Samsung's South Korea AI-RAN Pilots Put Industrial 5G to the Test

Samsung will supply KT's shipyard pilot and lead SK Telecom's refinery pilot from October. The whole government program is worth about 17.2 billion won, so the contracts matter as a reference design, not as revenue.

Samsung has signed contracts with KT and SK Telecom for two AI-RAN projects under South Korea's Hyper AI Network program, taking the role of sole global vendor on KT's project and main vendor on SK Telecom's. Deployments begin in October on 5G standalone private networks at HD Hyundai Samho's Yeongam shipyard and at SK Incheon Petrochem.

The workloads are specific. At the shipyard, KT will trial AI-assisted welding and painting robots, including a quadruped welding robot, and autonomous robots that maintain telecom equipment. At the petrochemical complex, SK Telecom will run autonomous patrol robots in hazardous zones and real-time video analysis that flags hazards. Neither operator nor Samsung disclosed contract values.

The money involved is small by design. The government budgeted about 17.2 billion won, roughly $11.6 million, for the whole program, split between the SK Telecom and KT consortia and covering operators, integrators, robots and software as well as radio equipment. Samsung's share is undisclosed. Even the entire program is less than 3% of the 600 billion won Samsung Networks booked in revenue in the first quarter of 2026 alone.

A shrinking unit needs a new category

That first-quarter figure is the context the announcement leaves out. Samsung Networks' revenue of about 600 billion won ($410 million) was down 25% from a year earlier and less than half the 1.28 trillion won it reported in the first quarter of 2019, according to Light Reading's analysis of Samsung's disclosures. Light Reading called it the unit's lowest first-quarter revenue since before 2018. Samsung blamed telecom-industry investment headwinds and promised to secure "profitability through structural cost discipline."

The second quarter was better. Samsung said networks earnings improved from both the prior quarter and the prior year on higher overseas sales. It did not publish a separate revenue figure, reporting networks together with the much larger mobile business, which recorded 33.2 trillion won of combined second-quarter revenue and a 0.7 trillion won operating loss. For the second half, Samsung said it aims to secure profitability "through new orders and structural cost discipline amid telco investment headwinds."

Investors treated the contracts as immaterial. Samsung Electronics rose 3.25% on September 23 as the KOSPI gained 0.9%, a move driven by an overnight rally in US chip stocks. KT closed down 0.75% and SK Telecom down 0.23%, little changed on a rising market.

For Samsung Networks, the strategic question is whether AI-RAN can become a spending category that operators fund once coverage is built. Industrial private networks that carry AI workloads are one of the few places where radio equipment can be attached to new revenue, because the factory pays for an outcome rather than a subscriber paying for coverage.

What Samsung is actually shipping

Samsung's platform is software-heavy. Its Network in a Server product runs virtualized RAN, a 5G standalone core and AI applications on Supermicro edge servers with AMD EPYC processors, managed through Samsung's CognitiV operations suite. GPU acceleration is added where a workload needs it.

That is a deliberate contrast with the Nvidia–Nokia approach. Nvidia announced an AI-RAN partnership with Nokia in October 2025 together with a $1 billion equity investment, built around GPU-accelerated base stations, with T-Mobile US among the operators planning trials in 2026. Samsung's pitch is that a CPU-first server can host the radio and much of the inference, keeping capital cost down. The Korean sites give it an early chance to show whether the lighter configuration handles real industrial loads.

The technical case for putting compute at the radio site is concrete. A welding robot inside a steel hull or a patrol unit in a refinery needs a short, predictable control loop and large uplink capacity for video, and the plant operator usually wants the data kept on site. "Physical AI requires networks supporting ultra-low latency, high-reliability communication and large-capacity uplink," the program's director at the National Information Society Agency told RCR Wireless. Humanoid-robot demonstrations are planned from 2027.

Vendor position is less exclusive than the headline

Samsung's release stresses its role in both industrial deployments. The wider program is more crowded. RCR Wireless lists Nokia, Ericsson and Korean supplier HFR alongside Samsung among the equipment vendors in the Hyper AI Network consortia. Samsung has won the two flagship industrial sites. It has not locked the Korean operators into a single AI-RAN supplier.

That matters because the prize is replication. A government-funded pilot at one shipyard and one petrochemical complex proves nothing commercially on its own. The value to Samsung comes if HD Hyundai's other yards, Korea's other shipbuilders, or chemical and steel plants buy similar systems on commercial terms, and if the design travels to operators abroad.

Who pays for the servers

Pilots of this kind rarely publish their economics, and these announcements do not. Once public funding ends, someone has to own the edge compute. There are three plausible models: the plant owner buys the servers as part of its automation spending; the operator finances them and recovers the cost through a multi-year managed-service contract; or the compute is bundled into a private-network connectivity price. Each puts the capital on a different balance sheet and gives Samsung a different customer.

The operator-financed model would matter most for Samsung Networks, because it turns telcos back into buyers of radio-site hardware and software. It is also the model operators are most wary of, since KT and SK Telecom would carry compute capex against a single industrial customer's willingness to renew.

CTOL view: No commercially funded follow-on will be announced while the government program runs. By June 30, 2027, neither Samsung, KT, SK Telecom nor HD Hyundai will announce an AI-RAN or private-5G edge-compute contract with a named industrial customer that is paid for outside the Hyper AI Network program. Shipbuilders buy outcomes, not radio equipment, and neither operator has shown willingness to carry compute capital spending before the pilots produce results. Readers can check the four companies' newsrooms. We are wrong if any of them announces such a contract on or before June 30, 2027.

Samsung's South Korea AI-RAN Pilots Put Industrial 5G to the Test

A shrinking unit needs a new category

What Samsung is actually shipping

Vendor position is less exclusive than the headline

Who pays for the servers

Sources

You May Also Like

Subscribe to our Newsletter