas, Republika, Tempo, and Hukumonline. GoTo also said the earlier Sahabat-AI releases had been downloaded more than 35,000 times on Hugging Face.^[8]

What remains unproven is not whether the project exists, but whether it can become routine infrastructure for ministries, enterprises, and provincial institutions rather than staying concentrated in flagship channels and a few ecosystem leaders.^[8]

• Distribution is now concrete: Sahabat-AI is on the open web and inside a mass-market payments app.^[8]
• The language layer is concrete too: five local languages are publicly named in the latest release.^[8]
• The project is open enough to matter for builders, with public downloads and a stated audience that ranges from startups and university labs to public-service institutions.^[8]

Indonesia’s talent challenge now has a clearer shape: the country needs a larger specialist base, a more AI-ready bureaucracy, and more region-specific capability building outside the main metropolitan centers.^{[10][11][17][18][20][21]}

At the national level, Komdigi continues to talk in terms of the 9-million digital-talent need by 2030, but the newer announcements are more concrete. AI Talent Factory is framed as a staged pipeline, while February 2026 updates showed 302 national internship participants being embedded into ministry work on AI and cybersecurity rather than being kept in ceremonial roles.^[10][21]

UNESCO’s work with KemenkoPMK adds the public-sector side of the picture. The December 2025 discussion on AI literacy for civil servants, following the Banda Aceh pilot, makes it clear that ministries are worried about bureaucratic readiness, not just private-sector employability.^[11][18]

The geographic detail is important. UNESCO’s Aceh work highlighted urban-rural gaps in competency and access, while a January 2026 BPSDM Komdigi update said regional ecosystem needs differed by place, naming tourism in Bali, startup ecosystems, and agriculture as examples. That is a more realistic picture of Indonesia than a single national readiness average.^[17][18][20]

By March 18, 2026 Indonesia can point to more than a strategy document. It can point to a local model, a sovereign-infrastructure buildout, ministry-level legal drafting, civil-service AI literacy work, and a public narrative around GovTech AI. The next step is a denser map of deployments.^{[8][9][11][12][15][23]}

The public-service layer is especially important because it gives AI a national-development justification. GovTech AI is being sold as a route to faster and cheaper public services, while the health discussion has already moved to sandboxing and ethics review before broad rollout. That is a sign of seriousness: the state is thinking in terms of use cases and safeguards, not only catchphrases.^[12][19]

The upside for Indonesia is real. The country has the scale, linguistic need, and political motivation to justify local infrastructure and local models. The risk is that the legal pipeline, ministry capacity, sector pilots, and provincial readiness all move at different speeds.^{[8][14][15][23]}