Nokia targets Korea with new AI-powered "Doksuri" lineup

Nokia has named its AI-powered base station Doksuri and said it will market its next-generation equipment under the Korean Eagle brand, targeting South Korea as its first major market ahead of MWC26.

Nokia said the ReefShark SoC-based Korean Eagle lineup offers about 30% better power efficiency and up to 25% lower weight in typical network environments.

The Korean Eagle uses a new mounting system to reduce installation time and that it is working to enhance RAN software with NVIDIA GPU accelerators in tests including T-Mobile, SoftBank, and Indosat.

NVIDIA has made a surprise US$1 billion investment in the company, which is seen as driving AI transformation in telecom networks.

🔗 Source: ZDNet

🧠 Food for thought

Implications, context, and why it matters.

The two-part strategy behind Nokia’s AI network push

The hardware efficiency pitch goes after the telecom sector’s biggest operating drain since radio access networks (RAN) can use nearly 70% of a carrier’s total energy consumption 1.
The “AI for RAN” plan applies artificial intelligence to automate RAN work and lower spend, matching operator plans to automate over 50% of RAN processes by 2027 1.
The SoftBank trial backs the “AI on RAN” effort that aims to make the RAN a distributed computing platform, using spare AI-RAN capacity for third-party AI tasks 2.
That setup could let operators rent unused network compute for outside AI workloads, opening new revenue beyond connectivity 3.

AI-RAN faces a cautious market and the ghost of Open RAN

Nokia and Nvidia still face doubt from possible buyers despite the announcements 4.
By late 2025, leaders at Verizon and Orange questioned whether graphics processing units (GPUs) made sense for RAN software given cost and power use, pointing instead to custom chips or newer central processing units (CPUs) 4.
The push comes after Open RAN (open radio access networks) lost momentum, after it fell short on promised cost savings 5.
To avoid the same outcome, Nokia needs to convince operators that revenue from selling spare compute will beat the high upfront spend and the added complexity 4.