Fermi Universe
Category: Foundation Models / LLMs
China's first startup focused on Q4AI (Quantum for AI), applying quantum physics and quantum computing methods to rebuild the underlying architecture of large language models. Fermi Universe was founded in 2025. The company is led by Li Wei (李伟). Based in Beijing, China. Team size: 11-50. Total funding raised: $14M. Latest round: Seed.
- Founded
- 2025
- Headquarters
- Beijing, China
- Team size
- 11-50
- Total funding
- $14M
Value proposition
Quantum-enhanced foundation models that improve reasoning performance and cut training costs without requiring fault-tolerant quantum hardware, deployable on existing GPU infrastructure.
Products and solutions
FermiQLLM 1.0 — world's first full-chain quantum-enhanced large model, built on the Qwen open-source base model, quantum enhancement across 5 stages (data representation, model structure, training, reinforcement, evaluation).
Unique value
World's first full-chain quantum-enhanced LLM (FermiQLLM 1.0) — systematically embeds tensor networks, quantum simulated annealing, and gauge degrees of freedom into the entire LLM pipeline (data representation, architecture, training, RL, evaluation) while remaining compatible with mainstream AI compute.
Target customer
Traditional LLM vendors (as both customers and partners) and their enterprise end-customers across industries.
Industries served
Foundation models / AI infrastructure, quantum-inspired optimization
Technology advantage
Quantum physics-inspired methods (tensor networks, quantum simulated annealing, gauge freedom) embedded into the LLM pipeline; does not depend on immature fault-tolerant quantum computers — runs on existing GPU infrastructure. Reported >15% inference performance gain, >25% training cost reduction in RL, and 10-20% composite benchmark improvement (MATH-500, GPQA-Diamond, BBH) vs. same-scale open base models.
How they differentiate
Unlike hardware-dependent quantum-AI players, Fermi Universe takes the "quantum-inspired on classical compute" route — embedding quantum physics/many-body methods into the full LLM pipeline so it works on existing GPUs without waiting for fault-tolerant quantum computers. Differentiates via a strong Tsinghua talent base (~half of employees) and academic collaboration with Tsinghua's Yangzhenning Institute. Closest domestic rival 清醒异构 focuses on quantum-inspired vision-language models (RiverONE) and AMD/Biren GPU adaptation, whereas Fermi Universe targets full-chain quantum enhancement of LLMs.
Main competitors
清醒异构 / Wake Systems (Shanghai, Tsinghua-affiliated quantum-inspired AI, RiverONE model), SandboxAQ (US, post-quantum crypto & quantum-inspired optimization, valued $5.3B), Xanadu (Canada, quantum ML frameworks)
Key partnerships
Collaboration with Tsinghua University Yangzhenning Institute for Advanced Study quantum research team (Prof. Yao Hong), Tsinghua Q4AI internship site.
Major milestones
Project launched March 2026 (4B model feasibility validation), company formally established May 2026, launched FermiQLLM 1.0 (world's first full-chain quantum-enhanced LLM), raised RMB 100M seed round (largest in China's Q4AI sector).
Growth metrics
Founded May 2026 (project started March 2026); ~RMB 1B post-money valuation on RMB 100M seed; ~half of employees from Tsinghua University.
Market positioning
China's first and largest Q4AI (Quantum for AI) seed round; ~RMB 1B post-money valuation. Positions itself as a foundational model company whose quantum-enhanced approach can serve both traditional LLM vendors and their customers.
Geographic focus
China (domestic Q4AI market), with awareness of overseas players (SandboxAQ, Xanadu)
About Li Wei (李伟)
Ex-NetEase Youdao and Renren (技术高管 / AI技术委员会主席); Tsinghua University CS NLP Lab (Class of 1998), BS & MS from Tsinghua; 1998 Hunan Province Gaokao science top scorer; translated "Foundations of Statistical Natural Language Processing" (2005).