Economy/Home · Economy

[Seohak!Star] NVIDIA, Rebellions, and SK Telecom Form 'AI Alliance'... GPUs Combined with Domestic NPUs May Reshape the AI Infrastructure Landscape

NVIDIA is joining hands with Korean AI semiconductor company Rebellions and SK Telecom to build Korea's own artificial intelligence (AI) infrastructure. The partnership is based on NVIDIA's high-speed interconnect technology 'NVLink Fusion' and will integrate domestic AI semiconductors, data centers

Oseong Kwon
Staff Reporter
14 min read
[Seohak!Star] NVIDIA, Rebellions, and SK Telecom Form 'AI Alliance'... GPUs Combined with Domestic NPUs May Reshape the AI Infrastructure Landscape
CBC News

NVIDIA is joining hands with Korean AI semiconductor company Rebellions and SK Telecom to build Korea's own artificial intelligence (AI) infrastructure. The partnership is based on NVIDIA's high-speed interconnect technology 'NVLink Fusion' and will integrate domestic AI semiconductors, data centers, and homegrown AI models into a single system.

On the 8th (local time), NVIDIA announced a partnership with Rebellions and SK Telecom to advance Korea's Sovereign AI infrastructure. The Ministry of Science and ICT is also participating in the cooperation, supporting the linkage of domestic semiconductor technology with global AI infrastructure.

■ GPUs and Domestic NPUs Divide Roles Within a Single System

The core of the partnership is connecting Rebellions' next-generation neural processing unit (NPU) 'Rebel R200', currently under development, to NVIDIA's AI infrastructure platform. Rebellions plans to use NVIDIA's NVLink Fusion technology to connect the R200 at high speed and to develop large-scale AI inference systems based on NVIDIA's standardized rack design platform 'MGX'. SK Telecom plans to adopt this system in its AI data centers and link it with its own AI models and services. The idea is to secure 'full-stack sovereign AI' competitiveness spanning semiconductors, data centers, AI models, and actual services.

Particularly notable is the division of roles between NVIDIA GPUs and the domestic NPU. The three companies plan to develop a 'disaggregated inference' architecture in which different semiconductors handle different stages of AI answer generation. Specifically, NVIDIA GPUs will handle the 'prefill' stage, where AI analyzes and processes input, while Rebellions' NPU will be deployed for the 'decode' stage, where actual answers are generated sequentially. The strategy is to assign each semiconductor the tasks it excels at, thereby increasing AI inference throughput and improving power efficiency.

Rebellions' R200, now in development, is being designed to process inference workloads for ultra-large AI models with low latency and high energy efficiency, aiming to support even Mixture of Experts (MoE) models with trillions of parameters.

NVLink Fusion enables such NPUs to be combined with NVIDIA's high-bandwidth, low-latency interconnect technology, making it possible to build rack-scale AI infrastructure in which multiple processors operate as a single system, going beyond the performance of individual chips. NVIDIA's MGX architecture also plays an important role. MGX is a standardized infrastructure structure that allows GPU and NPU systems to share key data center design elements such as rack size, power supply, and cooling. Rebellions expects this will reduce the burden of system development and commercialization while leveraging existing manufacturers and component supply chains.

■ 'Korea Can Become a Global AI Leader' — NVIDIA CEO Jensen Huang

NVIDIA founder and CEO Jensen Huang said, 'Korea has the talent, technology, and industrial capacity to become a world leader in AI.' Huang explained, 'Sovereign AI enables each country to transform its own language, culture, and expertise into intelligence,' adding, 'Through NVLink Fusion, Korean semiconductor companies can connect their own designed ASICs (application-specific semiconductors) to the NVIDIA AI factory platform and integrate them closely with NVIDIA's computing technology.' He emphasized that NVIDIA will work with the Ministry of Science and ICT, Rebellions, and SK Telecom to build a computing foundation that allows Korea to secure its own AI capabilities and drive next-generation industrial innovation.

Park Sunghyun, co-founder and CEO of Rebellions, assessed the announcement as a turning point in connecting Korea's AI semiconductors and AI models to global system technology. He stated that by combining Rebellions' semiconductor and system design technologies with NVIDIA's high-speed interconnect technology and rack-scale design capabilities, and SK Telecom's AI models and data center operations expertise, they will realize low-latency, high-efficiency inference for large-scale AI models. He also expressed his will to play a leading role in positioning Korea's AI semiconductor technology as a key pillar supporting the global AI industry.

■ SK Telecom's 'Full-Stack Sovereign AI'... Government Calls It 'Significant Progress'

SK Telecom views the partnership as an opportunity to concretize its sovereign AI strategy. SK Telecom plans to utilize the new infrastructure for its self-developed AI model 'A.X K2' and public AI services linked to the government's 'AI for All' project. SK Telecom CEO Jung Jae-heon said, 'This is significant in that we have confirmed the potential of next-generation sovereign AI infrastructure by combining global AI infrastructure technology with domestic AI semiconductors.' He added that SK Telecom will strengthen its full-stack sovereign AI competitiveness by connecting its AI data centers, proprietary AI models, and public AI services, while also contributing to the growth of the domestic AI ecosystem.

The government also sees the partnership as a growth opportunity for the domestic AI semiconductor industry. Deputy Prime Minister and Minister of Science and ICT Bae Kyung-hoon evaluated, 'This cooperation is significant progress in that domestically developed NPUs can now enter the computing infrastructure market together with NVIDIA, a global leader in AI computing, through NVLink Fusion.' The government plans to support the partnership so it leads to tangible results, while also expanding cooperation with NVIDIA in various areas such as physical AI and securing GPU supply.

■ Commercialization Timing and Performance Effects Still 'In Progress'

The announcement is also noteworthy in that NVIDIA is strengthening an open infrastructure strategy that accommodates semiconductors designed by external companies while maintaining its GPU-centric AI ecosystem. For domestic AI semiconductor companies, this opens a path to enter the large-scale AI systems market by leveraging NVIDIA's interconnect technology and data center infrastructure ecosystem.

However, the partnership is centered on next-generation system development and future data center adoption. The actual commercialization timeline, deployment scale, and improvements in inference performance and power efficiency are expected to be confirmed through future development and operations. Attention is focusing on whether the combination of NVIDIA's GPU technology, Rebellions' domestic NPU, and SK Telecom's AI data centers and proprietary models will allow Korean-style sovereign AI infrastructure to enter a new phase.

[Notice] 'Seohak!Star' is overseas stock content, including U.S. stocks, provided by CBC News. We will continue to strive to provide readers with a wide range of information. All investments are made at each individual's own judgment and discretion, and the final responsibility lies with the investor.

Oseong Kwon
Staff Reporter

CBC Globe publishes verified stories with editorial review, source checks, and tenant-specific publication standards.