Z.ai Revenue Soars 400% in H1 2026 on 27-Fold API Surge

Z.ai Co. Ltd. (2513.HK), the Hong Kong-listed large-model AI company also known as Zhipu, reported on August 31 that its first-half 2026 revenue surged 399.7% year-on-year to 954 million yuan ($142 million), surpassing its full-year 2025 total in just six months. The results mark the most dramatic revenue acceleration among China's publicly listed AI model companies.

API Business Becomes Dominant Revenue Engine

The growth was overwhelmingly driven by the company's open platform and API business, which generated 825 million yuan in revenue — a 27.4-fold increase from the same period last year. This segment now accounts for 86.5% of total revenue, up from just 15.2% a year earlier, completing a fundamental business model shift from project-based on-premise deployments to usage-based cloud API services.

Revenue from enterprise-level agents jumped 304.4% to 55.6 million yuan, while the legacy enterprise general-purpose large models segment contracted 54.6% to 67 million yuan. Localized deployment revenue fell 20.5% to 130 million yuan, dropping from 84.8% of the revenue mix to just 13.5%.

The company's MaaS (Model-as-a-Service) platform surpassed 7.4 million enterprise and developer users as of the announcement date, up 144% from the beginning of the year. Token call volume grew more than 40-fold over the same period, paid daily active users increased 603%, and the top ten customers' average daily call volume surged 98-fold. The platform now covers 233 countries and regions.

Cost Efficiency Gains and Margin Improvement

A critical enabler of the revenue surge was a dramatic reduction in inference costs. Z.ai said unit token inference costs dropped 80% from the start of the year, achieved through model architecture optimization and large-scale inference across a cluster of more than 100,000 domestic chips. Revenue per unit of compute investment improved 14-fold.

The open platform and API business gross margin improved from -0.4% in the same period last year to 24.6%, a swing of roughly 25 percentage points. API average pricing rose approximately 101%, reflecting both higher-value model tiers and strong demand.

Management disclosed at the earnings call that the MaaS platform's annual recurring revenue (ARR) reached $1.6 billion by the end of August, a 60% increase from $1 billion in early July. Back-calculating from this figure, Z.ai's single-month revenue in August was approximately 900 million yuan — nearly equal to the entire first-half total.

Losses Narrow but Profitability Remains Distant

Despite the revenue explosion, Z.ai is not yet profitable. Net loss for the period narrowed 12.1% to 2.07 billion yuan from 2.36 billion yuan a year earlier. However, adjusted net loss — excluding items such as equity incentives and book-value changes of financial instruments — widened 12.1% to 1.96 billion yuan from 1.75 billion yuan. Operating loss also expanded to 2.15 billion yuan.

R&D spending increased 33.6% to 2.13 billion yuan as the company continued investing heavily in its GLM family of large language models. The gap between API revenue and R&D expenditure remains substantial, indicating that the scale effect at the corporate level has not yet materialized. Sell-side analysts estimate the profitability turning point for Chinese large-model companies will arrive between 2028 and 2030.

Strategic Pivot to Coding and Agentic AI

Z.ai's trajectory represents a notable turnaround. In mid-2025, the company — then known primarily as Zhipu AI — was losing ground to DeepSeek and MiniMax, with an estimated 30% of customers shifting to DeepSeek at one point. The company responded by consolidating its model development around a single large-parameter model trained on a combination of reasoning, coding, and agentic data.

This pivot proved decisive. Z.ai released GLM-5 in February 2026, followed by GLM-5.1, GLM-5.2, and GLM-5.3 in August. GLM-5.2 achieved state-of-the-art status among open-source models on several benchmarks, ranking behind only proprietary models Claude Fable 5, Claude Opus 4.8, and GPT-5.5 on Artificial Analysis' overall leaderboard. GLM-5.3 tied Moonshot AI's Kimi K3 for first place among open-source models and ranked in the same performance tier as Claude Fable 5 and GPT-5.6 Sol.

Coding emerged as the key commercialization scenario. As foundation model companies worldwide moved into the coding market popularized by Anthropic, Z.ai was among the first Chinese companies to establish a meaningful position, giving it a first-mover advantage in converting model capability into billable API usage at scale.

Domestic Compute Chain and Market Position

Z.ai has built a complete chain of "self-developed models — domestic computing power — large-scale applications." GLM-5.2 achieved full compatibility with Huawei Ascend and Cambricon domestic compute platforms on its launch day, strengthening the domestic substitution narrative. U.S. export controls on Anthropic implemented in June 2026 further pushed some overseas developers toward Chinese models.

The stock, which went public in January 2026, rose 2.93% to HK$1,230 following the results announcement, trading roughly nine times above its IPO price. The shares had peaked at HK$2,980 on June 22 — a 24-fold gain from the issue price — before retreating approximately 65% as IPO lock-up periods expired and speculative premiums cooled. China Merchants Securities projects revenue of 35.4 billion yuan in 2026, 81.1 billion in 2027, and 150.2 billion in 2028.