From Model Breakthroughs to Industry Deployment: 2025 AI Panorama & 2026 Outlook

On November 21, 2025, LinkX held its Annual General Meeting (AGM) in Wuhan.

Over a century ago, Wuhan embraced the spirit of "daring to think and act," launching China's journey into modern industry — in 1890, the nation's first modern steelworks rose on this land, planting the initial seeds for domestic industrial development. In 1911, on this same land, a group of young people under 30 fired their rifles on the city walls, igniting the prelude to a new era.

Today, "historic opportunities for transformative change are once again before us," noted LinkX partner Wang Pu in his opening remarks. The choice of Wuhan carries deep significance: as global technology competition enters a white-hot phase, Chinese enterprises and capital are entering a new window of opportunity to "dare to think and act."

The AGM covered global technology competition, AI technological leaps, industry deployment practices, and the Hong Kong IPO ecosystem. Speakers included Dr. Xue Jing, Special Researcher at the China-US Relations Research Center of Tsinghua University; Zhang Mingchen, Managing Partner of LinkX; Deng Hui, VP of Google China; Yang Yang, Director of Huatai United Securities; and founders from multiple portfolio companies.

Voices from academia, industry, and capital markets converged on a consensus: 2025 is not only an acceleration year for technological change, but a critical starting point for a new cycle in the global technology industry.

Global Competition

US-China Driven, AI-Led: 4 Keys to Winning the Industrial Revolution

Dr. Xue Jing pointed out that global technology competition is entering a China-US dominated era. China and the US are the world's two largest manufacturing nations — in 2024, China accounted for 31.6% and the US for 15.9% of global manufacturing. Among 125 frontier technologies, the two countries together lead in 110.

Behind this competitive landscape is the lock-in of technology, productivity, and comprehensive national power. "Modern technological breakthroughs can no longer be achieved by a handful of scientists alone — they require the support of the entire national system, where capital, talent, resources, and governance capacity are all indispensable." The leading positions of China and the US in AI development are precisely the embodiment of this technological lock-in.

Xue Jing believes that technology competition is the main theme of future great-power rivalry. The battlefield is the next industrial revolution, and AI will play a key role. To win, he emphasized four priorities: fundamental scientific exploration; leveraging existing industrial manufacturing capabilities across diverse application scenarios; consolidating the engineer dividend; and finding new pathways for globalization.

From Breakthrough to Application

LinkX's Frontier Insights and Strategic Positioning in AI

In the new technology competition landscape, China is moving from "application extension" to "source participation." Against this backdrop, LinkX delivered an outstanding 2025 — among its 30 portfolio companies, over 90% completed new funding rounds, 17 have grown into quasi-unicorns in their respective segments, and 2 are advancing toward IPO.

Discussing AI developments in 2025 and the outlook for 2026, LinkX founding partner Zhang Mingchen shared deep insights from 30 experts at the frontier of research, industry, and entrepreneurship:

Four Breakthroughs of 2025

• Strong foundation models and reinforcement learning brought reasoning capability improvements, represented by DeepSeek V3 and R1;
• AI achieved new heights in multimodal understanding and generation, represented by Gemini 3, Sora 2, and Nano Banana;
• AI Coding began scaling into production applications, represented by Claude Code;
• Agents began handling complex tasks.

Five Directions for 2026

• AI Agents are expected to truly replace certain types of human labor in specific domains;
• Embodied intelligence will achieve breakthroughs, with robots demonstrating superhuman motor capabilities;
• Fundamental architecture innovations like Full Attention will further advance AI training and inference capabilities;
• In AI for Science, AI will evolve from a tool to an AI scientist capable of conducting deep research;
• Online learning and other breakthroughs will give models self-evolution capabilities.

Zhang Mingchen noted that 2025 has been a pivotal year of rapid AI development. Whether in model capability improvements, infrastructure growth, application exploration, or capital enthusiasm, all signs indicate AI is transitioning from technological breakthrough to industry application.

As a value amplifier transforming traditional industries, AI is demonstrating the powerful impact of "Roll Up" — globally, traditional industries are opening up vast scenarios and demands, expecting to leverage AI technology to improve overall efficiency and profits, rapidly achieving asset value growth.

This trend aligns closely with LinkX's philosophy. Zhang emphasized: "Our investment in AI is not merely an investment activity — we aim to build an integrated loop of industry, academia, research, and investment, driving the full flow of key elements in the ecosystem, ultimately achieving overall value amplification." To this end, LinkX co-established the T-ONE Innovation Lab with Tsinghua University.

Zhang further stated that from the rapid development of portfolio companies, outstanding entrepreneurs and teams have demonstrated unique speed and direction. The future of AI is not about "whether it will arrive," but "how fast." Entrepreneurs must seize the initiative — run earlier and run steadier.

Exploring the Intelligence Ceiling

AGI Breakthroughs: Foundation Model Capability, Alignment & Reasoning, AI-Native Productivity

Beyond industry and capital-level assessments, understanding AI's evolution from a technical standpoint was equally important at the AGM. Participants shared systematic thinking on the path to AGI, reaching the following consensus.

The AGI inflection point has not yet arrived. The industry remains in a technology-first phase, rather than rapidly pivoting to product orientation. The more important task now is to continue exploring the intelligence ceiling of models. Over the past two years, the industry has experimented extensively with multimodality and voice, but practice has proven that what truly determines the intelligence ceiling remains text and the commonsense knowledge it carries.

From first principles, achieving AGI requires two keys at the model layer: continuously raising the intelligence ceiling and effectively activating existing capabilities. At the application layer, the question is what AI's essence truly is.

First, foundation model capabilities will continue to improve.

