Z.AI Co., Ltd. (02513) Report Interpretation

Overview

This earnings review assesses Z.ai’s H1 2026 results and concludes that the apparent revenue shortfall was driven by a strategic reduction in on-premises business rather than weakness in the more important API business. Bernstein sees accelerating API ARR, improving inference economics and continuing model iteration as the central drivers of its Outperform view.

Core views

Z.ai reported H1 2026 revenue of RMB954 million, up 399.7% year on year but 17.2% below Bernstein’s RMB1.152 billion forecast and 29.6% below RMB1.355 billion consensus. Bernstein attributes the miss primarily to on-premises deployment revenue, which fell 20.4% year on year to RMB129 million and was 62.1% below its estimate. The report regards this shortfall as a consequence of management scaling back a historically disliked business line, rather than as the key indicator of the company’s underlying growth. The more important cloud/API business performed better than expected. Cloud-based revenue was RMB825 million, up 2,735.7% year on year, 1.6% above Bernstein’s RMB812 million forecast and 18.3% above consensus of RMB697 million. Cloud gross profit was RMB203 million, 7.2% above Bernstein’s estimate, and cloud gross margin was 24.6%, compared with Bernstein’s 23.3% model. Total gross profit was RMB252 million, implying a 26.4% group margin. Bernstein argues that the solid API revenue and margin delivery demonstrate progress in the revenue mix despite the consolidated revenue miss. Costs and losses were also better than expected. H1 R&D expense of RMB2.1 billion rose 33.6% year on year and was below Bernstein’s RMB2.6 billion model. GAAP operating loss was RMB2.147 billion, below Bernstein’s RMB2.808 billion estimate and consensus of RMB2.926 billion, while adjusted net loss of RMB1.964 billion was below Bernstein’s RMB2.172 billion estimate and RMB2.750 billion consensus. The report therefore sees the results as satisfactory on the measures that matter for longer-term growth, although it notes that a negative market reaction to the headline revenue miss would not be surprising. Bernstein places substantial weight on the recent acceleration in ARR. Management’s guidance for approximately US$1.6 billion of monthly ARR in August implies roughly US$130 million, or RMB900 million, of API revenue in that month—more than total H1 API revenue. The report traces ARR growth from US$250 million in March to about US$500 million in June, then US$1 billion and US$1.6 billion in subsequent months, linking the trajectory to GLM-5.2, GLM-5.3 and GLM-5.3-flash launches. A weekly run-rate above US$2 billion would mean year-end 2026 ARR materially exceeds Bernstein’s prior expectations. Management’s discussion focused on expert-domain data, including cyber, finance and co-work, larger base-model scale, reinforcement-learning depth and serving infrastructure. Bernstein sees GLM-5.3-flash as a precursor to the next, much larger pre-training cycle; it considers the quality of that next model important to the subsequent growth path. The report also expects the company’s objective of raising the ratio of revenue to total compute spend to support some H2 inference-margin improvement, even with hardware-cost inflation. Its modelling assumes pricing for the larger pre-train below Kimi K3 levels because of management’s emphasis on token affordability. Reflecting the stronger API and ARR outlook, Bernstein raises revenue estimates to RMB7.905 billion for 2026E, RMB27.950 billion for 2027E and RMB52.921 billion for 2028E—40.4%, 49.8% and 19.1% above its prior forecasts, respectively. It also improves 2026E GAAP operating income from a RMB4.771 billion loss to a RMB3.181 billion loss, and forecasts positive GAAP operating income of RMB568 million in 2028E. The target price rises from HK$1,350 to HK$1,500, using unchanged valuation anchors: a blend of 25x 2030E P/E discounted back at 14% annually and a DCF valuation. Bernstein reiterates Outperform.

Analysis framework

Bernstein separates the H1 revenue miss by deployment segment, compares actual revenue, gross profit, margins, R&D and losses with its own estimates and consensus, then evaluates management’s ARR and model-development commentary for the forward growth path. It updates forecasts and values the company using a blend of discounted 2030E earnings and discounted cash flow.

Methodology notes

• Industry AnalysisVolume-price decomposition

  Segment-level comparison of cloud/API and on-premises revenue, gross profit and margins.

  The report separates the headline revenue miss into cloud/API and on-premises components to determine whether the shortfall reflects weaker core demand or a deliberate business-mix change.

• Valuation methodsP/E and PEG Valuation

  25x 2030E P/E discounted back at 14% annually.

  Bernstein applies a forward earnings multiple to its 2030 estimate and discounts the resulting value back to derive part of the price target.

• Valuation methodsDCF (Discounted Cash Flow)

  Discounted cash flow valuation.

  The DCF values projected free cash flow using the report’s long-term operating assumptions and contributes to the blended HK$1,500 target price.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Z.AI Co., Ltd. (02513.HK)

  Primary covered company; Bernstein rates the shares Outperform.

  Strengths

  API revenue exceeded Bernstein’s and consensus expectations, API gross margin was above Bernstein’s forecast, and ARR accelerated alongside recent model launches.

  Weaknesses

  Total H1 revenue missed estimates because on-premises deployment revenue fell sharply.

  Comparison

  Cloud/API revenue of RMB825 million exceeded Bernstein’s RMB812 million estimate and RMB697 million consensus, while total revenue was below both benchmarks.

  Risks

  Enterprise AI adoption in China may develop more slowly than expected; competition from rival AI model developers and changing sentiment toward Chinese AI labs may affect perceptions.

Key data

• H1 2026 total revenueRMB954 millionUp 399.7% year on year; 17.2% below Bernstein’s RMB1.152 billion estimate and 29.6% below RMB1.355 billion consensus.
• H1 2026 cloud/API revenueRMB825 millionUp 2,735.7% year on year; 1.6% above Bernstein’s estimate and 18.3% above consensus.
• H1 2026 cloud/API gross margin24.6%Above Bernstein’s 23.3% forecast.
• H1 2026 on-premises revenueRMB129 millionDown 20.4% year on year; 62.1% below Bernstein’s estimate.
• H1 2026 adjusted net lossRMB1.964 billionBelow Bernstein’s RMB2.172 billion estimate and RMB2.750 billion consensus.
• August monthly ARR guidanceUS$1.6 billionImplied approximately US$130 million, or RMB900 million, of API revenue in August.
• 2026E revenue forecastRMB7.905 billion40.4% above Bernstein’s prior estimate.

Impact & implications

Bernstein argues that the results strengthen the case for judging Z.ai primarily on API adoption, ARR expansion, model development and inference economics rather than on the shrinking on-premises business. The forecast increases and higher target price reflect its view that the rapid API run-rate is more consequential for the company’s future trajectory than the H1 headline revenue miss.

Risks

• Enterprise AI adoption in China could progress more slowly than expected.
• Competition from rival AI model developers could pressure Z.ai’s growth outlook.
• Investor perceptions of Chinese AI labs can shift materially with recent model releases, particularly during bearish AI sentiment.

What to watch

• The pace of API ARR growth following the reported US$1.6 billion monthly ARR guidance for August.
• The quality and market impact of Z.ai’s next, much larger pre-trained model.
• Whether a higher revenue-to-compute-spend ratio improves H2 inference margins despite hardware-cost inflation.
• The continued transition away from on-premises deployment toward cloud/API revenue.