Building local LLMs helps in preservation of cultural heritage: Indosat’s CEO

Indonesian telecom giant Indosat Ooredoo Hutchison (Indosat) is a 57-year-old telecommunications brand in Indonesia that serves over 100 million customers. In 2022, Indosat Ooredoo and Hutchison 3 Indonesia merged to create the present-day company. In the last six months, the company has embarked on an Artificial Intelligence (AI) as the Northstar strategy, which has set its sights on leveraging this technology to transform itself and contribute to broader national and regional development.

Vikram Sinha, the chief executive officer of Indosat in a recent interview with TechCircle spoke about the company’s AI strategy, important partnerships, and tech sovereignty.

AI as the Northstar

Indosat’s AI strategy has three pillars. The first pillar is transforming Indosat into an AI-native telco by collaborating with partners like Google and Mckinsey. Sinha said that this initiative has already rendered some initial wins. So far, the company has deployed around 14 AI use cases, with a target of scaling up to nearly 200.

“We invest heavily in capital expenditure for our network. Previously, with machine learning, our prediction accuracy was around 70-72%. Now, it has increased to 88%, enabling us to achieve savings of approximately $25 million in just the past six months.” Additionally, the company has been able to improve churn rate and Average revenue per unit (ARPU).

The second pillar of the AI North Star strategy focuses on becoming an AI-native tech group, emphasising advancements in AI, cloud technologies, and security, particularly in India. 

The third pillar emphasises Indonesia's evolution from being a technology consumer to becoming an enabler. The goal is to position Indonesia as a secure and innovative technology hub, drawing lessons from India's journey, which has demonstrated significant growth and transformation at scale.

“Since announcing our AI initiative in August, we have made significant progress by focusing on building the right data foundation, unified data platform, and architecture. By November, we had fully established our architecture and unified data platform on Google Cloud Vertex,” said Sinha. 

To support our AI initiatives, Indosat launched our own large language model, Sahabat AI in partnership with Software-as-a-Service (SaaS) firm GoTo, tailored specifically for the Bahasa language, apart launching its own sovereign AI cloud.

“While there's been considerable debate about whether companies should build their own Large Language Models (LLMs), we strongly believe that it's crucial to preserve a country's culture and heritage. By developing a local language model, we ensure that we can accurately represent and preserve the history, traditions, and nuances of the language while enabling deeper engagement with our audience,” said Sinha.

Notably, many Indian tech leaders including Infosys’ co-founder Nandan Nilekani have said in the past that India needn’t waste resources on building their own LLMs.

AI investment

Indosat launched “AI Factory” under its subsidiary, Lintastarta, equipped with GPU computing capabilities for efficient data training. In August 2024, Lintastarta announced GPU Merdeka, a “GPU-as-a-service” platform powered by Nvidia. Indosat has already made an investment of $300 million towards the AI factory and for procuring high-performant GPUs from Nvidia.

“Currently, we are in the process of integrating Blackwell GPUs, which is expected to go live by June or July. Previously, our investment focused on H100 GPUs, forming the backbone of our AI Factory operations.”

Additionally, on the AI-native telco front, last year Indosat allocated around $550 million to network capex, he said. “The strategic adoption of AI is now driving productivity gains and enabling us to allocate funds more efficiently. Historically, our telco capex ranged between $600 to $800 million, but AI-driven optimisations are helping us better manage investments and reinvest in growth areas.”

Partnership with AIonOS

Sinha’s visit to India also coincided with the signing of a memorandum of understanding with former Tech Mahindra head CP Gurnani-led AIonOS. This MoU aims to advance Indonesia’s AI landscape as part of a government-to-government (G2G) initiative between India and Indonesia. 

To be sure, AIonOS is a joint venture primarily backed by InterGlobe Enterprises, along with the Assago Group chaired by Gurnani. The MoU builds on the partnership between the two companies. Last year in June, shortly after the formation of AIonOS, the companies announced a partnership to offer a range of solutions, including Sovereign AI, which is a comprehensive suite of AI solutions on a resilient digital infrastructure. 

Under the partnership signed this week, the firms will also set up a Centre of Excellence in Jakarta, with Chinese AI firm Deepseek as a partner. recently gained significant attention for developing advanced AI models at a fraction of the cost of its competitors. In January, DeepSeek introduced its AI model, DeepSeek-R1, which demonstrated capabilities comparable to leading models like OpenAI's ChatGPT, despite being developed with an investment of less than $6 million. This efficiency has disrupted the AI landscape, leading to significant market reactions, including a notable drop in Nvidia's stock value.