2026 Zhongguancun Forum Annual Conference | Hard Technology Supports the Rise of Unicorns

（Source: Beijing Business Daily）

Even though this is a conference focused on the development of “unicorns,” Zhou Hongyi still decided to start with that “lobster” that has “two horns.”

The lobster is OpenClaw, an open-source AI agent framework that went viral across the internet at the start of 2026. It is named after its red lobster icon. It takes AI from “chat-style interactions” to “autonomously executing tasks,” able to simulate operating a computer and complete complex workflows across multiple software applications. It quickly broke out beyond the tech industry, becoming a hallmark product of the commercial deployment of AI agents.

On March 29, at the site of the 2026 Zhongguancun Forum Annual Conference Global Unicorn Enterprises Conference, this 360 Group founder, Chairman and CEO, together with other participating guests, using topics like the viral lobster, and through experience and data, told Beijing Business Daily and other media the key logic behind high-quality growth of unicorns—hard technology.

At the same time, this conference also released the China Unicorn Enterprise Development Report (2026). The term “hard technology” became the core defining trait of China’s unicorn enterprises. With coordinated efforts in policy, ecosystem, and services, the industry is being pushed into a new cycle of high-quality development.

1. Unicorns anchor to the hard-technology cycle

As the core defining trait of China’s unicorn enterprises, hard technology is, through multi-dimensional coordination—policy guidance, ecosystem building, and service assurance—pushing the industry away from old modes of telling stories and hype, and into a new cycle of high-quality development.

Xu Hanhong, Chairman of the Board of the Zhongguancun Unicorn Enterprise Development Alliance, disclosed on site the core development data for China’s unicorn enterprises in 2025: the total number of unicorn enterprises nationwide reached 416, with a combined total valuation of $1.6 trillion. Both the number and valuations refreshed records from the past three years.

Among them, the artificial intelligence track—with 69 companies and a total valuation of $638 billion—has become the core engine of the industry. Hard-tech companies account for 322 of the total, making up 77.4% of all unicorns.

At the technical level, unicorn enterprises build competitive barriers based on core patents and R&D. The average number of patent applications for hard-tech unicorns is 2.2 times that of non-hard-tech enterprises. The core of founding teams is largely composed of university research strengths and technical experts from major tech companies.

As capital allocation becomes increasingly rational, it is also becoming increasingly focused. The financing market shows a “higher volume, lower total amount” pattern: in 2025 there were 201 financing events, but the total amount was only $18.19 billion. Later-stage rounds became the main force in financing, while the share of funding in early stages (Series A and earlier) was only 8%. In terms of track distribution, the four major hard-tech tracks together account for nearly 80% of financing scale, with artificial intelligence ranking first. In terms of investor participation, the share of market-driven VC/PE participating events reached 86%, with state-owned capital/guidance funds following closely at 69% participation. More than half of hard-tech unicorns received early pre-planning from state capital in the “seed stage,” becoming a core force in nurturing enterprises.

Yue Deyu, Director of the Beijing High-Tech Industry Development Promotion Center, also highlighted Beijing’s unicorn cultivation practices. In 2025, Beijing ranked first nationwide for six consecutive years with 116 unicorn enterprises and a total valuation of $741.9 billion. Within the region, the share of hard-tech unicorn enterprises exceeds 75%. Tracks such as artificial intelligence, health & pharmaceuticals, and intelligent equipment are core strengths. Relying on the “Unicorn 10 Articles” special policy, a concierge-style enterprise service package, benchmark incubators, and characteristic industrial parks, Beijing has built a cultivation system covering the entire lifecycle of enterprises. At the same time, it keeps a close eye on frontier technology tracks such as scientific intelligence, embodied intelligence, brain-computer interfaces, synthetic biology, and low-altitude technologies, continuously stockpiling high-quality potential unicorn enterprises.

2. Unicorns come from technology evolution

In the wave of unicorn growth dominated by hard technology, AI agents are one of the representative breakthrough directions.

Zhou Hongyi defined 2026 as the “Lobster Year” when agents will fully break into the mainstream. The recently booming OpenClaw agent, though still in a transitional development stage and the technology itself still has many immaturities, has completed an extremely important science-communication mission—just like the DeepSeek boom more than a year ago, which enabled traditional entrepreneurs and the general public to truly understand the core value of agents. After industry discussion for a long time, “moving large models from chatting to doing work” finally got its first tangible carrier.

