AINov 11, 2026, 7:26 AM UTC·1 min read

Lenovo Launches AI Token Factory to Tap China's Inference Demand

Lenovo has launched an AI token factory that sends complex tasks to larger models and simpler interactions to smaller ones, targeting China's inference market.

Rows of AI servers in a data center illustrating Lenovo's token factory for inference workloads. — Image: techinasia.com

Lenovo has launched an AI token factory aimed at capturing growing demand for AI inference in China. The system sorts incoming requests by difficulty: complex tasks go to larger AI models, while simpler interactions are handled by smaller ones.

How the routing approach works

The logic behind the design is straightforward. Large models can handle demanding reasoning, but running them for every query is expensive. Many everyday requests do not need that much capacity, so a smaller model can answer them faster and at lower cost. By matching each request to a suitably sized model, the system is meant to use computing resources more efficiently.

The term "token factory" reflects the unit in which AI usage is measured. Tokens are the chunks of text that models read and generate, and inference, the stage where a trained model answers real user requests, consumes them continuously. Framing the product as a factory signals that Lenovo wants to produce tokens at scale and at a manageable cost.

Why it matters

Inference is where AI models meet real users, and demand for it is rising as more businesses put AI into everyday workflows. For a hardware and infrastructure company like Lenovo, the move is a way to position itself in that part of the AI market rather than only supplying the machines underneath it.

For enterprise customers, the appeal of tiered routing is cost control: paying for large-model capacity only when a task truly requires it. Whether the approach delivers the savings and performance Lenovo is aiming for will depend on how well the system judges which tasks are complex.

The announcement leaves open details such as pricing, supported models and customer availability. Those specifics will determine how competitive the offering is in China's crowded AI infrastructure market.

Apps worth a look

SUMO: the number puzzle

Tap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.

Dialo: learn languages by speaking

Talk to an AI conversation partner; every sentence comes back corrected and explained in your language. 9 languages, A1 to C2.

ediviewer: read EDI files like documents

Open EDIFACT and X12 messages, see every segment explained, convert to JSON/XML, export PDF and Excel. Runs locally, no upload.

giftdetect: the right gift in minutes

Pick who it is for, the occasion and their personality. The search engine reads product descriptions from local shops and ranks what fits. Free, no sign-up.

Tags:lenovoai inferencetoken factorychinaai infrastructure