On April 27, 2026, RE:AI, the sovereign AI cloud business of Singtel Digital InfraCo, signed a strategic partnership with Mistral AI to build AI services for financial services, defense, government shared services and healthcare in Singapore, run on Singtel’s own GPUs and data centers instead of on servers overseas. Bill Chang, chief executive of Singtel Digital InfraCo, signed the agreement with Mistral AI chief executive and co-founder Arthur Mensch at the Milipol TechX 2026 event at Singapore’s Sands Expo and Convention Centre, The Business Times reported. According to Singtel’s announcement, the deal pairs Mistral’s open-source software stack with RE:AI’s GPU platform and sets up a joint engineering hub for testing industry use cases.
Chang summed up the pitch in one line to The Business Times: “We are looking to bring AI to the data and not data to AI in other countries, especially for sovereign customers’ needs with data governance that is required to be here.” That sentence is the whole product. Singtel is not selling a better model; it is selling a guarantee about where the model runs and where the data stays.
What Singtel and Mistral agreed to
Everything below comes from Singtel’s April 27, 2026 release:
- Shared infrastructure: Mistral’s open-source stack, including training capabilities, foundation models and agentic workflows, will be integrated with RE:AI’s GPUs to offer multi-tenant GPU-as-a-Service and AI-as-a-Service for essential industries and public sector deployments.
- Where it runs: customers will deploy from the upcoming AI-ready data centers of Nxera, Singtel’s data center arm, in Singapore, Johor and Batam, which are linked by Singtel’s subsea cables.
- Mistral as a customer too: Mistral will use RE:AI as cloud infrastructure to roll out Mistral Compute offerings for its internal use and for joint go-to-market work.
- A joint engineering hub: the two companies will set up an Applied AI Centre of Excellence where Singtel, Mistral and partners build secure, compliant AI environments inside Singtel’s infrastructure.
- Singtel tests it on itself first: Singtel Singapore is the center’s pilot customer, trialing customer care chatbots, network operations copilots and an AI-assisted tool for evaluating network tenders. Government agencies and enterprise partners are to follow as early adopters.
The deal also plugs into work Singtel had already announced. On February 24, 2026, Singtel Digital InfraCo launched a Centre of Excellence for Applied AI with NVIDIA, and under the April agreement Mistral will contribute resources to that center’s projects where applicable. On March 12, 2026, RE:AI partnered with Cohesity on a service that turns years of backup files into a searchable knowledge base while keeping that information in Singapore.
Sovereign AI cloud versus public cloud: where the line really sits
RE:AI is not new. Singtel launched it on October 10, 2024 as an AI cloud service combining GPUs, storage and AI tools with Singtel’s networks, delivered through its Paragon orchestration platform. What has changed since is the emphasis. Singtel now describes the whole stack, from data centers to subsea and terrestrial connectivity to orchestration, as sovereign infrastructure that keeps customer data inside Singapore.
That is the core of the “AI-ready infrastructure versus public cloud” question, and the honest answer is narrower than the marketing on either side:
- Singtel’s case: its October 1, 2026 release names “concerns about sending sensitive business data to AI models hosted overseas” as a barrier, and says models hosted on RE:AI keep data within Singapore to help customers meet data sovereignty and regulatory requirements.
- Singtel’s own caveat: the same release says that for less sensitive workloads, customers can also use non-sovereign AI clouds to optimize cost and performance. Even the sovereign provider is not arguing that everything belongs on a sovereign cloud.
- The public cloud’s answer: hyperscalers are competing on the same ground. On August 28, 2025, Google Cloud said its Singapore cloud region now offers expanded data residency guarantees, so customers can store data and run machine learning processing for AI workloads entirely within Singapore. It also made Gemini generally available on Google Distributed Cloud air-gapped, an on-premises setup disconnected from the public internet, and said Singapore agencies CSIT, GovTech Singapore and HTX would be among the first worldwide to get it.
So “public cloud” no longer automatically means “data leaves the country.” The real questions for a buyer are more specific: in which country is the data stored and processed, who operates the hardware, and which workloads are sensitive enough to need the strictest option.
Did Singtel and Google Cloud launch a sovereign cloud together?
