About Us
Computing x Power x Carbon - an intelligent computing operator built on compute-power-carbon synergy
Our Story
Founded in 2025, 8341TokenFactory is the technology transfer platform of the Juyongguan Key Laboratory of Energy Security and Intelligent Technology (a provincial and ministerial-level key laboratory) at North China Electric Power University. Drawing on the laboratory's long-running research in artificial intelligence, energy management and carbon emission control, we turn that work into AI capabilities and computing services developers can call directly.
Our technology comes from two lines of work. The first is artificial intelligence - model aggregation, inference scheduling and energy-efficiency optimization. The second is energy - clean power supply, data centre energy management and carbon emission control. In the laboratory these two were always done together: AI has to run, and running costs power; keeping that power cheap and green depends on energy management and carbon reduction.
8341TokenFactory combines both into one platform: Compute-Power-Carbon Synergy underneath, a unified LLM API on top. Developers integrate once and can call all mainstream models, while benefiting from the cost advantage created by energy management and carbon control. In essence we do one thing - make AI affordable, and make AI clean.
Our Mission
Turn the laboratory's technology into AI and energy services that developers can genuinely afford and use, lowering the barrier to AI while managing the energy consumption and carbon emissions of computing at the same time.
Computing Infrastructure
The company has built large-scale computing centers in Beijing, Ulanqab, Weihai (Shandong), Shaoguan (Guangdong), Guizhou, Horinger (Inner Mongolia), Zhongwei (Ningxia), and other locations. Leveraging green clean energy and natural cooling conditions, we provide efficient, low-carbon intelligent computing services to users nationwide. Through our self-developed computing scheduling algorithms, we achieve cross-regional intelligent routing and dynamic load balancing, enabling users across the country to experience millisecond-level low-latency responses with virtually no perception of delays caused by physical distance. Our accumulated expertise in silicon-carbon phase-change technology for heat dissipation and energy efficiency underpins our cost advantage in Compute-Power-Carbon Synergy.
Global Service Nodes
The platform has deployed 10 service nodes across China, Asia-Pacific, North America, and Europe, enabling nearby access and cross-region scheduling so users in core cities like Beijing, Shanghai, and Guangzhou as well as overseas users all get millisecond-level low-latency responses.
Nearby access · Intelligent cross-region routing · Automatic failover on single-node anomalies, 99.9% service availability
Core Values
Innovation
Continuously tracking cutting-edge AI technologies, integrating the latest models at the earliest opportunity to provide users with the most advanced AI capabilities.
User First
Developer experience at the core — clean APIs, comprehensive documentation, and responsive customer support.
Security & Reliability
Enterprise-grade data security, strict permission management, and encrypted transmission to keep your data safe.
Open & Win-Win
Embracing openness, compatible with mainstream API standards, building a thriving AI ecosystem with partners.
Core Technology: Compute-Power-Carbon Synergy
We do not train models, we sell tokens. Our moat is the physical layer - compute-power-carbon synergy - not software-level engineering optimization.
Compute-Power Co-scheduling Engine
Collects spot electricity price curves, wind and solar generation forecasts and grid demand-response signals in real time, then uses multi-agent reinforcement learning to globally minimize electricity cost and carbon emissions under latency constraints. Deferrable requests are moved to western nodes with low power prices and abundant Compute-Power-Carbon Synergy - measured savings of about 65%.
LoRe - Physics-driven Inference Acceleration
Brings condensed-matter physics methods into AI inference, using cluster-bath dynamic approximation to compute high-information-density tokens exactly while approximating the rest globally. Measured 8-15x speed-up and 12-44x memory reduction, giving domestic GPUs high-end inference capability and cutting token cost by another 5-10x.
TokenRouter - Token-level Routing (developed with a Tsinghua University team)
Not every request deserves a top-tier model: easy questions go to small models, and only hard ones escalate. TokenRouter lets each token decide on its own. Measured decoding throughput improves by up to 64x and end-to-end latency drops by up to 63x, while routing only 5% of tokens to the large model already matches full large-model quality. The same quality with far fewer tokens - your bill goes down, quality does not.
Core Team
The team combines PhDs from Peking University, Tsinghua University and North China Electric Power University with senior experts from leading technology companies, covering AI algorithms, chip-level engineering, renewable energy and cloud commercialization.
CEO - Compute-Power Synergy
PhD in AI from Peking University and national-level talent; former head of an AI platform at a leading technology company; 15 years in the AI industry. Author of the compute-power-carbon strategy.
CTO - Chips & Inference
Engineering PhD from Tsinghua University; former head of AI and chip R&D at a leading technology company; 20+ invention patents. Leads inference acceleration and heterogeneous compute management.
Compute-Power Synergy - Green Power & Algorithms
PhD from North China Electric Power University and contributor to a National Science and Technology Progress Award; focused on direct green power supply and compute-power scheduling algorithms.
Head of Operations - Commercialization
Former senior product expert at a leading technology company, with a track record of driving nine-figure RMB revenue; leads commercialization and customer growth.
FAQs About the Platform and Compute
Common questions about our company background, self-built compute, and technical capabilities