Lenovo Launches 'Hybrid Token Factory' in Zhangjiakou: Green Power In, Tokens Out, 70% Cost Cut

type

Post

status

Published

date

Nov 14, 2026

slug

summary

Lenovo Launches 'Hybrid Token Factory' in Zhangjiakou: Green Power In, Tokens Out, 70% Cost Cut

As AI adoption expands globally, enterprises are facing a paradox: while computational power availability increases, the cost of running AI agents is rising sharply. A recent report from the *Economic Observer* noted that a mid-sized bank spent nearly RMB 9 million on tokens in the first half of the year, equating to 8% of its total IT budget for the previous year. Consequently, organizations are becoming more cautious about "burning tokens," seeking a balance between performance and cost efficiency. Against this backdrop, Lenovo Group has introduced a new industrial model designed to redefine how AI inference is scaled and managed.

From Green Energy to Industrialized Token Production

On November 10, Lenovo Group officially unveiled its "Lenovo Wanquan Hybrid Token Factory" brand and signed a cooperation agreement with the Zhangjiakou Municipal Government in Hebei Province. The project aims to establish the first "Hybrid Token Mother Factory," leveraging the region's abundant renewable energy resources. Zhangjiakou is recognized as China's top city for wind and solar green power, with renewable energy capacity exceeding 45 million kilowatts by the end of 2025, accounting for nearly 90% of the local total power installation.

The initial phase of the factory is planned to include a 15,000 P heterogeneous computing cluster and an intelligent computing center with a capacity of approximately 100 MW. Once operational, the facility is projected to produce 2000 quadrillion tokens annually, with a peak output capability of 200 million tokens per second. Lenovo plans to expand this scale by more than 10 times by 2030. This "Green Power In, Token Out" approach shifts the focus from merely supplying hardware to industrializing the production of high-value, cost-effective AI services.

A Three-Dimensional Hybrid Supply Model

Unlike traditional data centers that focus solely on maximizing raw compute power, Lenovo's "Hybrid" model emphasizes aligning computational resources with specific business needs. The company identifies three core dimensions to this approach:

1. Source Hybridization: Combining private, self-generated tokens with authorized public cloud services to balance data sovereignty and cost efficiency.

2. Edge-Cloud Coordination: Routing lightweight tasks to edge devices, latency-sensitive tasks to nearby nodes, and elastic demands to the cloud.

3. Model Ecosystem Mixing: Dynamically selecting between open-source and proprietary models based on task quality requirements and cost constraints.

To manage this complexity, Lenovo developed the Token Engine, which utilizes dynamic co-location of models and computing resources, task-level resource pooling, and intelligent model routing. This is supported by the xCloud platform, which provides unified management for both private and public cloud resources, ensuring that compute supply follows business demand rather than being dictated by static hardware configurations.

Optimizing Efficiency and Security at Every Layer

Lenovo applies a three-level optimization strategy to maximize value. First, the Wentian Heterogeneous Intelligent Computing Solution improves compute efficiency by over 30% through optimized cluster communication and operator tuning. Second, the facility focuses on full-lifecycle cost control, aiming for a Power Usage Effectiveness (PUE) of below 1.1 through modular prefabrication and liquid cooling. Third, the Token Hub dynamically allocates resources so that simple tasks do not consume expensive high-end models.

Security remains a critical component. Lenovo implements three barriers: model safety via whitelist certification, content security through the Token Trust capability to intercept sensitive data, and privacy protection using Trusted Execution Environments (TEE). This ensures that sensitive data remains "usable but invisible" to the platform operator, complying with strict data governance standards.

The "Effective Token" Metric

Lenovo argues that throughput and unit cost are insufficient metrics for AI value. The company introduces the concept of the "Effective Token," which measures the quality of task completion. An AI system that requires multiple iterations and re-inferences to achieve a result generates more tokens but less actual value than one that delivers a correct answer immediately. By focusing on "Multi, Fast, Good, Cheap, and Green," the factory aims to prove its efficiency through internal practice, where Lenovo has achieved approximately 70% comprehensive cost optimization. The ultimate goal is to deliver 99.9% token "good product rate" while ensuring task precision and green sustainability, transforming AI infrastructure from a cost center into a streamlined industrial production line.

Source: QbitAI, https://www.qbitai.com/2026/11/510094.html