Introduction

Similar to the constantly updated rankings of large models over the past four months, the funding amounts and valuations of AI startups are also breaking new records.

At the beginning of the year, Zhiyu and Minimax both went public in Hong Kong, with Zhiyu’s stock price soaring more than sixfold, surpassing Baidu’s market cap—despite Zhiyu’s revenue being only 0.56% of Baidu’s last year and still facing significant losses.

As the stock market rises, pre-IPO AI startups have once again become the focal point for investors.

On May 6, it was reported that Kimi (Dark Side of the Moon) is about to complete a new round of $2 billion financing, with a post-financing valuation exceeding $20 billion, quadrupling in just six months. According to reports, Kimi has raised over 37.6 billion RMB, making it the AI startup with the highest cumulative financing.

DeepSeek, which has just begun external financing, has seen its valuation double in just a few weeks. Initially rumored to be valued at only $10 billion, it surged to $20 billion after Alibaba and Tencent entered the fray. Recent reports suggest that the National Integrated Circuit Industry Investment Fund is negotiating to invest in DeepSeek, with the latest valuation potentially reaching $45 billion to $50 billion.

This year, we may witness an unprecedented capital frenzy in artificial intelligence. Overseas companies like OpenAI, Anthropic, and Elon Musk’s SpaceX, which merged with xAI, are all set to go public this year, collectively valued at over $3 trillion. Following the listings of Zhiyu and Minimax, there have been repeated rumors that Kimi and Jietiao Xingchen are preparing for IPOs in Hong Kong.

However, this is far from the peak of the large model boom. Given the increasing computational expenses and Token consumption in the market, this funding battle may just be beginning.

Kimi’s Funding Success

Kimi was once the most prominent AI startup in China, attracting significant investment from major players like Meituan, Alibaba, and Tencent. It was also known for its aggressive spending, at one point exceeding ByteDance and Tencent in monthly expenditure.

However, after the launch of DeepSeek-R1 in early 2025, Kimi fell into a period of silence and uncertainty. Its flagship model, Kimi K1.5, struggled to make an impact in the shadow of DeepSeek, and its previous reliance on aggressive spending strategies faced criticism, leading to a pessimistic market outlook.

The past year has been a period of internal adjustment for Kimi. They paused their aggressive consumer-facing spending strategy and refocused on algorithms and model development, emphasizing open-source community engagement and model programming capabilities. Six months later, Kimi released and open-sourced its flagship model Kimi K2, significantly enhancing its code and agent capabilities, regaining market recognition. The K2.5 and K2.6 models launched this year have continued to drive revenue growth amidst the lobster craze.

As model capabilities improved, the capital market also renewed its investment. After completing over $300 million in Series B funding in August 2024, Kimi had not updated its funding status for 15 months. However, on the last day of 2025, Kimi announced it had completed $500 million in Series C funding, led by IDG, with Alibaba, Tencent, and Wang Huiwen as oversubscribed existing investors.

In a way, the emergence of DeepSeek has helped Chinese AI startups find differentiated competitive paths. Companies like Zhiyu, Minimax, and Kimi have avoided retracing the old path of competing as general AI assistants, instead focusing on model code and agent capabilities, leveraging the open-source community and overseas markets for renewed growth.

According to reports, following the K2.5 model update, Kimi’s ARR surpassed $100 million in early March and continued to grow to over $200 million in April. In comparison, as of March this year, Zhiyu’s MaaS platform ARR reached 1.7 billion RMB (approximately $250 million), a 60-fold increase year-on-year.

Although Kimi is slightly behind Zhiyu and Minimax in the IPO process, the soaring stock prices of the latter two may allow Kimi to raise more funds in the primary market. Amidst FOMO sentiment, the yet-to-be-listed Kimi has become the last lifeboat for investors. After all, one only learns to cherish what they missed.

