August 31, 2026 (InvestinChina.asia) — Z.AI Co., Ltd. (formerly Beijing Zhihpu Huazhang Technology Co.), trading as Zhihpu (02513.HK), on Monday reported its first interim results since listing, with revenue soaring 399.7% year on year to 953.9 million yuan, even as the underlying business model underwent a dramatic restructuring that dragged overall gross margin lower and kept adjusted losses in the red.
The results are being read by the market as both a validation of China’s large-model-as-a-service (MaaS) commercialization thesis and a reminder that the path to profitability for frontier AI labs remains long. Revenue of 953.9 million yuan (approximately $142 million) already surpassed the company’s full-year 2025 top line. Yet adjusted net loss widened 12.1% to 1.96 billion yuan, underscoring the tension between explosive growth and persistent cash burn.
Revenue Structure: A Complete “Blood Change”
The defining story of the half is a wholesale shift away from one-off localized deployment contracts toward usage-based cloud consumption. According to the company’s filing with the Hong Kong Stock Exchange, revenue from the open platform and API services — billed per token consumed — exploded 2,735.7% year on year to 825.2 million yuan, accounting for 86.5% of total revenue, up from just 15.2% a year earlier and 26.3% at the end of 2025.
By contrast, revenue from enterprise-grade general-purpose large model deployments (localized, project-based) collapsed 54.6% to 67.0 million yuan, its share shrinking from 77.3% to 7.0%. Enterprise agent revenue grew 304.4% to 55.6 million yuan, while technical services and other income surged to 6.1 million yuan.
In its own words, the company described the transition as follows: as GLM’s coding capability crossed the threshold from simple code completion to autonomously delivering entire engineering projects (what it calls “Agentic Engineering”), “the model is no longer just a tool, but an entity capable of independently completing a task.” Usage-based billing thus became the natural — and more valuable — form of monetization.
| Revenue Segment | H1 2026 (mn yuan) | Share | H1 2025 Share | YoY |
|---|---|---|---|---|
| Open Platform & API | 825.2 | 86.5% | 15.2% | +2,735.7% |
| Enterprise Agents | 55.6 | 5.8% | 7.2% | +304.4% |
| Enterprise LLM (Local) | 67.0 | 7.0% | 77.3% | -54.6% |
| Technical Services & Other | 6.1 | 0.7% | 0.3% | +1,166.7% |
| Total | 953.9 | 100% | 100% | +399.7% |
Source: Z.AI 2026 Interim Results Announcement, Hong Kong Stock Exchange, August 31, 2026.
Profitability: Margin Compression Masks Unit Economics Improvement
Gross profit rose 163.7% to 251.6 million yuan, but overall gross margin halved from 50.0% to 26.4% — a direct consequence of the revenue mix shifting toward the lower-margin cloud segment, which was still in early-scale mode. However, the cloud business’s own gross margin improved from -0.4% to 24.6%, turning positive for the first time.
The company’s operating metrics tell the story behind the margin recovery:
- MaaS platform token consumption grew more than 40-fold from the beginning of the year;
- API average selling price rose approximately 101% year on year;
- Per-token inference cost fell 80% from the beginning of the year, driven by architecture optimization and domestic chip adoption;
- Revenue generated per unit of computing input (training + inference) improved 14-fold year on year.
Bottom Line: Headline Loss Narrows, But Adjusted Loss Widens
The apparent “loss reduction” requires careful reading. Period loss narrowed 12.1% to 2.07 billion yuan, but this was aided by certain non-recurring items. Stripping those out, adjusted net loss actually widened 12.1% to 1.96 billion yuan.
The divergence is explained by the absolute growth in R&D spending: research and development expenses rose 33.6% to 2.13 billion yuan — more than double the half-year revenue. Management attributed the increase to sustained investment in training the GLM series of models to push the boundaries of model intelligence. Still, the adjusted net loss ratio compressed dramatically from 917.9% to 205.9% of revenue, a 3.5-fold improvement in loss intensity — evidence that scale is rapidly diluting fixed cost burdens.
Running-Rate Signal: ARR Hits $1.6 Billion
Looking beyond the half-year figures, the company disclosed that as of August 31, 2026, its MaaS platform annualized revenue run-rate (ARR) reached $1.6 billion on a monthly-track basis and $2.0 billion on a weekly-track basis. The platform now serves over 7.4 million users across 233 countries and regions.
The ARR figure, while striking, represents an annualized extrapolation of current-period billing velocity rather than revenue already recognized. The gap between the $1.6 billion run-rate and the 953.9 million yuan (approximately $142 million) actually booked in H1 sets up the central question for the coming quarters: can Zhihpu convert this explosive usage momentum into sustained, recognized recurring revenue?
What the Market Is Watching
Zhihpu’s shares have experienced significant volatility in recent months. At its peak, the company’s market capitalization touched 1 trillion Hong Kong dollars, but the stock has since pulled back considerably. The company was formally included in the MSCI China Index on the same day its results were released.
The H1 report validates the core thesis that domestic large-model companies can achieve meaningful monetization through cloud APIs, with Zhihpu demonstrating the most radical pivot among China’s AI labs. Yet three issues remain unresolved:
- Profitability horizon: With R&D spending at 2.2x revenue, the timeline to operating breakeven remains undefined;
- Margin recovery: Whether the 26.4% blended gross margin can climb back toward 40%+ as cloud scale economics mature;
- Competitive moat: Whether GLM’s coding and agentic capabilities can stay ahead of fast-followers including DeepSeek, Alibaba’s Tongyi, and ByteDance’s Doubao, all of which are competing aggressively in the MaaS space.
The company indicated that the commercial acceleration driven by the GLM 5.2, 5.3, and 5.3-Flash releases since late June has not yet fully flowed into H1 recognized revenue, positioning the second half as the critical proving ground. Management is guiding toward a year-end ARR target of $2.0 billion.