Samsung, KT test 6G AI-RAN on commercial network

Samsung Electronics and KT have completed a successful field test of 6G AI radio access network (AI-RAN) technology on KT’s commercial network in South Korea.

The test, carried out by Samsung Research and KT’s Future Network Laboratory, involved about 18,000 users in various locations in Seongnam, Gyeonggi Province.

This marks the first successful demonstration of 6G AI-RAN on a live commercial network, after earlier simulation tests in June.

The technology uses AI to automatically optimize network configurations for individual users, and aims to reduce connectivity issues.

Both companies said AI-RAN could play an important role in managing higher data usage expected with future 6G networks.

🔗 Source: Samsung

🧠 Food for thought

Implications, context, and why it matters.

Field test metrics remain undisclosed, leaving readiness unclear

Samsung and KT ran a Seongnam trial on KT’s live network with 18,000 users. They shared no throughput, latency, or call drops. That makes progress hard to judge even with past gains of 8.65% throughput and fewer drops 1.
Focus was AI-based user-level optimization, which tunes radio settings for each device. Samsung earlier hit an 87% detection rate and 92% accuracy for connection failures 1, though Seongnam’s scale results remain unknown.
It may involve 3.5 GHz mid-band spectrum 2 and 32T32R or 64T64R massive Multiple-Input Multiple-Output (MIMO), meaning 32 or 64 transmit/receive antennas 2. 6G work starts around 2026, which points to pre-6G validation 3.

Telecom edge vendors see near-term demand as AI-RAN pilots expand

AI radio access network (AI-RAN) trials are expanding beyond Korea. That drives demand for GPU plus Neural Processing Unit (NPU) edge hardware, RAN Intelligent Controller (RIC), and Machine Learning Operations (MLOps) 4. Qualcomm, Intel, and NVIDIA could gain 5.
Integrators plus test firms gain as NTT DOCOMO, KDDI, Verizon, and SoftBank move to trials 5. Orange raised concerns about GPU RAN and backed distributed computing infrastructure, which creates demand for hybrid CPU plus GPU setups 6.
Energy-efficient edge gear has an advantage, since operators question GPU power use 7. NVIDIA GPUs draw about 300 watts versus about 40 watts for alternatives 7. That opens room for lower power AI accelerators that deliver AI-RAN gains without steep energy costs.