This year, Kimi’s funding journey has accelerated significantly. Reports indicate that in January and February, Kimi completed three rounds of funding (including the Series C funding announced at the end of December), and with the latest $2 billion round, its total funding in less than six months has exceeded $3.9 billion. Based on cumulative funding amounts, Kimi’s 37.6 billion RMB has surpassed that of publicly listed Zhiyu and Minimax.

Even non-top-tier AI startups have benefited from this funding wave. At the end of February, reports indicated that Jietiao Xingchen was undergoing a new round of Pre-IPO financing, which would be delivered in two tranches, with a total financing amount of up to 3.2 billion RMB. Even so, related amounts are hard to secure, and many investors are already queuing up.

Ongoing Funding Battle in the AI Sector

Even though Kimi’s funding amounts and valuations are reaching historical highs, these funds may still be insufficient to support the future needs of large model competition.

As model capabilities continue to improve, Token usage is also skyrocketing. According to data from the National Bureau of Statistics, in March this year, China’s daily average Token usage exceeded 140 trillion, growing more than a thousandfold in just two years.

The explosive growth in Token usage has also significantly increased computational cost pressures. In March, major cloud service providers in China, including Tencent Cloud, Alibaba Cloud, and Baidu Smart Cloud, raised prices within a week, with some models in Tencent Cloud’s mixed Yuan series increasing by over 460%.

Since the beginning of the year, several large model companies have raised their API call prices. Kimi’s latest K2.6 model has an input price of $0.95 per million Tokens (cache miss), up approximately 58% from K2.5’s $0.60; the output price is $4, up about 33% from K2.5’s $3.

After the release of GLM-5.1 in April, Zhiyu’s input price per million Tokens (cache miss) increased again by 10%. Zhiyu’s CEO Zhang Peng stated in the first quarter report that API call pricing increased by 83% in the first quarter, yet demand continues to outstrip supply.

The capital expenditures of major internet giants have also significantly increased. Earlier reports indicated that ByteDance expects its capital expenditure to reach approximately 160 billion RMB this year, with about half allocated for AI chip procurement. Last year, Alibaba announced plans to invest over 380 billion RMB in cloud and AI hardware infrastructure over the next three years, exceeding the total investment of the past decade. Just months later, Alibaba announced it would further increase its investments.

Not to mention the larger-scale computational infrastructure plans of overseas companies like OpenAI, Meta, and Google. According to recent testimony from OpenAI co-founder and president Greg Brockman in court, OpenAI’s computational spending is expected to reach $50 billion this year, with plans to accumulate $600 billion in computational investments by 2030. Meta has also significantly raised its capital expenditure expectations for this year, with total expenditures expected to reach a historic high of $145 billion.

Startups are feeling even more pressure regarding computational costs. According to previously disclosed financial data from Zhiyu, over 70% of the company’s R&D expenditures are allocated to computational services. While revenues for large model companies continue to grow, their computational costs are also skyrocketing.

The talent war in the large model market is becoming increasingly intense. Over the past year, researchers from DeepSeek have become targets for major internet companies, resulting in team members frequently leaving. Reports indicate that ByteDance recruited former DeepSeek researcher Guo Dayan for nearly 10 million RMB in annual salary to lead the Agent team at ByteDance Seed.

DeepSeek, which has historically refused external financing, is now also attempting to raise more funds. Following the news of its financing, the company’s valuation has been continuously rising, with the latest valuation possibly reaching $45 billion to $50 billion.

Even though Kimi’s founder Yang Zhilin has consistently stated that the company is not in a hurry to go public, rumors about Kimi’s IPO have never ceased. In March, Bloomberg reported that Kimi is considering an IPO in Hong Kong and has had preliminary discussions about listing plans with CICC and Goldman Sachs. In early February, Jietiao Xingchen, which was reported to have initiated Pre-IPO financing, also hinted at plans to go public in Hong Kong within the year.

Zhiyu, which has successfully gone public in Hong Kong, is also advancing its A-share listing plans. According to previous announcements from the China Securities Regulatory Commission, Zhiyu intends to pursue an IPO on the Sci-Tech Innovation Board, with Guotai Junan Securities and CICC as its IPO advisory institutions.