Rebellions, a Seongnam, South Korea-based developer of purpose-engineered AI accelerators for hyperscale inference, has raised $400M in pre-IPO funding led by Mirae Asset Venture Investment. The round brings total funding to $850M at a $2.34B valuation. The capital will fund launches of RebelRack and RebelPOD rack-scale systems and accelerate global expansion into the US, Japan, and Saudi Arabia.
Inference Revenue Tops $1T Forecast
The timing aligns with Nvidia's projection of a $1T AI inference revenue opportunity, announced just weeks prior at GTC. South Korea's push intensifies, with Samsung committing a record $73B to AI chips the same week as Rebellions' close. Rebellions positions as a software-centric alternative, emphasizing open-source stacks over proprietary lock-in.
Power Constraints Stall AI Scale
Data centers face surging power demands as inference workloads dominate. Current GPU-heavy setups struggle with density and efficiency for LLMs and MoE models at production scale. Korean telcos like SK Telecom already deploy thousands of Rebellions ATOM cards for LLM services handling 50M daily API calls.
Chiplets Unlock Rack-Scale Efficiency
Rebellions' Rebel100 NPUs use chiplet architecture with UCIe interconnects and 144GB HBM3E, fabbed by Samsung. RebelRack packs 32 accelerators for 64 PFLOPS FP8, air-cooled for enterprise DCs. Full-stack software supports PyTorch 2.x, vLLM, and Triton without forks, enabling faster adoption.
As Sunghyun Park, co-founder and CEO, noted:
“AI is now measured by its ability to operate in the real world at scale, under power constraints, and with clear economic return. That shifts the center of gravity toward inference infrastructure and software that makes that infrastructure usable.”
Open Source Speeds Deployment Cycles
Rebellions differentiates via open-source compatibility, contrasting monolithic competitors. RebelServer integrates 8 RebelCards with AMD EPYC CPUs for up to 2 PFLOPS FP8. Partnerships with Arm Total Design and Red Hat OpenShift AI validate ecosystem fit.
Government Backs Sovereign Challenger
Mirae Asset led with growth-stage conviction, joined by Korea National Growth Fund's first 'K-Nvidia' investment of $165M. Strategic backers Arm, Samsung Ventures, and SK Hynix provide IP, foundry, and HBM synergies post-merger with SK Sapeon. This mix signals national champion status for Korea's AI sovereignty.
AI Chips Explode to $1.1T Market
The AI chip market stands at $121.73B in 2026, projected to reach $1.10T by 2035 at 27.88% CAGR. Rebellions ranks No. 2 globally in CB Insights’ AI inference startups, ahead of Groq, SambaNova Systems, FuriosaAI, Tenstorrent, and Cerebras. Chiplet trends favor scalable inference over wafer-scale or GPU dominance.
MIT Alums Drive Hardware Wins
Co-founder and CEO Sunghyun Park designed ASICs for SpaceX Starlink and ML accelerators in Samsung Galaxy S9, backed by MIT PhD. Kyeong-Jae Lee, head of forward deployed engineering, architected at Lion Semiconductor, acquired by Cirrus Logic for $335M. Their exits and pedigrees anchor Rebellions' chiplet leadership.
RebelPOD Targets Sovereign Deployments
Rebellions launches RebelRack and RebelPOD now, with PoCs underway at US labs like Meta and xAI. Aggressive hiring spans NPU software engineers to IR managers in Seongnam. Post-funding, JPMorgan advises on potential 2026-2027 IPO.