With a valuation of hundreds of billions and losses of tens of billions, why do investors still line up to grab shares?

In May, the AI track is witnessing an unprecedented capital feast.

Yuezhi Anmian is close to finalizing a $2 billion financing round, with its post - investment valuation exceeding $20 billion. Jieyue Xingchen is about to conclude a nearly $2.5 billion financing round, having completed the removal of its red - chip structure and is just one step away from a Hong Kong IPO. DeepSeek's valuation has skyrocketed. It jumped directly from $10 billion at the beginning of the year to over $50 billion, and it plans to raise up to 50 billion RMB in financing.

In just one week, the total financing of three leading large - model companies approached 100 billion RMB, completely igniting the enthusiasm of the entire venture capital circle.

All this madness can be traced back to the listings of Zhipu AI and MiniMax four months ago.

Since its listing on the Hong Kong Stock Exchange on January 8th, Zhipu's stock price has been rising steadily. Its market value increased by over HK$50 billion in just one week. As of the close on May 18th, Zhipu's latest market value exceeded HK$500 billion, while MiniMax's latest market value also exceeded HK$200 billion.

As the stock prices soared, early investors reaped dozens of times of excess book returns, and many core employees of several companies achieved financial freedom.

Alibaba's financial report data more intuitively confirms the value of this wealth - creating feast. On May 13th, Alibaba released its 2026 fiscal year financial report. Its annual interest income and net investment income increased by 322% year - on - year, reaching 87.512 billion RMB, hitting a new high for this type of income since 2022.

Relying on its layout of holding about 13.66% of MiniMax's shares and the Alibaba Group and Ant Group together holding about 6% of Zhipu AI's shares, AI track investment has become an important driving force for the significant increase in Alibaba's income.

This wealth - creating event, like a stone thrown into a lake, has completely activated the entire AI primary market. As a result, investors are flocking in frantically, for fear of missing the next "wealth - creating target" that will change the industry landscape.

Financing for large AI models: Investors are queuing up to enter

The joke "queuing up to wait for Liang Wenfeng to answer the phone" that circulates in the investment circle reflects the madness of the large AI model track.

DeepSeek's financing plan of over 50 billion RMB instantly ignited the enthusiasm of the entire industry, but it has left countless investors in anxiety as they can "see but not touch" it.

An investor in the hard - tech field told Tech Planet that his institution has been closely monitoring DeepSeek's developments, but the difficulty of getting involved is far beyond expectations. On the one hand, the financing threshold is high; on the other hand, there are too many investors competing for limited opportunities. "It's not the companies seeking capital. Investors are chasing after Liang Wenfeng, and he gets to make the choices throughout the process."

This first - round external financing since DeepSeek's establishment is enough to support a valuation myth with just a series of keywords: a financing target of 50 billion RMB, a valuation of 350 billion RMB, the founder's personal investment of 20 billion RMB, the leading investment by the National Large - scale Fund, and the rumors of giants like Alibaba and Tencent rushing to get involved, making it a top - tier target that is "out of reach" in the primary market.

The story of companies having the initiative to choose capital is not unique to DeepSeek. The above - mentioned investor said that the other two companies with high financing enthusiasm in the large - model track have also stepped out of the stage of simply needing money and entered the stage of carefully selecting capital providers.

Public information shows that Jieyue Xingchen is about to finalize a nearly $2.5 billion financing round. Its shareholder lineup includes consumer electronics industry giants such as Huaqin Technology, Longcheer Technology, OmniVision Technologies, and ZTE Corporation. It has also received a strategic investment from the "Hong Kong version of Temasek", Hong Kong Investment Corporation (HKIC).

This round of financing for Yuezhi Anmian is led by Meituan Longzhu, with a single investment of over $200 million. Institutions such as China Mobile and CPE Yuanfeng are following up. With the support of state - owned and industrial capital, institutions that have never been involved before have no hope of getting in. "Moreover, this is a round close to the IPO, and the investment shares are already in short supply," the above - mentioned investor added.

Besides the popular companies, some investment institutions are also turning their attention to large models in vertical fields and physical applications. A VC investor said that such startup projects have a lower capital entry threshold, clear implementation scenarios, and smoother commercialization paths, so there is no need to compete for the scarce shares of leading companies.

Meanwhile, senior entrepreneurs with mature industry experience have also become high - quality targets that capital is eager to pursue.

Lin Junyang, the former technical leader of Alibaba's Qianwen large model, recently officially entered the startup scene, focusing on the fields related to world models and embodied brains.

According to public information, he has now assembled a core team composed of technical talents from ByteDance, Tencent, and overseas. The project has also officially started seeking financing, and its current valuation is about $2 billion. First - tier investment institutions such as Sequoia Capital China and Gao Rong Venture Capital have actively initiated contact and negotiation.

Rapid revenue growth

Around 2024, the capital market generally had a cautious and pessimistic attitude towards large - model startup companies. At that time, the industry consensus was that these companies had insufficient revenue - generating capabilities, were deeply in long - term losses, and were constantly facing business pressure from large Internet companies. The investment return was dim, and the risk of capital investment was very high.

However, in just over a year, the market expectations have reversed. Leading companies such as Zhipu, MiniMax, Yuezhi Anmian, and Jieyue Xingchen have rapidly emerged, completely getting rid of the positioning of ordinary startups and becoming leading unicorns pursued by capital. Some companies have even entered the ranks of companies with a market value of over 100 billion.

A large - model practitioner attributes this change to "attention". In his view, AI has become the core driving force for industrial and economic development, even though most companies in the industry have not yet achieved large - scale profitability.

