Zhipu 1H26 Results: Narrower Losses, But Have the Storm Clouds Really Cleared?

Figure 1: Zhipu 2022 - 1H26 Revenue

As we noted in previous article, cloud deployment was set to be the next step in accelerating Zhipu's commercialisation — and this set of interim results shows the revenue mix shifting even faster than expected.

Zhipu's revenue for 1H26 reached CNY 950 million, up 399.7% y-o-y. Within this, cloud deployment revenue reached CNY 830 million, not only posting a 2,735.7% y-o-y increase but also overtaking on-premise deployment — previously regarded as Zhipu's core business — in terms of revenue share, reflecting a clear shift in the company's business model within a short period.

Driven by continuous model iteration, Zhipu drew broad market attention ahead of the official launch of GLM-5.3 by releasing it anonymously as the high-performance model “Ox Alpha”. This not only built anticipation for the new model ahead of launch, but once again demonstrated the strength of Zhipu's models. The model currently scores 45 on Artificial Analysis, ranking seventh globally among all models, and leading Moonshot AI's Kimi 3 by 3 scores. In fact, since GLM-5.2, Zhipu's models have progressively earned validation from developers and professional coding users; strong model performance combined with notably lower API costs has helped Zhipu rapidly build its user base and strengthen user stickiness. According to Zhipu's results briefing, registered users have now surpassed 7.4 million, up 144% since the start of the year. This reflects that, driven by improving model capability, Zhipu is increasingly capturing high-frequency subscription use cases, unlocking the associated commercial value and driving scaled growth in the cloud deployment business.

Figure 2: Zhipu 2022 - 1H26 Gross Margin

As part of the compute cost can be passed on to customers, on-premise deployment has carried a higher gross margin than the cloud business since 2023. In particular, under project-based billing, large project revenues once pushed Zhipu's overall gross margin above 60%. However, as we have previously noted, because Zhipu does not operate its own cloud infrastructure, the company must rely on third-party cloud service providers to deliver cloud deployment — and this compute cost scales directly with API call volume, diluting the benefit of revenue growth. Although Zhipu recently completed construction of a 1GW-scale data centre, the benefits of internalising compute costs have yet to show through, and with model call demand still rising rapidly, this pain point remains difficult to ignore.

Furthermore, as more domestic AI model companies roll out subscription products to capture enterprise customers, the appeal of on-premise deployment — with its higher upfront investment cost — may decline, pushing enterprise customers toward private cloud subscriptions or lower-cost AI services instead. As a result, even though Zhipu's revenue has risen quickly on the back of the cloud business, and average API pricing has risen by 101% in tandem, the y-o-y increase in cost of revenue (636.2%) — far outpacing revenue growth — has dragged overall gross margin down to 26.4%.

China AI Model Competition: GLM 5.3 Was Challenged by Kimi, Qwen and DeepSeek

Before the revenue mix fully shifted, on-premise deployment had helped Zhipu build an edge with enterprise clients. Zhipu's move into the cloud deployment arena may mean the company now has to contend with an even fiercer battle among domestic models.

Figure 3: Market Share by AI Company on OpenRouter

According to OpenRouter, Zhipu's share of developer API call volume stands at 9%, trailing DeepSeek, Google, and OpenAI — indicating that even as Zhipu's models begin to gain traction and recognition within the developer community, the company has yet to demonstrate a clear “leadership position” in terms of market share.

Figure 4: Comparison of China's Leading Flagship AI Models

Looking purely at the domestic market, competition among Moonshot AI, Alibaba, DeepSeek, and MiniMax has never let up. Comparing Zhipu's latest flagship model, GLM-5.3, against peers, Zhipu overlaps considerably with several mainstream models in terms of modality and product positioning. When different providers' functional positioning converges, customers find it easier to switch between platforms — indicating very low switching costs — which makes it harder for Zhipu to rapidly capture and defend market share. Pricing tells a similar story: Zhipu's API pricing is lower than that of Kimi and Qwen, and with the compute cost pain point still unresolved, even though GLM-series model optimisation has already brought down unit inference costs, this pricing strategy remains insufficient to support a recovery in profitability.

