China’s artificial intelligence large language model sector has witnessed another mega fundraising round. Just after Zhipu announced it had completed HK$31.4 billion (approximately $4.0 billion) in financing, another Hong Kong-listed AI model company, MiniMax (0100.HK), followed suit on July 10 with an announcement that it plans to raise approximately HK$16 billion (approximately $2.0 billion) through a share placement and convertible bond issuance. At the same time, MiniMax founder and CEO Yan Junjie issued an internal letter declaring that he will no longer receive any compensation from the company until artificial general intelligence (AGI) is achieved, and will allocate shares equivalent to 5% of total share capital from his personal holdings for employee incentives and to support the open-source community.
This marks MiniMax’s first large-scale refinancing since its Hong Kong Stock Exchange debut in January this year, just six months after its IPO. According to the announcement, the fundraising consists of two components: the company plans to place 35.6 million Class A shares at a price of HK$267.99 per share, expected to raise approximately HK$9.54 billion (approximately $1.2 billion); and simultaneously issue convertible bonds with a total principal amount of HK$6.5 billion (approximately $829.2 million). The combined net proceeds are expected to reach approximately HK$15.96 billion (approximately $2.0 billion).
In terms of capital allocation, MiniMax has clearly designated 80% of the net proceeds (approximately HK$12.77 billion) for strengthening AI infrastructure construction and large model R&D, 10% for the global commercialization of its AI-native product Harness, and the remaining 10% to supplement general corporate working capital. The company explained that as of June 30, 2026, approximately 77% of the funds raised from its earlier IPO designated for AI infrastructure and model R&D had already been used, with the pace of capital consumption exceeding initial post-listing expectations, primarily due to accelerated model iteration, growing customer scale, and rising user demand over the past six months.
Financial data shows MiniMax remains in a high-investment phase. In 2025, the company achieved operating revenue of $79.04 million, including $53.08 million from AI-native products and $25.96 million from its open platform and other AI enterprise services. Although gross margin improved to 25.4%, the company posted an adjusted net loss of approximately $250 million for the year due to continued investment in foundational model R&D and computing infrastructure.
Founder’s “Zero Pay” Pledge and Equity Incentives
On the same day as the fundraising announcement, Yan Junjie sent an internal letter to all employees, sending a clear signal of long-term commitment. He announced that, effective immediately until the day AGI is achieved, he will no longer receive any compensation from MiniMax. Additionally, he will allocate shares equivalent to 4% of the company’s total share capital from his personal holdings to incentivize team members who commit to long-term development with the company, and another 1% to establish a dedicated fund to continuously support the development of relevant open-source communities. Yan stated that he will devote all his time, energy, and resources to the long-term goal of AGI going forward.
Following this fundraising, MiniMax’s shareholding structure will change. After the share placement, total share capital will increase from 314 million shares to 349 million shares, with Yan Junjie’s stake decreasing from 25.22% to 22.65%, while newly introduced placees will hold approximately 10.19%. If the HK$6.5 billion convertible bonds are subsequently fully converted into Class A shares, Yan’s stake will further decline to 21.46%, with bondholders receiving approximately 5.26% of the shares.
Zhipu Simultaneously Ramps Up: “TouchHigh Plan” Targets AGI
Just one day before MiniMax’s announcement, the other AI model giant, Zhipu (02513.HK), also made major moves. According to a report by The Paper, Zhipu founder Tang Jie issued a company-wide internal letter on July 11 titled “The Giant Wave Has Arrived,” announcing the launch of a two-year strategic investment plan called “TouchHigh,” which abandons the pursuit of short-term application monetization to focus entirely on next-generation AGI.
In the letter, Tang reflected on Zhipu’s growth journey: from guarding an academic search system on a desktop computer in 2006, to betting on a hundred-billion-parameter model to create GLM-130B in 2021, to the January 2026 H-share listing day when he declared a “return to zero” to refocus on foundational model research. “Others ring the bell; we return to zero. This is not a posture, this is conviction — since the destination is AGI, short-term interests or industry tailwinds are merely scenery along the path to the endgame,” Tang wrote.