This 360 Group founder, Chairman and CEO, sharing with media including Beijing Business Daily, said that the lobster’s biggest feature is actually “having the nerve.” To achieve its purpose (such as instructions), it will use any means—even “leaving no stone unturned.” It dares to break from conventional AI logic and “brutally” deconstruct task workflows. It is precisely this innovative trait that is driving structural change across the entire industry.

Based on this, the information infrastructure for agents may face reconstruction. Most of today’s software and internet services are designed for human-computer interaction, and their operating logic does not fit how agents are used. For example, for an agent to log into an e-commerce website, it still needs to take screenshots one by one and click one by one. A dedicated browser built specifically for agents, a dedicated email address, a search engine, and an e-commerce platform may instead become entirely new entrepreneurial tracks, bringing potential disruptive opportunities.

Meanwhile, the software industry may break away from the traditional SaaS model and transform toward the underlying, atomic, modular direction—turning into “skill packs” that agents can flexibly call. Software delivery would also shift from selling finished products to assembling on demand. Zhou Hongyi used a vivid analogy to explain it: the traditional software industry is like a pre-made meal manufacturer—after making everything at the source, it is simply assembled for the customer. In the future, agents will be like “chefs,” flexibly combining ingredients according to customer needs and optimizing—or even rewriting—the service logic.

Of course, although agents are great, they are not perfect at present. Zhou Hongyi acknowledged that agents still have issues such as a high learning curve, insufficient runtime stability, and data security risks. It is like Sun Wukong in “The Monkey King causing havoc in Heaven”—he can complete tasks efficiently, but he could also disrupt the workflows enterprises have already set. Especially in enterprise-level vertical scenarios, customized agents and agent security protection solutions still have many gaps. These technical pain points that urgently need to be addressed can also be new directions for the industry and newly emerging unicorns to attempt breakthroughs.

3. AI deployment needs to “avoid the virtual and go for the practical”

Following the application wave brought by the lobster, when agents become the mainstream form for AI deployment, how enterprises avoid industry traps and make AI truly empower business becomes a hands-on problem that unicorn enterprises need to focus on. In his sharing, Zhou Hongyi also joked and predicted, “In the future, if you want to raise a lobster to help you bargain on e-commerce, merchants can also raise a bunch of lobsters to bargain back with your lobsters.” Behind this is also a reflection that, during AI deployment, it is necessary to balance innovation with the core need for practicality.

Feishu CEO Xie Xin pinpointed the common “illness” of the current AI product market with a real usage experience for media including Beijing Business Daily. He had tried late at night to use another company’s AI product to send messages on his behalf. The product completed most basic operations—including opening the software and entering text—very much matching what people would call “good propaganda effects,” but it went wrong in understanding the intent behind the final step—“having lunch tomorrow noon.” At 12 a.m. in the middle of the night, he even accidentally called a video call invitation to an acquaintance of the opposite sex he hadn’t contacted in a long time.

Facing frequent, similar “crash” scenarios, Xie Xin believes that most AI products on the market today—whether from domestic and overseas big tech companies or startups—only emphasize the functions they claim to have, but never mention real-world usage outcomes. Yet AI products inherently have characteristics of being immature and unstable. Even though the industry calls for labeling product maturity levels, few companies respond. The current state of confusion creates significant challenges for enterprises when choosing AI tools.

He suggests that when enterprises choose AI tools, they cannot rely solely on vendor copy and demo effects. Looking at real feedback from real users in real scenarios is far more reference-worthy than surface-level promotion.

Regarding the most troublesome AI budget issue for enterprises, Xie Xin proposed that forward-looking teams can try to break away from the past mindset of calculating short-term ROI (return on investment). They don’t have to be fixated on short-term input-output returns. After all, at this stage, once you total projects like electricity and compute, the ROI truly does not match proportionally.

But if you think from another angle and calculate the “cost of no action”—what an enterprise will lose or give up one year from now or three years from now if the team doesn’t invest in AI and doesn’t embrace industry upgrading. The iteration speed of AI technology is extremely fast, and nobody can predict how the industry will change a month from now. Rather than obsess over short-term gains, it may be better to treat agents as digital employees for long-term planning, which perhaps better matches the pace of the AI era.

Beijing Business Daily reporter Tao Feng Wang Tianyi

大量资讯、精准解读，尽在新浪财经APP