NVIDIA pushes sovereign AI deeper into Southeast Asia

NVIDIA used its AI Day Singapore event, held at the Raffles City Convention Centre on 22 and 23 September, to argue that Southeast Asian governments are moving AI out of pilot projects and into production. The company frames the shift as national strategy rather than procurement: it says it is working to enable all nations to be AI nations, and that keeping that transition local needs open models, developer tools and a partner ecosystem as much as it needs accelerators.

Public safety and public services first

Singapore’s HTX, the Home Team Science and Technology Agency, is researching NVIDIA’s Nemotron 3 Super and Nemotron 3 Nano Omni models for public safety work. NVIDIA says the larger of the two suits complex reasoning and agentic workloads, while the Omni model’s combined vision, audio and language handling could underpin multimodal tools built on the agency’s own data.

NCS is pairing Nemotron models with NVIDIA’s video search and summarisation blueprint to push agentic AI into enterprises and the public sector, and is also working on physical AI for humanoid robotics. ST Engineering is building its AI stack on NVIDIA’s NeMo tooling and its cuOpt optimisation software.

Region-specific models rather than imports

Localisation is the clearest theme of the event. Malaysia’s YTL AI Labs is fine-tuning Nemotron models for enterprise and citizen services, while Vietnam’s Viettel AI has been fine-tuning Nemotron 3 Super for Vietnamese. NVIDIA says that work topped both the VMLU benchmark and Viettel’s own product benchmark, and that the model is set for adoption in Legal AI, an agent harness that will serve the group’s own staff as well as outside customers.

In Thailand, the Big Data Institute and iApp Technology, both part of the ThaiLLM Collaboration, are adapting Nemotron 3 Nano by fine-tuning OpenThai 2.0 Legal with Thai legal data through NeMo. The result is released as open source for Thai developers, and it already powers Thanoy, iApp’s legal assistant, which NVIDIA says has around 43,000 users. FPT Smart Cloud co-developed Nemotron-Personas-Vietnam, an open dataset grounded in Vietnamese demographic and cultural data, and Brunei’s Antrique built a platform aimed at productivity in the country’s food sector.

AI Singapore is expanding its SEA-LION model family to include Nemotron models and NeMo tools, and Hummingbird Bioscience with LynxKite is building a toxicity knowledge graph on Nemotron 3.5 Lightning and NeMo Retriever. Cosmos world models and the video search blueprint are doing quieter work: Malaysia’s ITMAX is using them to improve city traffic operations, and Thailand’s AS-TECH is applying the same stack to passenger flow at airports.

Sea becomes the region’s first Vera Rubin customer

Sea Limited, the Singapore-headquartered group behind Shopee, Garena and Monee, is the first enterprise in ASEAN to adopt NVIDIA’s Vera Rubin platform. NVIDIA describes the move in terms of reach rather than benchmark figures: Shopee already uses AI to cut the effort of writing product listings and to improve discovery, Monee applies it to fraud detection and credit risk assessment, and Garena uses it in its games. Sea says the platform will let it run larger models and more agents across services that reach hundreds of millions of users.

Our opinion

Sovereign AI is an awkward phrase for a business that sells the same accelerators to everyone, but the Singapore line-up shows what it means in practice. Viettel fine-tuning a model for Vietnamese and iApp releasing a Thai legal model as open source are not declarations of independence from American silicon. They are attempts to stop depending on a handful of providers for the language layer underneath public services, which is where the political sensitivity actually sits. The detail worth watching is Nemotron’s role as the common base: when a dozen organisations in five countries fine-tune the same open family, they get portability and NVIDIA gets a moat built out of other people’s local data. Sea’s Vera Rubin adoption matters for the same reason. A regional champion signing up as the first ASEAN customer is a reference NVIDIA can take to every ministry in the neighbourhood, and reference customers, not benchmarks, are what win national AI contracts. The claims here are NVIDIA’s own and mostly unquantified: no throughput figures, no deployment dates and no named government buyers beyond the agencies it names. Treat the direction as real and the scale as still to be proven.