Not that we could find. The two companies do work together: Singtel lists Google Cloud as a key cloud partner and helps business customers build on Google Cloud Platform. But as of September 30, 2026, we found no announcement from either company of a joint sovereign cloud offering in Singapore. The RE:AI partnership releases Singtel published in 2026 that we reviewed involve Mistral AI, NVIDIA, Cohesity and WEKA, and Google Cloud’s August 2025 Singapore sovereign AI announcement names government agencies and companies such as DBS Bank, with no mention of Singtel. In other words, Singtel helps customers use Google’s public cloud while competing with Google on sovereign AI.
What has happened since the Mistral deal
- June 10, 2026: RE:AI signed a memorandum of understanding with WEKA to build sovereign AI infrastructure across Singapore and the wider ASEAN region, using WEKA’s NeuralMesh storage system to keep GPUs supplied with data.
- October 1, 2026: RE:AI launched AI Token-as-a-Service, which Singtel calls a first-in-Singapore managed service. Customers buy token volumes by subscription and use them across a catalog of models that includes open-weight models such as Qwen, Mistral, GLM, Kimi and MiniMax, with selected frontier models when required. The service routes each job to a suitable model and gives token-level visibility for governance.
Who is actually on notice
Where we land: the pressure from this deal falls less on hyperscalers, which are already offering in-country options, and more on regional firms in regulated sectors that have been treating “we cannot send this data abroad” as a reason to wait. A local provider with a named model partner, a test center and published use cases removes that excuse. Banks, hospitals and agencies in Singapore now have to decide which of their AI work is truly sensitive, because the answer determines which cloud they pay for.
The workforce side is worth watching too. Chang told The Business Times that Singtel is “massively retooling and reskilling all of our employees, especially in Singapore,” and the first pilots at the Mistral center are tools for Singtel’s own customer care, network operations and network tender teams. The fair test of those pilots is whether those teams handle more and better work, not whether they shrink.
The lesson travels well beyond Singapore. A small business will never buy GPU capacity from Singtel, but it faces the same question at a smaller scale every time it pastes customer information into an AI tool. Our guide to what happens to your data when you use an AI tool covers the plain-language version, and Singtel’s own split, sensitive work on the strictest option and everything else on the cheaper one, is a sensible default at any size. Singtel’s new service also bills by the token, the same usage-based pricing that sits behind many AI tools’ API plans; our breakdown of per-seat versus usage-based AI pricing explains how to read them before the bill arrives.
What the announcements leave out
Three things a buyer would want are not in any of the releases cited here. None publishes a price for RE:AI’s GPU or token services. None names an outside customer of the Singtel and Mistral center beyond Singtel Singapore itself. And none gives an opening date for the center or for the Johor and Batam data centers it will draw on. Until those appear, the partnership is a credible plan with a working platform underneath it, not yet a track record.
Questions about Singtel’s sovereign AI push
What is Singtel RE:AI?
RE:AI is the sovereign AI cloud business of Singtel Digital InfraCo. Singtel launched it on October 10, 2024 as an AI cloud service, and it now offers GPU-as-a-Service and, since October 1, 2026, AI Token-as-a-Service, hosted on Singtel infrastructure so that customer data remains within Singapore. (Sources: Singtel RE:AI launch release, Singtel Token-as-a-Service release)
What does the Singtel and Mistral AI partnership include?
Signed April 27, 2026, it combines Mistral AI’s open-source stack with RE:AI’s GPUs to offer GPU-as-a-Service and AI-as-a-Service for financial services, defense, government shared services and healthcare, and sets up a joint Applied AI Centre of Excellence with Singtel Singapore as its pilot customer. (Source: Singtel and Mistral AI release)
Are Singtel and Google Cloud launching a sovereign cloud in Singapore?
We found no such announcement as of September 30, 2026. Singtel lists Google Cloud as a key cloud partner for business customers, but its 2026 RE:AI sovereign AI releases name Mistral AI, NVIDIA, Cohesity and WEKA as partners, and Google Cloud announced its own Singapore sovereign AI options, including Gemini on Google Distributed Cloud air-gapped, without Singtel. (Sources: Singtel Google Cloud partner page, Google Cloud Singapore announcement)