In addition, Zhipu has designated the Enterprise-level Agent as one of its core future product lines — an agent built on the GLM model to help enterprises automate defined business processes, with particularly strong performance in software development and cybersecurity. However, this product may be better suited to enterprises with a strong technical background and substantial R&D needs; for enterprises with more fixed workflows or lower R&D requirements, AI tools that integrate with their existing office ecosystem may better match operational needs.

One of the main competitors here is Alibaba's Qwen model. Because Alibaba's DingTalk platform is already a leading intelligent office and enterprise management platform domestically, Qwen can integrate more easily into DingTalk customers' existing workflows, lowering enterprises' training costs. A complete product ecosystem not only gives Qwen more diverse testing scenarios but also lowers subsequent customer-acquisition costs and speeds up model penetration. Precisely because of this, even where Zhipu's model performance has gained a degree of recognition, model performance alone may not be sufficient to guarantee Zhipu's competitiveness at the level of product adoption and go-to-market execution.

R&D Investment Is Running Hot

Figure 5: Zhipu R&D Expenses as a Percentage of Revenue

As of 1H26, thanks to Zhipu's rapid revenue growth, R&D expense as a percentage of revenue fell to 223%, while the net loss margin improved from 648.6% a year earlier to 217.1%. According to Zhipu, the company has set “Fully Self-Training” as its goal — using models to drive the self-optimisation of models. Should this technology vision be successfully realised, it could help lower training costs. Separately, Zhipu completed the acquisition of a 60% equity stake in XCore Sigma (Zhongke Jiahe) during 1H26; by leveraging XCore Sigma's technology in heterogeneous compute, the deal may help Zhipu make greater use of domestic chips, thereby lowering inference costs.

That said, Zhipu's technology roadmap will continue to push up R&D expense in the near-to-medium term, and it will take considerably longer to validate the technological outcomes and assess their impact on financial performance. In addition, investment activity such as equity acquisitions and data centre construction will place further pressure on Zhipu's cash flow in the near term. We believe that, at this stage, Zhipu's financial flexibility remains constrained, and therefore cannot rule out the possibility of further external financing going forward, nor overlook the potential dilution risk this poses to existing shareholders.

Storm Clouds Have Yet to Clear — Target Price Maintained

Figure 6: 12-Month Forward P/S of Zhipu

In summary, revenue growth of close to 400% and an annualised recurring revenue (ARR) of USD 1.6 billion are undoubtedly bright spots in this set of interim results. However, this rapid revenue expansion cannot fully mask the operational and financial concerns Zhipu continues to face. As Zhipu shifts its operational focus toward the more fiercely contested cloud and API battleground, the company may need to keep launching models that meaningfully outperform peers in order to defend its market position and technological lead. Yet sustaining that lead means R&D and capital expenditure are unlikely to scale back in the near term, while the process of achieving greater self-sufficiency in compute is still ongoing — keeping cash flow under sustained pressure. This leaves Zhipu caught in a cycle: it must keep proving its capability, which requires ever-greater R&D investment, which in turn deepens financial strain.

Given this, we maintain our view on Zhipu's fair price-to-sales ratio and profit growth forecast. We expect that, amid this year's high base effect, intensifying competition among domestic large language models, and Zhipu's API pricing already sitting below peers, revenue growth is likely to begin slowing from 2027 onward. Combined with the ongoing risk that liquidity and the scarcity premium could fade, we believe the valuation of Zhipu is still high, which will continue to limit its upward momentum.

Taking all of this together, based on our existing forecast framework, our end-2028 target price stands at HKD 968, implying a potential downside of 4.9% versus the closing price on 10 September 2026. Although Zhipu's share price has recently fallen sharply over several consecutive sessions and is now trading near HKD 800, as we have previously noted, Zhipu's share price is highly volatile. Therefore, even though the stock has pulled back significantly from both its peak and our target price, investors should still exercise caution in assessing the current situation and the company's future development.

Table 2: Valuation and revenue-per-share forecast of Zhipu

Declaration:
For specific disclosure, at the time of publication of this report, IFPL (via its connected and associated entities) and the analyst who produced this report hold a NIL position in the abovementioned securities.

This research report was prepared with the assistance of artificial intelligence (AI) tools. iFAST Financial Pte Ltd does not rely exclusively on AI for content generation; the content of this report – including all investment theses, ratings, price targets and conclusions – has been independently reviewed and verified by the research analyst(s) to ensure accuracy and professional integrity.