Multiverse Computing raises up to $570M to shrink LLMs

Multiverse Computing closed a Series C of up to $570 million at a roughly $1.7 billion valuation, about five times its prior round, per Bloomberg.

The bet behind that number is that inference cost, not model capability, is the binding constraint on AI deployment in 2026.

Multiverse doesn't train models, it shrinks them. The target is the gap between what a full-size LLM costs to run and what companies can afford on cloud, on-prem, or edge hardware.

Forgepoint Capital International, BNPP Solar Impulse Venture Fund, and Bullhound Capital co-led the round, according to Bloomberg.

CompactifAI, Multiverse's compression product, applies tensor-network methods borrowed from quantum physics to large language models. The company says the technique cuts model size by 80 to 95 percent with minimal accuracy loss.

Bloomberg's report doesn't say what minimal accuracy loss means in practice at the higher end of that range. It also doesn't say whether the result holds outside Multiverse's own benchmarks.