Track Hyper | Tianxi AS Major Upgrade: Lenovo Restructures Industry Logic

Author: Zhou Yuan / Wall Street News

On February 25th, at the YOGA AI PC product tasting event held in Mile, Yunnan, Zhang Hua, Senior Vice President of Lenovo Group and General Manager of the China Consumer Business Group, announced that the Lenovo Tianxi Personal Intelligent System (Tianxi AS) has achieved significant upgrades through deep integration with the DeepSeek large model capabilities, and Lenovo's edge-side GenAI (generative AI) capabilities have also seen substantial enhancements.

According to Zhang Hua, the Lenovo Tianxi AS will debut the DeepSeek large model based on personal cloud deployment in early May this year, at which point the hybrid AI deployment of Tianxi AS will be fully realized.

As a result, Lenovo becomes the world's first terminal brand to deploy and operate the DeepSeek large model on the AI PC edge; on the same day, Lenovo launched six new products in the AI PC "YOGA Yuanqi Series" equipped with the Tianxi AS system.

This move signifies that Lenovo is beginning to promote a "dual-wheel drive" of edge-side deployment and hybrid AI architecture, marking a key turning point for Lenovo's AI PC from "cloud dependency" to "edge-cloud collaboration."

Solving the Infeasible Triangle of Large Models

Through distillation technology, Lenovo has achieved localized operation of the DeepSeek-R1 large model with 7 billion parameters on consumer-grade devices.

On the AI PC (laptop) edge, users can perform operations such as document summarization, translation, and writing without calling the cloud, which not only ensures data privacy (such as handling sensitive internal corporate information) but also supports offline scenarios (such as flights and outdoor work).

In addition, Zhang Hua revealed that the DeepSeek large model based on personal cloud deployment will debut in May this year.

Thus, the Tianxi AS can achieve adaptive model invocation through a three-layer architecture of edge-side personal large models, personal cloud large models, and public cloud large models.

First is the edge-side model (7B parameters), primarily handling tasks that require high real-time performance and are privacy-sensitive (such as meeting minutes generation); second, the personal cloud model (planned for release in May this year), relying on users' private servers to handle personalized needs (such as business decision analysis); third, the public cloud model (DeepSeek-R1 full version), which can leverage cloud computing power to complete complex tasks (such as multilingual document translation).

With the architectural innovation of "edge-cloud collaboration + hybrid AI," it not only reduces cloud computing costs (with edge-side processing accounting for over 60%) but also ensures data security through layered encryption technology, solving the "infeasible triangle" of large models: "high performance, low cost, and secure reliability."

"Tianxi AS will deeply integrate DeepSeek across edge-side personal large models, personal cloud large models, and public cloud large models, and hybrid AI deployment will be fully realized," Zhang Hua stated. With the significant upgrade of Tianxi AS capabilities, Lenovo's moto AI phones and AI computers with DeepSeek edge-side personal large models will also be released soon, and native AIoT devices based on Tianxi AS will be accelerated for launch.

The Lenovo Tianxi AS system has already connected to over 1,700 AI applications, covering vertical fields such as healthcare (e.g., electronic medical record analysis), law (contract clause review), and education (personalized learning plan generation) Taking the personal intelligent agent "Lenovo Xiaotian" in Tianxi AS as an example, its L3-level intelligent agent can achieve a code generation error rate reduced to below 2%, comparable to the level of a mid-level engineer, in terms of specialization, personalization, and automation; based on local vector database analysis of user behavior, the recommendation accuracy improves by 40%; and the efficiency of cross-device task scheduling (e.g., mobile phone to PC file relay) increases by three times.

In fact, Lenovo's collaboration with DeepSeek has gone beyond technical integration, forming a closed loop of "model-hardware-application," achieving a resonance in the industrial chain from "single-point breakthrough" to "ecological win-win."

Creating a New Paradigm of Hybrid Architecture

Lenovo has been laying out its technology roadmap for AI PCs for a long time.

