South Korean startup FuriosaAI has officially launched its second-generation artificial intelligence chip, Renegade, positioning itself as a formidable contender in the global semiconductor market. The new chip is strategically designed to address the industry's shift towards AI inference, where running trained models efficiently is paramount. By prioritizing power efficiency and cost-effectiveness, Renegade aims to provide a sustainable alternative to established market leaders for data center operations.
A Strategic Focus on Inference
The launch comes as the AI industry increasingly focuses on the operational costs of inference, which involves running AI models in real-world applications. FuriosaAI CEO Paik June-ho noted that by 2030, inference workloads are expected to consume approximately 70 percent of the new AI data center capacity added globally. Consequently, the company has centered its strategy on enabling operators to execute these repetitive tasks at the lowest possible total cost.
FuriosaAI's core mission is to make AI computing more sustainable and accessible as the ecosystem scales. The Renegade chip was developed to handle high volumes of real-time AI queries while consuming less power, directly addressing concerns that computing costs could outpace revenue growth. This approach targets a critical need for economic viability in the deployment of large-scale AI services.
Performance and Power Efficiency
From its inception in 2022, the Renegade chip was developed with a strict power envelope of around 180 watts as a primary design constraint. Instead of pursuing maximum peak performance, the company optimized the chip to efficiently run diverse and evolving AI models within this thermal limit. This design choice reflects a deliberate strategy to balance high performance with practical operational requirements for data centers.
Internal benchmarks indicate that Renegade can manage up to 7.4 times more concurrent users than Nvidia’s RTX Pro 6000 GPU at the same power level. The company claims this superior efficiency can lead to a 40 percent reduction in the total cost of ownership for data center infrastructure. To ensure broad compatibility and cost-effectiveness, the chip utilizes the standard PCIe 5.0 interface rather than a proprietary high-speed connection.
Advanced Hardware and Production Scale
Manufactured on TSMC's 5-nanometer process, the Renegade chip integrates approximately 40 billion transistors and is paired with SK hynix's fourth-generation high-bandwidth memory (HBM3). This advanced hardware configuration delivers up to 512 teraflops of compute performance and 1.5 terabytes per second of memory bandwidth. These specifications place it firmly in the category of high-performance AI accelerators.
FuriosaAI commenced mass production in January 2024, a significant achievement for a startup in the capital-intensive semiconductor industry. The company has already started producing an initial run of 4,000 units this year, with plans to expand to 20,000 units to meet growing customer demand. This ability to reach commercial scale is a crucial differentiator in the competitive AI hardware landscape.
Ecosystem and Strategic Collaborations
To support its hardware, FuriosaAI is expanding its software stack, which includes compilers and software development kits (SDKs) tailored for scalable inference architectures. The company is actively collaborating with major domestic partners, including LG AI Research, LG Uplus, and Samsung SDS, to drive adoption. These partnerships are instrumental in validating Renegade's performance in real-world deployment scenarios and building a robust ecosystem.
In a notable development, Samsung SDS will launch a subscription-based Neural Processing Unit (NPU) service in July, making it the first domestic cloud provider to offer such a service. This platform will allow customers to access Renegade chips through the Samsung Cloud Platform, diversifying its offerings and supporting South Korea's national push for sovereign AI infrastructure. The initiative aligns with the country's broader “K-Moonshot” effort to foster national technology champions.
The introduction of the Renegade chip marks a pivotal moment for FuriosaAI and South Korea's ambitions in the global AI industry. By delivering a powerful yet efficient solution for the rapidly growing inference market, the company addresses a critical need for sustainable AI computing. With strong government backing, key industry partnerships, and a clear technological vision, FuriosaAI is well-positioned to secure a significant role in the future of AI hardware.