Google Cloud brings on-premise Gemini AI to Singapore

Google Cloud is bringing its “Gemini everywhere” vision to Singapore with the launch of its most powerful generative artificial intelligence (AI) models on-premise and guaranteeing data residency for its cloud-based AI services in a bid to support the nation’s AI ambitions.

Announced at its AI conference in the city-state last week, the move lets organisations with strict data sovereignty and security requirements run Gemini models in their own datacentres for the first time through Google Distributed Cloud (GDC) air-gapped, a fully managed offering that’s completely disconnected from the public internet.

This directly addresses the needs of government agencies and regulated industries that handle highly sensitive data. Singapore’s Government Technology Agency (GovTech), the Centre for Strategic Infocomm Technologies (CSIT), and the Home Team Science and Technology Agency (HTX) will be among the first worldwide to use the technology.

Speaking at a media briefing, Goh Wei Boon, chief executive of GovTech, stressed the importance of data sovereignty. “We care a lot about where the data goes when we deliver digital services,” he said.

“We do not want to send our data to services all over the world; we hope for it to stay in Singapore. Bringing Gemini to the GDC will allow us to be able to do processing for highly sensitive systems in a specific datacentre within our control.”

GovTech is also planning to experiment with Google’s Agentspace platform for public officers in a dedicated AI agents sandbox that will be developed with Google, Cyber Security Agency of Singapore and the Infocomm Media Development Authority.

“We’re going to start with a small number of Agentspace licences, put in place the guardrails, experiment in the AI sandbox, and be very sure that it’s safe before we will scale it to the public,” said Goh.

We do not want to send our data to services all over the world; we hope for it to stay in Singapore. Bringing Gemini to the GDC will allow us to be able to do processing for highly sensitive systems in a specific datacentre within our control

Alongside the on-premise Gemini offering, Google Cloud has also expanded its data residency guarantees in its Singapore public cloud region. This allows organisations to not only store data at-rest, but also perform all machine learning processing for services such as Gemini 2.5 Flash and Vertex AI Search entirely in Singapore, a key enabler for GovTech’s use of AI on its Government on Commercial Cloud (GCC) platform to develop confidential cloud-eligible applications.

In the financial sector, DBS Bank, a long-time user of AI, sees the capabilities as a way to further industrialise its AI efforts. Nimish Panchmatia, chief data and transformation officer at DBS Bank, noted the significant business impact of the bank’s AI initiatives, which are expected to generate over S$1bn in economic value this year.

Reflecting on the DBS’s AI journey, Panchmatia noted that the bank has developed over 350 use cases and 1,500 models, and takes a measured approach to AI safety, employing a human-in-the-loop model and even using one large language model to judge the output of another model to prevent hallucinations. Currently, about 30% of the bank’s traditional AI and machine learning workloads run on Google Cloud.

In the retail sector, FairPrice Group has partnered with Google Cloud to launch its Store of Tomorrow concept, deploying a suite of AI agents to transform the shopping experience. For example, at the new FairPrice Finest outlet in Punggol Digital District, smart shopping carts can generate shopping lists, highlight offers, navigate through the store and enable seamless payment. The store also uses Google’s Vision AI to alert staff when shelves need restocking.

Thomas Kurian, CEO of Google Cloud, noted Singapore’s position as a hub for regional AI innovation. “Singapore is the launchpad for AI in Asia, and today we’re making Google’s most capable models and AI tools available here,” he said. “By bringing Gemini everywhere – in the public cloud with data residency guarantees, in on-premise or disconnected environments, and at the edge – we are unleashing a wave of innovation for both public sector organisations and businesses across industries.”

Addressing space constraints in Singapore that could hamper Google’s growth in the city-state, Kurian noted that in most countries, it’s not space, but power that limits the amount of compute capacity.

“We’ve brought a new tensor processing unit system that’s about 2.2 times more power efficient,” he said. “That allows us to roughly double the capacity that you can serve. Right now, we don’t have an issue and we’re actively looking at broadening our footprint here.”