Looking back, Lenovo's AI PC roadmap is divided into three stages: the first is from the fourth quarter of 2023 to the first quarter of 2024, which belongs to the "hardware foundation" period.

During this stage, Lenovo launched products such as the ThinkBook Plus Gen5 Hybrid, completing the layout of heterogeneous computing power with CPU+GPU+NPU, but has not yet integrated edge-side model deployment.

The second stage, from the second quarter of 2024 to the first quarter of 2025, can be called the "formation of intelligent agents and ecosystems" period: a landmark event occurred in April 2024 when Lenovo released its first complete AI PC (with the built-in personal intelligent agent Lenovo Xiaotian) and launched the "AI Space" application ecosystem (with over 1,700 applications integrated).

Starting from the second quarter of 2025, a new phase begins. The edge-side deployment of DeepSeek marks the start of this phase.

According to Zhang Hua, in May 2025, Lenovo will release the DeepSeek model based on personal cloud at the TechWorld conference, achieving advanced functions such as enterprise-level data analysis and multimodal content generation.

DeepSeek enables a 70 billion parameter model to run on consumer-grade hardware through model compression (reducing parameter count by 80%, with accuracy loss <5%) and operator optimization (reducing inference latency to the millisecond level), lowering the unit computing cost to 10% of that of cloud models.

Driven by this underlying technology, the business logic of terminal companies has also shifted from "selling devices" to "selling services."

Wall Street Journal noted that Lenovo Group Chairman and CEO Yang Yuanqing, as early as 2017, promoted Lenovo's 3S transformation with his visionary insight and deep market understanding.

The core of this transformation refers to Lenovo's expansion and transformation from a traditional PC-centric business model to an "overall service" direction.

The so-called 3S refers to Smart Devices, Smart Infrastructure, and Smart Services.

Currently, Lenovo's AI PC adopts a "hardware + subscription" model: basic functions are free, while enterprise-level AI tools (such as intelligent contract review) are charged on demand. This model has increased the average transaction value by about 20% and improved user retention by 35% With Huawei (Pangu large model) and Xiaomi (MiLM) following up on edge deployment, the Chinese AI PC market is showing a pattern of "Lenovo leading, multiple strong players emerging."

However, Lenovo still maintains a dominant position due to its first-mover advantage (global PC market share of 24.7%) and its full-stack self-research capabilities (from GPU adaptation to system optimization).

Lenovo categorizes the AI levels of personal intelligent agents into five levels: L1 Assistive, L2 Assistant, L3 Collaborative, L4 Expert, and L5 Autonomous.

Lenovo's Xiaotian is expected to reach L3 level (autonomous task execution) this year, possessing specialized, personalized, and automated capabilities, but achieving L4 level (predictive decision-making) still requires further breakthroughs in three areas: data closed-loop, multimodal fusion, and energy consumption optimization.

Among them, the data closed-loop needs to expand the user behavior dataset (currently only covering 10% of office scenarios); in terms of multimodal fusion, the edge-side image/audio models have not yet achieved deep coupling with text models; energy consumption optimization must continue, as device endurance faces significant pressure when running high-load large models.

Lenovo plans to improve the energy efficiency ratio of edge-side models by over 50% by the end of 2025 through quantum computing chip simulation (in collaboration with Benyuan Quantum) and upgrades to neuromorphic computing architecture.

Lenovo's deployment of the DeepSeek large model on the edge is not only a technological breakthrough but also creates a new industrial paradigm of "edge-cloud-personal cloud" hybrid architecture.

With the dual dividends of accelerated AI terminal penetration and deepening domestic substitution, Lenovo is expected to leverage its full-stack layout advantage to be the first to open the trillion-level AI PC market.

From an industry perspective, terminal companies, including Lenovo, will face three major challenges in the future: the speed of killer application incubation, control of edge-side computing cost, and the ability to co-build an open ecosystem.

To what extent and at what speed these three challenges are met may define the competitive landscape of terminal intelligence in the next decade