The development builds on Sahabat-AI, an Indonesian large language model launched in mid-November. Indosat led the Sahabat-AI project, which is part of a drive to preserve local culture and dialects amid a global surge in AI development, currently led by models trained in the U.S. and China.

This latest model operates in Bahasa Indonesia, the country’s national language, as well as five other local languages including Javanese, Balinese and Bataknese. These languages can boast tens of millions of speakers: Approximately 70 to 80 million people speak Javanese, roughly equal to the number of people that speak Italian. People across the Indonesian archipelago of some 17,000 islands may often converse in local dialects rather than Bahasa Indonesia. 

“Indosat is proud to lead the development of Indonesia’s sovereign AI capabilities. Sahabat-AI is more than a model—it’s a national asset powered by collaboration and built for all Indonesians,” Indosat CEO Vikram Sinha said in a statement

The company added that all data and GPU infrastructure used to serve the model are stored within Indonesia territory or on users’ own servers.

Indosat’s subsidiary, Lintasarta, launched GPU Merdeka, a “GPU-as-a-service” platform powered by Nvidia’s H100 GPUs in late August. 

‘Telco to techco’

Indosat is one of Indonesia’s leading telecoms companies, serving nearly 100 million consumers across the country. Yet the firm is pursuing a “telco to techco” strategy, leveraging its broad consumer base to start serving other internet-based services. (Other Southeast Asian telecoms companies, like the Philippines’ Globe Telecom, are also foraying into new businesses like fintech).

In an interview with Fortune in mid-May, Sinha revealed that Indosat is already providing AI services to more than 20 Indonesian firms, including banks and commodities companies.

Still, the company hasn’t revealed how much it has invested in Sahabat so far. 

The firm is launching its consumer-focused Gen AI application right as it’s getting squeezed by rising living costs. The majority of mobile subscriptions in Indonesia are pre-paid, so customers can easily scale back their spending as budgets tighten.

Revenue for the three months ending March 2025 came in at 13.58 trillion rupiah ($833.3 million), down 1.8% from the same period a year ago.

Yet a local language LLM could help Indosat reach rural Indonesians that don’t speak English or even Bahasa Indonesia. Getting more Indonesians hooked on AI services will increase data use, and encourage them to top up their mobile spending a little more. 

Sinha said the company wanted to “support and solve their problems” by offering an AI search tool. 

There are also monetization opportunities down the line. Data collected from queries, especially in the several dialects, will help Indosat fine-tune its AI model and the company could eventually market its “Gen AI-as-a-service” applications to business like call centers. 

Sinha added that that could be a more premium AI search function that could require a paid subscription as well.