From Cautious Observer to Full-Stack AI Challenger: Why Lenovo Bets Big on S...

Lenovo Venture Partner Song Chunyu

AsianFin— Chinese tech giant Lenovo is stepping confidently into the AI battleground, no longer content to watch from the sidelines.

At its Tech World 2025 conference, Chairman and CEO Yang Yuanqing introduced the company's boldest vision yet: the super intelligent agent, an evolution of traditional AI assistants equipped with perception, cognition, autonomy — and the promise to redefine human-machine interaction.

More than just futuristic talk, Lenovo backed its ambitions with action. Its venture capital arm, Lenovo Venture, unveiled a sweeping investment record: over RMB 1 billion deployed across more than 20 companies since it announced a RMB 2 billion commitment to the AI terminal ecosystem last year.

The portfolio spans from foundational model developers to application-layer innovators, reflecting a deliberate strategy to build what Song Chunyu, CIO and Senior Partner at Lenovo Venture, calls full-stack AI capabilities.

Lenovo's newfound momentum contrasts sharply with its earlier restraint during China's large model frenzy. Asked last year whether Lenovo had invested in any foundational model firms, Song calmly replied: "Not yet. We need time to see clearly who can truly pursue AGI."

Today, he's ready to make the call. "The AI industry is at a turning point," he told TMTPost Focus in an exclusive interview. "Open-source large models will capture 80% of the market, while closed-source models will drop to 20%."

He credits the shift to companies like DeepSeek, whose open-source models have slashed deployment costs while improving interoperability — enabling agents to collaborate and perform increasingly complex tasks. Song believes the next AI boom will center on these intelligent agents, both vertical and general-purpose.

Song, a seasoned tech investor and former engineer, has a reputation for shunning hype in favor of long-term fundamentals. His investment philosophy is rooted in three pillars: elite talent density, founder influence, and capital-raising strength.

"These teams need to be able to raise RMB 10 billion at an early stage," he said. "Without that, you can't afford the computing power required to train competitive models."

Among Lenovo Venture's portfolio: chipmakers like Cambricon, Hygon, and Moore Threads; AI infrastructure platforms such as OpenCSG and Zhipu AI; and application developers including SmartMore and YouMind. Even Europe's Mistral AI makes the list, reflecting Lenovo's global outlook.

Song calls this a multi-layered, high-coverage approach — building not just AI models or hardware, but a vertically integrated AI ecosystem. At the heart of it all is Lenovo's own super intelligent agent platform, which integrates tools, data, and compute modules to enable seamless deployment across Lenovo's AI PCs and other terminals.

Song believes AI's most profound real-world impact will come from embodied intelligence — intelligent systems with physical interaction capabilities, like autonomous vehicles and service robots.

He predicts that "generalization in limited and vertical scenarios" will arrive within two years, especially in industrial settings like factories or pharmacies. But he cautions against overhyping humanoid robots. "The future will be multi-species — wheeled, multi-armed, or hybrid forms. It won't all be humanoid."

He also sees a pivotal moment coming in autonomous driving: "Cars are essentially four-wheeled robots. Embodied intelligence will hit commercial viability here first."

Looking ahead, Lenovo Venture aims to deepen its stake in what Song calls "the next AI iPhone moment" — a reference to the transformative impact of consumer-friendly innovation. For Lenovo, this centers on super intelligent agents and AI PCs — devices capable of hosting and evolving with AI workloads, tailored for individual users.

Already, Lenovo's ecosystem partners are plugging into this vision. SmartMore powers its text-to-image capabilities, YouMind supports creative content generation, and OpenCSG enables user-friendly model tuning. These collaborations, Song said, are embedded directly into Lenovo's software architecture — accelerating feedback loops, user adoption, and performance optimization.

To further this strategy, Lenovo offers two engagement paths: direct integration for portfolio companies and an open "Xingchen Program" for external startups seeking to join the ecosystem.

While Song estimates a six-month lag between China and the U.S. on AI algorithms, he believes China's technical talent remains a formidable force. "In algorithms, nothing can stop the innovation of Chinese technologists," he said.

For an industry long dominated by U.S. giants, Lenovo's full-stack AI vision — methodical, global, and ecosystem-driven — could mark a new chapter in China's AI ascent.

"No rush, and nothing will be missed," Song concluded — a mantra that may well define Lenovo's transformation from careful observer to full-stack AI contender.