Chinese AI firm Z.AI (stock code: 02513.HK) announced on September 13 that it has secured $5 billion in financing, comprising approximately $2 billion in share placement and around $3 billion in convertible bond issuance. The final allocation structure reveals that overseas institutions formed the dominant force in this offering, with investors spanning major capital markets across Asia, Europe, and the Americas.

Global top-tier long-term funds and multi-strategy funds participated enthusiastically, including several of the world's Top 10 investment institutions managing assets exceeding $100 billion each. Multiple institutions submitted individual demand surpassing $500 million. Several existing long-term shareholders of Z.AI continued to participate and increase their positions, with some core institutions allocating to both equities and convertible bonds simultaneously. Notably, certain traditional long-term value investors who had adopted a cautious stance during the IPO phase have now shifted to actively building positions this time around.

During the placement and bond issuance process, the top 20 investors collectively received over 85% of the share placement allocation and more than 88% of the convertible bond allocation, with the final distribution concentrated among core institutions. This marks yet another fundraising round for Z.AI in 2026, following its Hong Kong Stock Exchange listing on January 8, 2026, when it issued 37.42 million shares at HK$116.20 per share, raising net proceeds of HK$4.173 billion.

Z.AI thus became the world's first listed large language model company. In July 2026, the company conducted its first placement at an issue price of HK$1,588 per share, issuing 19.78 million shares to raise total proceeds of HK$31.4 billion. After deducting listing-related expenses, net proceeds amounted to HK$31.375 billion. With the latest capital injection, Z.AI's total 2026 fundraising has exceeded HK$70 billion.

Where the new capital is heading

Z.AI stated that the proceeds will be directed toward the next-generation GLM, a fully self-training system, and related computing infrastructure, covering training, production inference, chip adaptation, and R&D talent development. Between February and August 2026, the company successively launched GLM-5, GLM-5.1, GLM-5.2, and GLM-5.3, maintaining a major upgrade cadence of approximately every two months. This round of financing will further support model training, production inference, and foundational engineering, allocating resources for subsequent technological iterations.

Sustained release of frontier models requires mutual reinforcement between R&D capability and long-term investment. Z.AI has demonstrated its model iteration capability through consecutive upgrades, and the addition of approximately $5 billion expands its capacity to arrange computing power, experiments, and talent investment across multiple R&D cycles. The defining feature of the "second half of AI" now extends to whether companies can consistently translate research breakthroughs into model outcomes. With technology accumulation paired with capital support, Z.AI has gained greater R&D autonomy while building reserves for scaling training experiments, advancing fully self-training exploration, and maintaining a frontier innovation pace.

Structure of the deal: $2 billion placement, $3 billion convertible bonds

Z.AI's placement and bond issuance adopted a "small equity, large debt" arrangement: equity represents only about 40% of the total, while approximately 60% comes from zero-coupon convertible bonds. The convertible bonds were issued at a price above face value, with the conversion price implying a double-digit premium over the market price, while carrying no cash coupon cost—locking in large-scale, low-cost, long-term capital while reducing immediate dilution.

For this placement, Z.AI set the placement price at HK$714 per share, with up to 21,965,000 new H-shares to be placed, representing approximately 4.72% of issued shares as of the announcement date and about 4.50% of issued shares after enlargement. The placement price represents a discount of approximately 9.96% to the closing price of HK$793.00 per H-share on September 11, 2026, and a discount of about 19.95% to the average closing price of approximately HK$891.90 over the five consecutive trading days ended September 11, 2026.

Assuming all placement shares are fully allotted, total proceeds from the placement are expected to reach approximately HK$15.683 billion (about $2 billion), with net proceeds of roughly HK$15.664 billion. Concurrently, Z.AI conducted its bond issuance, with gross proceeds expected around $3.016 billion (approximately HK$23.648 billion) and net proceeds of about $3.01 billion (approximately HK$23.6 billion). Based on the initial conversion price (subject to adjustment), the net issue price per conversion share is approximately HK$895.51.

The convertible bonds feature a zero-coupon structure issued at 100.5% of principal, with investors paying a premium for potential conversion value—reflecting positive expectations regarding the company's long-term equity value. According to informed sources, the investor lineup includes European long-term funds managing over $1 trillion, Southeast Asian long-term capital with nearly $100 billion in assets under management, Chinese growth-oriented long-term institutions, and multiple global top-tier private equity firms managing hundreds of billions of dollars each.

Nearly 30 long-term investors participated, with multiple existing long-term shareholders continuing to add positions. Meanwhile, a group of core institutions allocated to both placement shares and convertible bonds, expressing their medium-to-long-term value assessment of the company through a cross-asset approach. These investors typically focus on fundamental research and long-term holding rather than short-term trading, making their concentrated entry a powerful endorsement of value. Some traditional long-term value investors who remained cautious about the large model sector during the IPO phase have now transitioned from observation to active position building in this transaction.

Additionally, nearly 20 technology-focused investment institutions participated, including specialized funds highly focused on AI, semiconductors, and digital infrastructure, as well as leading investment platforms with deep roots in Asian technology assets. These represent the most professionally sophisticated capital with the deepest understanding of the sector and the highest research barriers—their allocation decisions cast a vote of confidence in the company's long-term scarcity within the large model industry chain.

