Fermi Universe Raises RMB 100M to Apply Quantum Methods to Foundation Models
The AMW Read
A newly funded Chinese entrant proposes quantum-inspired optimization across the foundation-model stack, but its performance claims remain internally reported and early-stage.
Fermi Universe Raises RMB 100M to Apply Quantum Methods to Foundation Models
Fermi Universe, a China-based Quantum for AI startup, has raised a reported RMB 100 million (about $14 million) in seed funding at an approximately RMB 1 billion (about $140 million) post-money valuation. The company says it has launched FermiQLLM 1.0, a quantum-enhanced large language model built on an open-source Qwen base model. Its approach does not require a fault-tolerant quantum computer: it applies methods including tensor networks, quantum simulated annealing, and gauge degrees of freedom within existing classical AI infrastructure.
The company claims that, against comparable open-source base models, FermiQLLM 1.0 improved aggregate results by 10% to 20% on MATH-500, GPQA-Diamond, and BBH; it also reports more than 15% better inference performance and over 25% lower training cost in continuous reinforcement learning. Those are company-reported internal results, not independently validated benchmarks. Still, the launch puts Fermi Universe on the map as an early attempt to turn quantum-inspired techniques into a model-layer optimization strategy rather than a future bet on quantum-hardware availability.
For builders, the relevant test is portability: Fermi Universe plans over the next six to 12 months to generalize its current single-base-model capability into a framework that supports multiple LLMs and automated optimization. For investors, the seed round is a modest but notable wager on whether measurable efficiency and quality gains can be reproduced across models, workloads, and independent evaluations before the approach becomes another specialized research technique.
