At a recent AI investment summit held in Beijing, centered on the theme of "Certainty Opportunities in the AI Infrastructure Era," Song Chunyu, Vice President of Lenovo Group, Chief Investment Officer and Senior Partner of Lenovo Capital, shared his insights on the AI industry cycle, full-stack investment strategy, and next-generation AI technology trends.

Song noted that generative AI remains in the early stages of the fourth industrial revolution. Current industry development is still predominantly in the first wave of AI infrastructure construction, while the true penetration of AI into every industry and sector has yet to begin. Meanwhile, as foundation models evolve rapidly, new questions are emerging as global focal points for the AI industry and investment community, including the capability boundaries between large models and Agents, and whether independent frontier laboratories can achieve new technological breakthroughs.

Decade of Sci-Tech Investment Anchored on Next-Gen AI

This year marks the tenth anniversary of Lenovo Capital. Song revealed that the firm has invested in 330 tech companies cumulatively, with 26 having achieved IPOs. AI stands as the largest heavyweight track for Lenovo Capital, with investments in over 150 companies, making it the best-performing sector over the past decade in terms of project count, investment amount, and returns.

As the global tech industry fund under Lenovo Group, Lenovo Capital positions itself as the "forward-looking watchtower" for the group's future technology pathways. In AI, the firm adopts a full-stack investment strategy, spanning from chip and semiconductor infrastructure and underlying materials, to foundation models and frontier models, then extending to digital-world agents, as well as physical-world applications like autonomous driving and embodied intelligent robots.

Song explained that on the computing power front, Lenovo Capital joined Cambricon's Series A round in 2017, followed by four consecutive rounds through to its strategic placement, and was among the earliest investors to complete a full lineup of four AI computing power companies on the STAR Market. In the foundation model and frontier model space, the firm has invested in companies such as Zhipu AI, StepFun, Kling AI, and AMI, which focuses on world models.

Looking toward next-generation AI, Lenovo Capital is extending its investments to deeper underlying technologies. Song noted that in AI computing, investments have moved to the material level, exploring cutting-edge directions like two-dimensional semiconductors and atomic-scale chip and memory design. At the system level, the firm is betting on all-optical interconnect technology to solve high-throughput connectivity challenges in future AI infrastructure. On the model front, key focus areas include AI self-evolution, AI for Science, and next-generation world models.

Song said one crucial goal for future foundation models is to enable AI to evolve into "high-dimensional scientists." A key mission behind these frontier investments is to provide technological support for Lenovo Group's core business AI transformation through early-stage tech deployment.

AI Still in Its Infancy, Industry-Wide Adoption Is the Real Battleground

In Song's view, as AI technology iterates rapidly, investors need to rethink the shifting industry boundaries between foundation models and application layers. One significant question is the capability boundary between large models and Agents.

Song mentioned that a popular discussion among Silicon Valley VCs and entrepreneurs is that "foundation models are eating Agents." As foundation models continuously strengthen their long-horizon reasoning and Agentic capabilities, some functions traditionally handled by independent agents may gradually become covered by foundation models. However, this does not mean Agents lose their investment value.

What concerns Song more is how far foundation model capabilities will ultimately extend, and whether a relatively clear capability boundary can form between foundation models and Agents. He believes this question is vital for assessing how AI truly enters every industry and for identifying AI application investment opportunities.

Another notable trend is that more top researchers are leaving large tech companies to establish independent frontier laboratories, exploring frontier models, AI for Science, and next-generation world models. Whether these new labs can grow into independent technological forces or will eventually be absorbed and integrated by large tech companies is becoming a key question for the global industry and investment community.

Song also pointed out that some voices in the market suggest the foundation model landscape is already largely settled, but this judgment remains to be seen. He does not rule out the possibility that emerging frontier labs could achieve technological breakthroughs, leading to differentiated foundation model experiences.

Addressing market debates around Nvidia's market value and the recent rally in storage stocks, Song offered his assessment: the AI industry is just beginning. He noted that generative AI based on the Transformer architecture has only been around for less than four years since breaking out in November 2022. Measured on the timescale of industrial revolutions, it is still at a very early stage.

"What everyone is seeing now is infrastructure—that is the first wave. But its actual entry into every industry has not truly begun yet," Song said. Taking coding as an example, AI has already started to penetrate programmers' work scenarios, yet a vast number of industries still await deeper integration. From the initial investment in AI infrastructure, to models and Agents genuinely entering corporate production processes, and then to AI forming closed commercial loops across more industries, there remains a long path for value release.