On 1 October 2026, Singtel Digital InfraCo's sovereign AI cloud unit RE:AI launched AI Token-as-a-Service (TaaS), the first managed service in Singapore to deliver AI models-as-a-service for enterprises running agentic AI workloads.

The token-based offering gives customers access to AI models, scalable infrastructure, orchestration and governance capabilities through a flexible subscription tied to token volumes. Enterprises allocate tokens once and consume them across internal queries, software development, content creation, voice and video services, and agentic workflows where AI agents automate market research or document review.

What problem is TaaS solving?

RE:AI's pitch centres on two pressures hitting enterprise AI budgets: rising token costs as agentic AI multiplies inference calls across multiple models and tasks, and the data sovereignty concerns that arise when sensitive workloads run on AI models hosted overseas. RE:AI's response is to host models on sovereign infrastructure inside Singapore, under the same regulatory jurisdiction as the customer.

The structure also addresses model management overhead. Enterprises no longer need to procure, integrate and operate each model individually — they consume tokens against a curated catalogue and let RE:AI handle the underlying orchestration.

What sits underneath the service?

TaaS runs on Singtel's RE:AI sovereign AI cloud and the group's integrated technology stack: AI data centres, terrestrial and subsea connectivity, and the Paragon orchestration platform. Paragon routes workloads across compute, storage and network resources according to policy and cost rules set by the customer.

Which models does TaaS support?

The catalogue combines leading open-weight models with selected frontier options. RE:AI's launch materials list Qwen, Mistral, GLM and Kimi among the supported open-weight models, alongside other open-weight alternatives and selected frontier models reserved for workloads with stricter capability requirements. An intelligent routing feature automates model selection per workload, weighing cost against performance.

Enterprises can also onboard proprietary models and route them through the same governance and orchestration layer. That capability matters for organisations that have already invested in fine-tuned models but lack the infrastructure to serve them at scale.

How does pricing work?

Customers commit to a token volume and draw it down across applications as workloads run. The model lets enterprises route premium frontier models only when the task demands them and default to cheaper open-weight alternatives for routine work — internal queries, code suggestions, content drafts. RE:AI frames the design less as a discount mechanism and more as a budgeting tool: finance teams get a single consumption contract instead of per-model licensing sprawl.

What does the launch signal?

RE:AI's first-to-market claim in Singapore reflects Singtel Digital InfraCo's strategy of converting carrier assets — data centres, subsea cables and regional connectivity — into recurring cloud revenue rather than pure connectivity sales.

What to watch next

RE:AI has not disclosed customer commitments, token pricing tiers or capacity targets attached to the launch. The next indicators will be enterprise case studies, particularly from sectors where cross-border data rules shape architecture choices, and any disclosure of throughput or inference volumes handled by TaaS in its first quarters of operation.