Infinigence AI

River AI is a full-stack AI company founded by xAI co-founder Igor Babuschkin, building an open AI stack. Its River API delivers LoRA fine-tuning and reinforcement learning for frontier open-weight models, with token-metered billing and instant deployment, letting developers and companies train, tune and serve their own models without dedicated training infrastructure.

Series A · $1.1B · 2026-08 · Valuation $5.0B

Hugging Face, founded in 2016 in Paris, France and New York, USA. It currently ranks as the fifth most valuable generative AI unicorn.

Undisclosed · $335M · 2023-08 · Valuation $4.5B

Lamini, founded in 2020 in Palo Alto, California, provides an enterprise platform for LLM tuning and inference, centered on Memory Tuning to reduce hallucinations, deployable on-premise or in any cloud on Nvidia or AMD GPUs. In June 2025 co-founder and CEO Sharon Zhou and several employees joined AMD; this was not an acquisition.

Undisclosed · $25M · 2024-05

TokenRhythm is an AI infrastructure startup founded in April 2026 by Wang Yunhe, former director of Huawei's Noah's Ark Lab and head of the Pangu large model, with CTO Han Kai, a former chief researcher at the same lab. It builds a routing harness layer between foundation models and applications via the TokenRhythm API, an OpenAI/Claude-protocol-compatible platform aggregating domestic models such as DeepSeek, GLM and Kimi, positioned as a Chinese counterpart to OpenRouter. It also maintains the open-source AI agent project OpenSquilla and is exploring next-generation agent-native models trained on real agent task feedback.

Angel · 2026-08

Visincept (视启未来) was founded in 2025 in Shenzhen, China, focusing on general-purpose visual large model research and applications. The company carries core IP such as DINO-X and offers MaaS (Model-as-a-Service), private deployment, and embodied intelligence vision solutions, with products including DINO-X, T-Rex Label, and CountAnything.

Angel · $51M · 2025-11

MetaCircle (Shanghai Yuanhuan Intelligent Technology Co., Ltd.), founded in 2026 in Shanghai, works on "AI for AI": using AI to optimize and automate the research and development workflow of AI models themselves. Its core team comes from Tsinghua University, Peking University and MIT.

种子轮及种子+轮 · 2026-10

Flower Labs, founded in 2020 in Hamburg, Germany, provides a federated learning framework for developers to implement federated learning in machine learning projects.

Series A · $24M · 2024-02

Glaive, founded in 2023 in San Francisco, USA, aims to train small language models for specific use cases using synthetic data.

Seed · $4M · 2023-08

Prem AI, founded in 2023 in Lugano, Switzerland, provides private, sovereignty-preserving local AI infrastructure that enables regulated institutions such as hedge funds and law firms to run AI agents and models on their own infrastructure, addressing privacy and compliance challenges for AI applications involving sensitive data. Media reports indicate the company is targeting a valuation of no less than $500 million in its current round.

CentML, founded in 2022 in Toronto, Canada, develops tools to reduce the cost and improve the performance of deploying machine learning models.

Seed · $27M · 2023-10