The “TouchHigh Plan” will focus on four major directions: tackling long-horizon task capabilities and developing next-generation memory architectures; building multi-agent collaboration systems to create “digital employee” clusters; implementing fully self-training systems and constructing high-quality synthetic data factories; and investing tens of billions of yuan in AI safety governance and advancing superintelligence alignment research. Tang particularly emphasized: “The more powerful the capability, the more robust the safety constraints must be. Superintelligence realization and alignment research must advance in parallel.”
Capital Market Performance Diverges Sharply
Notably, both Zhipu and MiniMax recently faced the expiration of their first batch of lock-up shares following their Hong Kong listings, but their capital market performances have diverged dramatically.
On July 8, the day Zhipu’s lock-up expired, its stock price surged 13.35%, with its market value briefly touching HK$906 billion (approximately $115.6 billion) the following day. In contrast, on July 9, the day MiniMax’s lock-up expired, its stock price plunged nearly 18%. As of the July 10 close, Zhipu traded at HK$1,640, with a total market value of HK$731.2 billion (approximately $93.3 billion); MiniMax traded at HK$268, with its total market value shrinking to just HK$84.2 billion (approximately $10.7 billion). Just three months earlier, MiniMax’s market value had reached as high as HK$410 billion (approximately $52.3 billion).
According to AI investor Guo Tao’s analysis for The Paper, the divergence stems from three factors: first, the difference in lock-up scale — MiniMax’s lock-up expiration covered 60% of shares, predominantly held by financial investors, creating significant selling pressure, while Zhipu’s was only 6%; second, shareholder attitudes — nearly 70% of Zhipu’s cornerstone investors committed to long-term holding, whereas despite some MiniMax shareholders pledging not to reduce holdings, the sheer volume of unlocked shares still triggered market concerns; third, market expectations — Zhipu received institutional favor for “converting model capability into commercial results,” while MiniMax was weighed down by a failed price increase and divided evaluations of its new M3 model.
MiniMax’s share price pressure is closely tied to the market performance of its next-generation large model, MiniMax M3. In early June, MiniMax released the M3 model and announced a permanent 50% API price cut a week later. According to The Paper, after the price cut news emerged, JPMorgan Chase downgraded MiniMax to “Neutral,” citing that the market did not recognize the product’s originally intended premium pricing, and the company’s failure to communicate with users in advance before adjusting prices also sparked user dissatisfaction on social media.
However, from a technical benchmark perspective, MiniMax M3 outperformed GPT-5.5 and Gemini 3.1 Pro on two authoritative benchmarks — SWE-Bench Pro, which measures professional software development work, and Terminal-Bench 2.1, which assesses command-line problem-solving — ranking among the top global models and first among Chinese open-source models.
The AI Model Race Enters a “High-Investment Cycle”
As Chinese AI model companies like Zhipu and MiniMax successively advance tens of billions in fundraising, industry competition is entering a new phase. In the past, AI model competition largely revolved around parameter scale and benchmark scores; now, as agents, multimodal capabilities, and enterprise applications gradually materialize, the computing infrastructure, capital reserves, and commercialization capabilities behind the models are becoming the new focal points of competition.
Looking at the strategic choices of the two companies, their paths have clearly diverged: Zhipu has explicitly stated it “does not pursue short-term monetization,” concentrating resources on foundational model research and safety governance; MiniMax, while continuing to ramp up R&D, is simultaneously advancing the global commercialization of its AI-native product Harness. Yet both founders have, without prior coordination, conveyed long-term commitment to the market through personal gestures — Yan Junjie announced zero pay until AGI, while Tang Jie wrote in his internal letter: “The giant wave has arrived; failing to reach the summit is failure.”
For AI model companies, how to sustain R&D investment, expand computing resources, and convert technical capabilities into commercial value will be key factors determining the next phase of the industry landscape. And as leading overseas players begin constructing computing clusters at the scale of millions or even two million chips, this race toward the “upper bound of intelligence” is clearly just beginning.