First-half 2026 revenue reaches RMB 950 million; MaaS ARR hits $1.6 billion

Z.AI recently released its 2026 financial results, showing first-half 2026 revenue of RMB 954 million (approximately $142 million), representing a 399.7% surge from RMB 191 million in the prior-year period. Revenue from the open platform and API services reached RMB 825 million (approximately $123 million), soaring 2,735.7% from RMB 29.1 million a year earlier and becoming the group's primary revenue source. Revenue share from usage-based open platform and API services increased from 26.3% at the end of 2025 to 86.5%.

Revenue from enterprise-level general intelligent agents reached RMB 55.56 million in the first half of 2026, up 304.4% from RMB 13.74 million in the prior-year period. Revenue from enterprise-grade general large models was RMB 67.04 million, compared with RMB 148 million in the same period last year. Gross profit reached RMB 252 million, a 163.7% increase from RMB 95.42 million a year earlier, while adjusted net loss for the period stood at RMB 1.964 billion.

The most significant change in this earnings report is that Z.AI's sustained investment in model capability and inference efficiency has begun to concentrate into platform scale and revenue growth for MaaS: higher intelligence ceilings enable the models to tackle more complex, higher-value tasks, while lower inference costs allow these capabilities to be deployed at greater frequency and scale.

As of August 31, 2026, Z.AI's MaaS platform annualized recurring revenue (ARR) reached $1.6 billion on a monthly basis and $2 billion on a weekly basis. Enterprise and developer users surpassed 7.4 million, token call volume grew more than 40-fold from the start of the year, paying daily active users increased 603%, top-ten customer daily call volume grew 98%, and average API pricing rose approximately 101%. In the first half of 2026, Z.AI invested simultaneously in training and inference, incorporating domestic chips into its primary inference computing infrastructure. The company achieved cost-effective large-scale inference on 100,000-level domestic chips, reducing unit token inference costs by 80% from the beginning of the year.

R&D focus: next-generation GLM and fully self-training

Fully self-training represents a key strategic focus of Z.AI's current R&D investment. This round of financing will further support the development of the next-generation GLM and the fully self-training system, exploring a framework where next-generation models train within environments constructed by previous-generation models, progressively forming a recursive self-improvement loop. Related research covers training data generation and filtering, executable and verifiable task environments, and capabilities such as long-horizon reasoning and self-verification, enabling models to participate in more complete training and improvement processes.

Around this objective, Z.AI stated it will simultaneously strengthen training and production inference facilities, advance domestic chip adaptation, operator development, and computing resource scheduling to improve effective output of computational resources. The new capital will support larger-scale training experiments and successive rounds of technological iteration, consolidating capabilities in model R&D, post-training, and foundational engineering innovation.

During Z.AI's first-half 2026 earnings call, founder and chief scientist Professor Jie Tang stated that any single-point leadership is time-limited—being first on a coding benchmark today could be matched within weeks or months. Therefore, the understanding of SOTA should not be framed as "how long the lead lasts" but rather rewritten as "how strong the capability for sustained leadership is."

Tang said that sustaining leadership in coding or leading on every benchmark is not the most fundamental question. The truly important issue is whether, when coding extends downward, the company can naturally transfer these capabilities to the next layer—particularly the co-work level with more complex, higher-value tasks. Tang revealed the vision for the new model GLM 6.0: self-evolution. The model can judge when to stop and when to correct itself.

Tang also emphasized that Z.AI is pursuing cost reduction and efficiency gains, lowering inference costs with the goal that from day one of the next model's release, users can fully utilize it—not simply creating a large model that is impractical to deploy.

Continuing to reach higher: targeting the next summit of AGI

Several months ago, Tang published an internal letter stating that Z.AI has never been a company that chases trends. It grew out of a laboratory, carrying with it two decades of that laboratory's methodology—summarized in three words: essence, counter-intuition, and focus. Think deeply enough to dare to choose differently; choose differently enough to hold on persistently.

"Others ring the bell at IPO; we reset to zero. This is not posture—this is conviction. Since the endpoint is AGI, short-term profits or industry trends are merely scenery along the road to the final destination," Tang wrote. He described Z.AI's choices as consistently "counter-intuitive"—"while the industry accelerates commercial monetization, we have decided to break through upward."

Tang expressed belief in a simple truth: those who truly reach the summit will pave the mountain into a road. The wave has arrived; the trend is irreversible. Z.AI intends to be the one meeting the wave head-on, reaching upward. "Not reaching the summit means failure. This time, what Z.AI aims to touch is the height that belongs to all humanity."

Z.AI has named this strategy the "Touch High Plan." At this historic juncture where artificial intelligence transitions from perception and cognition toward full general intelligence, Z.AI states it will challenge the current physical and algorithmic limits of technology with a "reaching high" posture. Over the next two years, Z.AI plans strategic investment—not pursuing short-term application monetization, but aiming directly at the next high ground of AGI.