About TokensChain
The intelligent traffic router
between China's compute
and global AI demand.
We don't own compute — we route it. As China's AI middleware MaaS + compute-scheduling infrastructure, we make Chinese cloud-GPU LLM inference as cheap, reliable and compliant as the power grid, for enterprises worldwide.
Build the AI middleware MaaS + compute-scheduling infrastructure that makes Chinese cloud-GPU LLM inference affordable and trustworthy for every enterprise, everywhere.
Become the Akamai of the AI era — the intelligent router between Chinese compute and global demand.
Compliance first. Efficiency above all. Customer success. Long-term thinking.
Our story
TokensChain is the Fireworks of China's compute.
Since 2025, generative AI has exploded worldwide — and especially in China. Yet enterprises everywhere hit the same three walls: runaway cost, compliance risk, and the pain of switching between models.
Fireworks captured that space abroad with software-layer routing, OpenAI compatibility and day-0 support for every new open model — becoming the default on-ramp to open-model inference for the rest of the world. The equivalent seat — routing China's compute to the world — stayed empty. China compliance is the moat, software is the tool, global routing is the opportunity.
TokensChain exists for that reason. We mirror Fireworks' market-validated playbook on China's clouds — aggregating GPUs from the country's top providers, OpenAI-compatible, day-0 model support — and layer on deep China compliance and global delivery so the rest of the world gets the same developer experience on Chinese compute.
Why now
A three-year window is opening.
Global LLM inference demand ignites; closed APIs become the default.
Software-layer aggregation of open models with OpenAI-compatible routing becomes the default on-ramp abroad.
DeepSeek, Qwen, Kimi, GLM, MiniMax and others close the gap with closed-source SOTA.
Bring the Fireworks experience to China's compute — and deliver China's compute to global developers.
Values
Principles guiding every line of code and every decision we make.
People-first, open collaboration
We win when our customers and partners win. We believe in open collaboration, long-term relationships and sustained investment in the developer community.
Pragmatism & precision
We solve real problems with clear, direct solutions. Performance, reliability and compliance leave no room for hand-waving — every number must hold up under scrutiny.
Innovation & excellence
We push boundaries responsibly. Every release and every feature ships to a production-ready bar, and we keep raising the developer-experience floor.
Open source
Giving back to the community through open source.
We open-source the core of our inference engine and runtime so any developer can build generative AI without friction.
SmartDiff
A lightning-fast inference engine for diffusion models, optimized for real-time image and video generation. Drop-in support for SD / FLUX and Chinese open models.
View on GitHubTokenRuntime
An AI-native runtime built for scalable inference workloads across large language and multimodal models. Designed for flexibility, observability and raw performance.
View on GitHubLeadership
A team that knows compute, models and global markets.
Our founding team comes from top tech enterprises, global investment firms and cross-border infrastructure organizations, with 15+ years in AI, cloud and global markets.
Milestones
From 0 to 1, then to N
Contact us
Let's talk.
Whether you're an enterprise customer, an investor, or thinking of joining us — we'd love to hear from you.