TL;DR:
- SAPEON Korea, SK Telecom’s AI semiconductor subsidiary, unveils the X330 AI chip model.
- South Korean mobile carriers aim to reduce reliance on NVIDIA by strengthening AI infrastructure partnerships with local companies.
- KT Corp. collaborates with AI chipmaker Rebellions Inc. to offer Mi:dm AI model to B2B customers.
- Microsoft enters the AI chip market with Maia 100.
- SK Telecom’s X330 outperforms NVIDIA’s L40S model, enhancing AI capabilities.
- KT achieves a sixfold increase in power efficiency and 50% cost reduction through collaboration with Rebellions.
- Rebellions plans to mass-produce “ATOM+” chips for ultra-large AI models in Q1 2024.
Main AI News:
In the race to establish a robust AI chip infrastructure, SAPEON Korea Inc., the AI semiconductor subsidiary of South Korean mobile giant SK Telecom Co., introduced its latest AI chip model, the X330. This unveiling signals a heightened competition among local mobile carriers in South Korea to expand their foothold in the AI chip industry. Instead of solely relying on NVIDIA Corp.’s infrastructure, Korean mobile carriers are forging strong collaborations with domestic companies to bolster their AI infrastructure capabilities.
SK Telecom is leading the charge by incorporating SAPEON’s AI chips and advanced spam detection technology into its services. Meanwhile, KT Corp. is collaborating with AI chipmaker Rebellions Inc. to offer its hyperscale AI model, Mi:dm, to its customer base.
As things stand, NVIDIA dominates the AI chip market, commanding a staggering 80 to 90 percent market share with its graphics processing units (GPUs). However, companies are beginning to explore the AI chip business to reduce their reliance on NVIDIA and gain autonomy over their technology. Notably, Microsoft Corp. unveiled its own AI chip, Maia 100, during Microsoft Ignite 2023 earlier this month.
SK Telecom’s strategic focus revolves around enhancing the competitiveness of its AI solutions through SAPEON. The recently launched neural network processing unit (NPU) X330 is a game-changer, significantly improving AI performance by supporting transformer-based large language model (LLM) operations. Impressively, its computational performance outpaces NVIDIA’s L40S model by nearly double, according to SAPEON.
Furthermore, SK Telecom has established a powerful 7.6-petaops NPU farm, leveraging SAPEON’s technology. During a recent press conference, SAPEON showcased how its semiconductors power SK Telecom’s services, including a vision AI technology that identifies reusable cups in the Happy Habit Project. The X330 stands out by quadrupling the performance of its predecessor, enabling it to handle multiple requests simultaneously efficiently. SK Telecom’s future plans include utilizing SAPEON’s chips to develop voice phishing prevention, as well as spam and smishing text detection technologies.
On the other front, KT is actively collaborating with Rebellions in the realm of business-to-business (B2B) transactions. In addition to employing Rebellions’ NPU ATOM for its in-house services, KT is extending the super-sized AI Mi:dm, powered by Rebellions’ technology, to its B2B clientele. In July 2022, Rebellions, a local AI semiconductor company, joined forces with KT in an AI full-stack alliance, with KT making a strategic investment of 30 billion won ($23.24 million). Similar to SAPEON’s X330, Rebellions’ ATOM is fully equipped to support transformer models and execute floating-point calculations for large language models. Prior to the Mi:dm’s launch, Rebellions collaborated with KT Cloud Corp. to commercialize cloud-based NPU infrastructure. Through this collaboration, KT has fortified its infrastructure and enhanced cost competitiveness. “By collaborating with Rebellions, we have achieved a sixfold increase in power efficiency and reduced inference costs by 50 percent,” proclaimed KT at the Mi:dm’s launch announcement.
Rebellions, looking ahead, intends to commence mass production of “ATOM+” chips optimized for ultra-large AI models in the first quarter of 2024. The future of AI chip infrastructure in South Korea is poised for exciting advancements, as mobile carriers and semiconductor companies alike gear up to shape the AI landscape.
Conclusion:
South Korean mobile carriers are aggressively expanding their AI chip infrastructure capabilities through collaborations with local companies like SAPEON and Rebellions. This strategic move not only reduces their dependence on NVIDIA but also positions them to capture a larger share of the AI chip market. Microsoft’s entry into the market further emphasizes the growing significance of AI chips. These developments signal a competitive landscape shift and potential innovation surge in the AI chip industry.
