On May 28, Tianjin. Lenovo's exhibit list at the 2026 World Intelligent Industry Expo is quite interesting: 1,000 intelligent agents running permanently, the cost of one million tokens is less than 1 yuan, DeepSeek's inference throughput has doubled, and the efficiency of wind resistance simulation for Geely Automobiles has increased by 28 times.
This is not the kind of "AI fully empowers enterprise digital transformation" rhetoric commonly found in press releases. Lenovo has laid out concrete specifics this time—what architecture is used, how much it costs, and what results have been achieved. For those involved in enterprise digitalization, these numbers are far more useful than descriptions like "strategic layout."
What Lobster Lake Talks About
Lenovo's main enterprise solution this time is called "Lobster Lake," which is essentially a privatized deployment of an intelligent agent cluster solution that can support up to 1,000 "lobsters" (intelligent agents) online simultaneously.
What caught my attention most was not the quantity, but the cost structure: the operating cost per million high-quality tokens is less than 1 yuan. This figure was almost impossible a year ago, but now it is a price that can be promised at the product level.
Agent cluster scale: 1000 private deployments, resident operation cost
The matter of private deployment deserves a separate discussion. Over the past two years, most enterprises have used AI through cloud APIs, paying based on call volume. As they continued using it, they found the bills were higher than expected, and more troubling were data security and compliance issues—in industries such as export processing, semiconductors, and finance, much of the data simply cannot leave the internal network.
The positioning of Lobster Lake is to lower this threshold: private deployment, data staying within enterprise boundaries, and self-owned computing power. This logic is opposite to the approach of encouraging public cloud usage in previous years, but it is more aligned with the actual procurement preferences of China's manufacturing industry and leading enterprises.
Hybrid AI is not a new term
In his opening speech, Yang Yuanqing used the term "hybrid AI," a phrase Lenovo has employed for at least two years, but this time the expression was somewhat different. He said the essence of global economic competition is "a comprehensive contest of AI + real economy ecosystem capabilities," with the key word being "ecosystem"—not technology, not computing power, but ecosystem.
Innovation in the era of artificial intelligence is no longer a breakthrough in a single product or technology, but rather systematic innovation—where technology, scenarios, ecosystems, standards, and governance are all indispensable.
I think this judgment is correct. In the past few years, one core reason for the high failure rate of enterprise AI projects has been focusing only on technology while ignoring scenarios and governance. The technology itself is actually no longer the biggest pitfall; the biggest pitfalls are the two issues of "which scenario to start with" and "who takes responsibility when problems arise."
Lenovo also announced two specific investments in Tianjin: the general server production line will begin mass production in September 2026, and the new-generation AI computing product R&D and manufacturing center will start mass production in the autumn of 2027, both located at the Lenovo Smart Innovation Service Industrial Park in Tianjin. These two timelines are quite concrete, not vague promises of "sometime in the future."
What the Qingtian Claw ran out
Lenovo Wanquan Heterogeneous Intelligent Computing Platform 4.0 is the underlying computing power support for the Qingtian AI platform. This exhibition showcased two specific cases:
| Scene | Result | Instructions |
|---|---|---|
| DeepSeek inference throughput | Double | Compared to the standard configuration, the processing capacity is doubled under the same computing power. |
| Geely Automobile Wind Resistance Simulation | Efficiency improved by 28 times | HPC high-performance computing scenarios, typical manufacturing demands |
The wind resistance simulation case is quite representative. In automotive development, CFD (Computational Fluid Dynamics) simulation has always been a time-consuming process, with a full simulation potentially taking several days. A 28-fold increase in efficiency means that work that originally took a week can now be completed within six hours. This is not just "a little faster"; it changes the entire pace of design iteration.
In the vertical scenario of intelligent manufacturing, Lenovo also demonstrated three industry-specific intelligent agents: assisted design, intelligent quality inspection, and Rubik's Cube intelligent customer service, as well as the urban super intelligent agent (a 1+N architecture already deployed in multiple cities such as Shanghai and Chongqing).
How should enterprises view these things
There are several issues worth thinking about carefully.
First, whether a privatized intelligent agent cluster is suitable for oneself. The logic of Lobster Lake is suitable for enterprises with high data sensitivity and existing IT infrastructure, such as leading manufacturing companies, financial institutions, and state-owned enterprises. For small and medium-sized enterprises, maintenance costs and configuration complexity may instead become a burden.
Second, the premise behind the figure of less than 1 yuan per million tokens. Lenovo provided self-tested data; actual deployment factors such as token usage, call frequency, and operational maintenance costs will all affect the real TCO (Total Cost of Ownership). This figure is a reference benchmark, not a final destination.
Third, the reproducibility of results such as 28x wind resistance simulation. HPC requirements vary greatly across industries; the CFD needs of the automotive industry are not on the same scale as those of semiconductor EDA simulation. The range of industries in which Lenovo can replicate this result is narrower than what is stated in the exhibition PPT.
However, none of this affects a basic judgment: the technical threshold for deploying enterprise-level intelligent agents is rapidly decreasing, and privatization solutions are becoming increasingly feasible in terms of both price and capability. The question in 2026 will no longer be "can it be done," but rather "who will lead and which scenario to start with."
For enterprises deploying ERPs like Odoo, the path to introducing intelligent agents is actually clearer than outsiders imagine—the ERP itself is the system with the densest enterprise data. Starting with procurement assistants, inventory anomaly alerts, and customer service automation, running privatized small models on the internal network precisely bypasses data security concerns.
