KT Cloud Begins Validating Rebellions' REBEL-100 AI Chip to Slash LLM Inference Costs

KT Cloud has begun testing Rebellions' next-generation AI chip, REBEL-100, aiming to validate its long-standing goal of cutting LLM inference costs by 50%.

This move is a crucial step for Rebellions, providing commercial validation with a major client and boosting the credibility of domestic NPUs as a viable alternative to GPUs.

The partnership is strengthened by KT's strategic investment in Rebellions, government support for domestic AI chips, and a strong product message focused on reducing total cost of ownership (TCO).

030200.KS

KT Cloud has officially begun validating 'REBEL-100', the next-generation AI chip from Korean startup Rebellions, in its data centers.

This test marks a critical moment in the race to solve one of the biggest challenges in the AI era: the soaring cost of running large language models (LLMs). For years, NVIDIA's GPUs have dominated the market, but their high price and power consumption have created a huge demand for more efficient alternatives, especially for LLM inference—the process of using a trained model to generate answers. Rebellions is tackling this head-on, positioning its rack-scale systems with a compelling message for decision-makers: up to 75% lower purchase cost and 6 times less power compared to GPU-based solutions.

This collaboration didn't happen overnight. It's the result of a carefully built alignment between customer and vendor. First, KT is not just a customer; it's a strategic investor that co-led Rebellions' Series B funding round in early 2024. This 'customer-capital alignment' creates a closed loop where investment de-risks the partnership and speeds up the journey from initial testing to large-scale deployment.

Second, the ecosystem around Rebellions has matured significantly. The company's 2024 merger with its rival Sapeon created Korea's largest AI chip unicorn, reassuring large customers like KT of its long-term stability. Furthermore, REBEL-100 is built on a reliable domestic supply chain, using Samsung's 4nm process and HBM3E memory from Korean tech giants, which minimizes production risks.

Finally, supportive government policies have paved the way. Initiatives like the national 'Micro Data Center (MDC)' project, which promotes the use of domestic AI chips, provide a structured framework for evaluation and adoption. This public-private partnership lowers the barrier for companies like KT to invest in homegrown technology. Therefore, KT's validation of REBEL-100 is more than just a technical trial; it's a culmination of strategic investment, supply chain strength, and national policy aimed at creating a sustainable AI infrastructure.

NPU (Neural Processing Unit): A specialized processor designed to accelerate machine learning and AI tasks, often more power-efficient than general-purpose GPUs for specific operations like inference.
TCO (Total Cost of Ownership): A financial estimate that includes not only the initial purchase price of an asset but also all direct and indirect costs of operating it over its lifespan, such as electricity, cooling, and maintenance.
LLM Inference: The process of using a pre-trained large language model to perform a task, such as answering a question or generating text. This is distinct from 'training,' which is the computationally intensive process of creating the model itself.