The investment of industry giants has made this trend even more obvious. Alibaba CEO Wu Yongming summarized at the earnings conference: "The first quarter of 2026 was a quarter for Alibaba where sowing far exceeded harvesting." He firmly believes that in the era when AI reshapes everything, today's investment will determine tomorrow's height.

Tencent also revealed the real development situation of the industry. Ma Huateng said at the shareholders' meeting when talking about the group's AI layout: "A year ago, we thought we were on the boat, but later we found that the boat was leaking. Now we feel like we're on it, but we can't sit down. We still hope the boat can go faster."

However, an investor said bluntly that the current capital layout in the AI field is more inclined to practical investment. Simply telling stories no longer works. Capital's focus has finally returned to physical products, and technological breakthroughs have become the core confidence for large - model companies to gain double recognition from the market and capital.

Now, the revenues of leading large - model startup companies such as Zhipu, MiniMax, and Yuezhi Anmian have all witnessed rapid growth.

Among them, Yuezhi Anmian is developing particularly rapidly. In mid - 2025, its subsidiary Kimi officially launched K2, the first domestic open - source large - scale model with a trillion - parameter base, and was the first to implement the practical capabilities of the second - order optimizer Muon in actual tests.

After successfully building the technological barrier, the company's commercialization process has accelerated. At the end of 2025, Yang Zhilin, the founder of Yuezhi Anmian, revealed in an internal letter that the company had a cash flow of over 10 billion RMB. From September to November 2025, the monthly average month - on - month growth rate of Kimi's paid users at home and abroad exceeded 170%. After the launch of K2 Thinking, it further drove business revenue growth, and its overseas API business revenue soared four times year - on - year.

In 2026, with the rapid popularity of the open - source AI agent framework OpenClaw, Kimi quickly completed the integration and implementation of technology, and the commercial results were very impressive. Since the end of January this year, the revenue generated by Kimi in just twenty days exceeded its total revenue for the whole of 2025.

According to Wang Xinyu, a partner of Meituan Longzhu, Kimi's ARR (Annual Recurring Revenue) exceeded $100 million in March and increased to $200 million in April.

Different from Kimi, which follows the commercialization and monetization route, DeepSeek adheres to full - route open - source and has no clear explicit revenue model for now.

However, an algorithm practitioner in the industry said that DeepSeek's advantage lies in the market share brought by open - source. Extreme open - source has lowered the threshold for developers to use, deploy, and conduct secondary development. Coupled with the advantages of free access and rich versions, DeepSeek has firmly established itself in the domestic market and is rapidly spreading in many developing regions around the world.

Valued at hundreds of billions, but in billions of losses

As of the close on May 18th, Zhipu's stock price was reported at HK$1,137, and its total market value exceeded HK$500 billion. Its market value soared nearly ten times in just four months. MiniMax's market value also once exceeded HK$250 billion recently, and the secondary market's enthusiasm for it is high.

However, behind the sky - high market values, the two leading companies still haven't escaped the dilemma of losses, and their business fundamentals have not kept up with the growth rate of their valuations.

The financial report data shows that in 2025, Zhipu's revenue increased steadily, with an annual total revenue of 724 million RMB, a year - on - year increase of 131.9%. However, the profit - making pressure continued to increase, and the annual net loss attributable to the parent company was 4.698 billion RMB, an increase compared to the loss of 2.956 billion RMB in 2024.

MiniMax also showed a trend of increasing revenue but expanding losses. In 2025, the company's revenue was $79.038 million, a year - on - year increase of 158.9%. Over 70% of its revenue came from overseas markets, and its globalization expansion was very successful. However, the annual loss was as high as $1.872 billion, a year - on - year increase of 302.3%.

It should be noted that the annual revenues of these two companies are less than one billion RMB, but their market values have soared to hundreds of billions of Hong Kong dollars. Against this backdrop, their price - to - sales ratios have been inflated, deviating from the industry's normal level.

The above - mentioned investor told Tech Planet, "These companies are priced by the capital market as if they were similar to mobile Internet infrastructure, but the revenue quality of most companies has not yet fully proven that they deserve such an infrastructure - level valuation."

The leading players are all facing their own core challenges.

Zhipu, which focuses on the government - enterprise and developer infrastructure track, although it has solid orders, still cannot escape the industry pain points of "heavy delivery and long project cycles". Its large - scale profit - making pace is limited by the project implementation efficiency.

MiniMax, which focuses on multi - modal and global AI applications, has quickly gained popularity with products like Xingye, Hailuo AI, and Talkie. However, it still cannot avoid the core questions of C - end AI applications: Can users continue to pay? Can the retention rate be stable?

Kimi, which excels in long - text processing and super workflows, attracts high - frequency use from academic researchers and professionals. However, can this high - frequency use be effectively converted into actual revenue? If it only stays at the level of high user activity but low payment conversion rate, it will instead create pressure.

Jieyue Xingchen, which is targeting the terminal intelligence track, aims to implement large models in physical terminals such as mobile phones, cars, and smart hardware. The track has broad prospects, but the interests of multiple parties in the industry are intertwined. The industrial chain cooperation model and profit distribution pattern have not yet been finalized, and large - scale implementation is still uncertain.

Facing such challenges, on the one hand, there is the soaring market value in the hundreds of billions, and on the other hand, there are the continuously expanding operating losses. Does this contrast between valuation and revenue - profit fundamentals carry the risk of a bubble? Who will ultimately foot the bill? This has become a key question that the entire industry urgently needs to answer.

This article is from the WeChat official account "Tech Planet" (ID: tech618). Author: Ren Xueyun. Republished by 36Kr with permission.