To explore the intelligence ceiling, it remains difficult to bypass the scaling path in the short term — including larger-scale commonsense data (real, synthetic, and distilled), larger model parameter sizes, and greater compute investment.

Second, alignment and reasoning become increasingly important for capability activation.

Even if a model possesses knowledge, its capabilities cannot be released without effective activation. Previously, alignment relied mainly on SFT and CoT, but reinforcement learning is gradually becoming the mainstream approach: constructing scenarios and data to let models self-optimize through RL, activating the intelligence ceiling.

The real challenge is that benchmark evaluations are limited, while real-world application scenarios are virtually infinite. With the emergence of numerous Agents, alignment complexity has increased significantly. If past model alignment was like testing language and math knowledge, Agents are closer to real work environments — diverse roles, complex situations, and even a degree of "emotional intelligence."

Third, from the application layer perspective, AI's essence is not creating new capabilities but replacing certain human work capabilities — making the Agent explosion almost inevitable.

Going forward, the application layer needs to clarify which capabilities and positions will be replaced, while being thin enough to quickly adapt to changes in foundation model capabilities.

Meanwhile, beyond the virtual world, corresponding to the Agent explosion is the development of embodied intelligence. Current "robot brains" already possess strong intelligence — the next key step is aligning large model capabilities with robot behavior. As hardware technology matures, embodied intelligence is poised to become a major breakthrough direction for the AI industry.

AI Accelerating Deployment

Frontier Innovators' Thinking and Practice

While discussions on intelligence ceilings and AGI pathways laid the foundation for AI's future, at the practical frontline, AI model capabilities are driving profound transformation across industries. Portfolio companies shared their valuable experiences and forward-looking insights on accelerating AI deployment.

Zijing Zhikang: AI Breaks the Healthcare "Impossible Triangle"

Zijing Zhikang focuses on AI-empowered healthcare services. Founder Liu Yang noted: "Difficulty and cost of medical care are global challenges. The 'impossible triangle' of quality, affordability, and accessibility is gradually being broken by AI."

AI brings revolutionary changes in two dimensions: accelerating the production of high-quality medical resources, and reshaping healthcare service organization through intelligent Agents. Liu emphasized that the evolvability of intelligent agents is core — AI can explore medical decision boundaries at low cost through self-iteration, gradually building a future AI healthcare operating system.

Qidian Lingzhi: Multimodal AI Companionship Reshaping Early Childhood Education

Qidian Lingzhi CEO Bao Ta shared the company's exploration in early childhood AI companionship. Young children don't need simple chat partners — they need multimodal playmates who can join them in pretend play, integrating vision, movement, and language for a "learn through play" education philosophy.

Bao emphasized a "software-first" rapid iteration approach, breaking the traditional linear path of hardware startups, and building continuous feedback loops through user co-creation. The Duoqi AI English learning robot has launched for pre-sale on JD.com with positive market reception.

Wuyun Innovation: Data-Driven, AI-Core Home Intelligent Terminals

Wuyun Innovation focuses on home data storage and intelligent management, proposing the "Everything to NAS" concept. CEO Guo Yanan stated: "In the AI era, data fragmentation has become a pain point. Storage should not be a data graveyard, but the brain of memory."

By locally deploying multimodal large models, Wuyun Innovation achieves lightning-fast workflows, multimodal intelligent search, and zero-wait sharing. The long-term goal is to build a home intelligent terminal, deeply integrating AI with data for true home intelligence.

Zhiying Guangnian: AI-Powered Full Production Crew for Video Creation

Zhiying Guangnian focuses on end-to-end video production optimization, lowering creation barriers so ordinary users can become top content creators. CEO Wang Tiesong noted: "When demand grows exponentially while supply remains linear, capacity bottlenecks become fully exposed."

The company proposes a "four-level" video production model: from basic material processing to narrative art, personalized customization, and content resonance, achieving personalized matching. The company has built information block understanding technology and domain expert models (Editing GPT).

The team's future goal is to provide creators with a closed-loop production system through simplified front-end interaction and comprehensive back-end support.

Long China, Long AI

New Opportunities in Technology Investment

Resonating with AI technology breakthroughs and primary market progress, the secondary market is re-evaluating China's technology growth with real capital. AI has become the core anchor for Hong Kong stocks' new round of valuation and liquidity.

Huatai United Securities Director Yang Yang shared insights on the Hong Kong IPO market. She noted that 2024 saw significant recovery in Hong Kong IPOs, with 87 companies listing. "New economy" companies are becoming the dominant force in Hong Kong's valuation system, with TMT and new energy companies occupying nearly half of listed seats.

The "A+H" dual-listing trend is becoming increasingly clear. From A-share companies planning Hong Kong listings to the Hong Kong market absorbing scarce technology assets, capital from both markets is forming new circulation pathways. "Long China, Long AI" is no longer a slogan but a new cycle validated by liquidity.

She believes technology companies listing in Hong Kong are moving from a "financing tool" to a "strategic lever" for international business. To seize these opportunities, companies should prepare in three ways:

First, proactively plan "H+A" pathways, leveraging Greater Bay Area policy advantages for cross-market value maximization;

Second, fully utilize the 18C policy window by early introduction of qualified "leading senior independent investors" to pass regulatory review with professional capital structures;

Third, present a "technology story of certainty" to mature investors with scarce technology, clear growth paths, and verifiable commercialization metrics, while locking in high-quality cornerstone investors before issuance.

Returning to the scene of LinkX's 2025 AGM, we once again witnessed the power of courage and transformation.

LinkX will continue to discover and support entrepreneurs who "dare to think and act," running with speed and direction in the new acceleration cycle, jointly driving a new round of development and innovation in China's technology industry.

We believe that 2025 is only a new beginning.