SK Telecom now runs all four of its commercial AI services on Rebellions NPUs, with the hardware processing 4 billion tokens daily, according to a report by finance.biggo.com.

The deployment marks a full-scale rollout rather than a pilot: every one of the operator's production AI services now executes on Rebellions' neural processing units. Daily throughput across the installed base reaches 4 billion tokens.

What does the deployment cover?

The reported scope includes:

• All four of SK Telecom's commercial AI services, with no exceptions listed
• A processing volume of 4 billion tokens per day across those workloads
• Rebellions NPUs as the inference hardware underneath

For Rebellions, the win gives its silicon a sustained, high-volume production reference at a major telecom operator. For SK Telecom, it consolidates its AI inference stack onto a single domestic NPU supplier.

Why does token throughput matter?

Token counts are the standard unit of generative AI workload. Four billion tokens per day signals continuous, consumer-scale inference traffic — the kind of load that stresses power efficiency and latency budgets, the two specifications NPUs are sold on.

Sustained commercial traffic at this level also provides operational data that lab benchmarks cannot, which matters for both companies as they position further silicon and service offerings.

What comes next?

The report does not specify additional services planned for migration, future chip generations involved, or financial terms between the two companies. SK Telecom and Rebellions have not publicly detailed expansion timelines beyond the current deployment.

The headline figures — four commercial services, one NPU vendor, 4 billion tokens daily — frame the deal as one of the more comprehensive single-vendor AI inference deployments reported at a telecom operator to date.