Samsung-Backed AI Chip Startup Rebellions Raises $400 Million in Pre-IPO Round at $2.3 Billion Valuation
South Korean fabless chip designer Rebellions has closed a $400 million pre-IPO round led by Mirae Asset and the Korea National Growth Fund, bringing total funding to $850 million as the company launches rack-scale inference platforms and prepares for a public listing.
South Korean AI chip startup Rebellions has raised $400 million in a pre-IPO funding round that values the company at approximately $2.34 billion, according to TechCrunch. The round was led by Mirae Asset Financial Group and the state-backed Korea National Growth Fund, which contributed roughly $165 million — its first direct investment under South Korea’s K-Nvidia initiative to strengthen the country’s domestic semiconductor sector.
The fresh capital brings Rebellions’ total fundraising to $850 million, with $650 million of that raised in the last six months alone. The company has hired JPMorgan to manage its South Korean IPO filing, with a listing expected in late 2026 or early 2027, as The Register reported.
Inference, Not Training
Founded in late 2020, Rebellions designs chips optimized for AI inference — the compute required to run trained models in production — rather than the training workloads that have defined Nvidia’s dominance. The company’s flagship Rebel100 accelerator uses a chiplet architecture manufactured by Samsung, delivering 1 petaFLOP of FP16 performance and 2 petaFLOPS at FP8 precision. Each card carries 144 GB of HBM3e memory with 4.8 TB/s of aggregate bandwidth, all within a 600-watt thermal envelope and a standard PCIe form factor that requires only air cooling, according to The Register.
The inference focus reflects a broader industry shift. As large language models move from research labs into production deployments, the ratio of inference to training compute is widening rapidly, creating an opening for challengers that can deliver better performance per watt on serving workloads.
Rack-Scale Platforms
Alongside the funding announcement, Rebellions unveiled two new infrastructure products designed to move the company beyond selling individual accelerators. RebelRack integrates four nodes with eight Rebel100 accelerators each, totaling 32 accelerators, 64 petaFLOPS of FP8 compute, and 4.6 TB of HBM3e memory in a single rack with quad-400 Gbps networking between nodes. RebelPod scales the architecture further, supporting 8 to 128 nodes connected via 800 Gbps Ethernet interconnect, as The Register detailed.
The rack-scale approach mirrors what Nvidia has done with its DGX and HGX platforms — offering turnkey infrastructure rather than discrete components. For Rebellions, it represents a shift toward selling complete inference solutions to data center operators.
Supply Chain and Strategic Backing
Rebellions’ relationship with Samsung extends beyond chip fabrication. Both Samsung and SK Hynix — the world’s two largest producers of high-bandwidth memory — are investors in the company, giving Rebellions advantageous access to HBM supply at a time when memory remains one of the tightest bottlenecks in AI infrastructure, TechCrunch reported. By manufacturing through Samsung rather than TSMC, Rebellions also avoids competing for foundry capacity with Nvidia, AMD, and the hyperscalers.
The company’s software stack leans on open-source frameworks including vLLM, PyTorch, and Triton, along with llm-d for disaggregated inference and Red Hat OpenShift compatibility. Rebellions is also a member of the PyTorch Foundation.
Global Expansion
Rebellions is moving aggressively beyond its domestic market, with new offices in Japan, Saudi Arabia, Taiwan, and the United States. The company is targeting AI research labs such as Meta and xAI rather than hyperscalers as its initial customer base, according to TechCrunch, with proof-of-concept trials already underway in the U.S.
“AI is now measured by its ability to operate in the real world — at scale, under power constraints, and with clear economic return,” founder and CEO Sunghyun Park said.
The K-Nvidia initiative backing from the Korean government signals Seoul’s ambition to position its semiconductor industry not just as a memory supplier but as a competitor in the logic chip market that powers AI workloads. For Rebellions, the path from $400 million pre-IPO round to public markets will test whether an inference-first strategy and Samsung manufacturing can carve out meaningful share in a market still overwhelmingly defined by Nvidia’s CUDA ecosystem.