趋境科技
Category: AI Infrastructure
AI infrastructure company building a high-performance Token-as-a-Service (ATaaS) platform for enterprise AI inference, spun out of Tsinghua University's HPC Institute. 趋境科技 was founded in 2023. The company is led by 艾智远 (Ai Zhiyuan). Based in Beijing, China. Team size: 101-500. Total funding raised: $50.0M. Key investors include 高瓴创投 (GL Ventures), Z基金/星连资本 (Z Fund/Star Capital), 华控科技 (Huakong Technology), 真知创投 (Zhenzhi Capital), 弘晖资本, 水木校友种子基金.
- Founded
- 2023
- Headquarters
- Beijing, China
- Team size
- 101-500
- Total funding
- $50.0M
Value proposition
Token-as-a-Service (TaaS) — shifting from model-centric MaaS to token-centric production, delivering high-quality AI tokens at scale with low latency (TTFT), high throughput (30-50 TPS), structured output stability, and predictable service quality for enterprise production environments.
Products and solutions
ATaaS (Approaching AI Token as a Service) platform — high-performance AI Token production service, 大模型知识推理一体机 (Large Model Knowledge Inference All-in-One Machine), KTransformers (open-source edge heterogeneous inference framework, 17k+ GitHub stars), Mooncake (distributed inference framework, co-built with Tsinghua/Kimi)
Unique value
"少模型、深优化" (fewer models, deeper optimization) approach — focuses on a select few high-productivity models with deep optimization rather than supporting hundreds of models; full-stack heterogeneous computing orchestration, cross-cluster cache sharing, and "以存换算" (storage-for-compute) innovation that reduces deployment costs by 10x and latency by 20x.
Target customer
Enterprise AI teams, large model companies (e.g., Zhipu GLM, Moonshot Kimi), financial services, cybersecurity, telecom, education, aerospace/defense sectors needing production-grade AI inference infrastructure.
Industries served
AI Infrastructure, Financial Services, Cybersecurity, Telecommunications, Education, Aerospace/Defense
Technology advantage
Full-stack heterogeneous computing orchestration; cross-cluster cache sharing; inference pipeline isolation; elastic scaling; "以存换算" (storage-for-compute) architecture; KTransformers (first edge heterogeneous inference framework, 17k+ GitHub stars); Mooncake distributed inference framework; 20+ years of Tsinghua HPC research heritage; 90%+ master's degree, 50%+ PhD team composition.
How they differentiate
Pioneered the TaaS (Token-as-a-Service) paradigm vs. traditional MaaS; "以存换算" (storage-for-compute) innovation reducing inference costs by 10x; deep optimization on select models rather than broad model support; full-stack heterogeneous compute orchestration from hardware to application layer; strong open-source ecosystem (KTransformers, Mooncake) with 17k+ GitHub stars.
Main competitors
硅基流动 (SiliconFlow), 无问芯穹 (Infini-AI), 清程极智 (Qingcheng Jizhi)
Key partnerships
Tsinghua University HPC Institute (郑纬民院士, 武永卫教授, 章明星副教授), 京东云 (JD Cloud) — joint next-gen inference engine, 华为昇腾 (Huawei Ascend) — full-stack tech & commercial collaboration, 阿里云 (Alibaba Cloud), 英特尔 (Intel), 金航数码 (Jinhang Digital, AVIC subsidiary), 长亭科技 (Chaitin Security), 中国电子云 (China Electronics Cloud), 商汤科技 (SenseTime), 月之暗面 Kimi (Moonshot AI), 智谱 AI (Zhipu AI)
Notable customers
智谱 GLM (Zhipu AI), 月之暗面 Kimi (Moonshot AI), MiniMax, 通义千问 Qwen (Alibaba)
Major milestones
2023-12: Company founded, 2024-08: Angel round led by 真知创投, 2024-10: Launched Large Model Knowledge Inference All-in-One Machine, 2025-02: Angel round led by 高瓴创投 & Z基金, 2025-06: Intel ecosystem partnership, 2025-12: Partnership with 金航数码 (AVIC), 2026-03: Launched ATaaS platform, 2026-05: Pre-A round of hundreds of millions RMB, KTransformers reached 17k+ GitHub stars, ATaaS platform processes ~1 trillion tokens daily
Growth metrics
ATaaS platform processes ~1 trillion tokens daily; KTransformers 17k+ GitHub stars; 30-50 TPS output capability
Market positioning
Tier-1 AI infrastructure startup in China, positioned as a "Token Factory" — differentiated from MaaS platforms by focusing on token production quality and efficiency rather than model breadth. Strong Tsinghua academic lineage with院士-level scientific advisory.
Geographic focus
China (domestic market primarily, with global open-source presence)
Patents and IP
Tsinghua University technology成果作价增资入股 (technology成果 valued and injected as equity); KTransformers open-source framework (17k+ GitHub stars); Mooncake distributed inference framework (co-developed with Tsinghua/Kimi/Alibaba/Ant Group)
About 艾智远 (Ai Zhiyuan)
Ex-深信服 (Sangfor Technologies) R&D Manager for Big Data, Digitalization, and AI Application departments. PhD in Computer Science from Tsinghua University (2013-2018), B.Eng from Harbin Institute of Technology (2009-2013). Led multiple multi-million-dollar project deliveries.
Latest news about 趋境科技
More AI Infrastructure companies
Official website: https://approaching-ai.com
