Z.ai's Revenue Quintupled to $142 Million in H1 — and Its Losses Are Finally Shrinking

Z.ai, the Beijing-based company behind the GLM family of large language models and China’s first publicly listed pure-play foundation-model developer, has reported first-half 2026 revenue of 953.89 million yuan — roughly US$142 million — up nearly fivefold (about 400%) from the same period a year earlier. It is the clearest financial evidence yet that China’s aggressive open-source AI strategy is converting into real commercial traction, and it arrives at a moment when the global industry is watching to see whether anyone besides Nvidia and the hyperscalers can actually make money from foundation models.

The headline growth number, however, hides a more complicated picture. The results slightly missed analyst expectations, the company remains deeply unprofitable, and its stock — up more than 800% since its January Hong Kong listing — is priced for perfection. The H1 report is thus best read as two stories at once: a genuine demand surge, and a stress test of whether price-war economics can ever produce a durable AI business.

The numbers

According to the company’s interim report, revenue for the six months ended June 30, 2026 reached 953.89 million yuan (about US$142 million), versus roughly a quarter of that in H1 2025. Gross profit rose 163.7% to about 252 million yuan, implying a gross margin in the mid-20s percent range — thin by Western AI-lab standards but a meaningful improvement for a company that once gave away inference nearly free.

The growth was powered overwhelmingly by API sales. The Information reported that Z.ai’s API revenue surged in the first half as developers flocked to its GLM MaaS (model-as-a-service) platform, which now serves a rapidly expanding base of enterprise clients and independent developers. This tracks with the company’s earlier disclosures: by July 2026 Z.ai was reportedly operating at a US$1 billion annualized revenue run-rate, with GLM API prices having risen a cumulative 83% during the first quarter even as usage grew around 400% — a rare example of a Chinese AI vendor successfully raising prices amid a price war.

On the bottom line, the net loss attributable to shareholders narrowed 12.1% to 2.07 billion yuan (about US$308 million), from 2.36 billion yuan a year earlier. The adjusted net loss moved in the opposite direction, widening 12.1% to about 1.96 billion yuan once share-based compensation and other items are stripped out. In other words: the core business is scaling, but the company is still burning well over US$300 million a half-year once everything is counted.

Why the stock wobbled

Markets greeted the results with a shrug. Shares initially rose about 2.9% to HK$1,230 but the report drew headlines for missing consensus: Bloomberg-reported analyst estimates had penciled in higher sales, and Z.ai fell short by roughly 29% on some tallies. The miss was attributed directly to China’s AI price war, which has pushed average enterprise inference prices to 2026 lows — recent research cited average inference prices between roughly US$1.16 and lower bounds per million tokens depending on model class, with Chinese open-source models undercutting Western frontier APIs by 60% to 90% on comparable workloads.

The valuation math is where things get uncomfortable. Z.ai’s market value has surged more than 800% since its January listing, peaking around US$137 billion — enough to outstrip many long-established tech giants — before settling near US$71 billion at the time of these results. Against half-year revenue of about US$142 million, that is a price-to-sales ratio in the hundreds. Bulls argue the public float is small, the run-rate is compounding fast, and GLM’s developer mindshare ( cemented by the open-weight GLM-5.3 family and the viral Ox Alpha reveal) justifies a premium. Bears point out that even the company’s own full-year guidance — full-year sales expected to expand around 514% — implies annual revenue of only a few hundred million dollars against a US$71 billion market cap.

The strategic story behind the surge

Z.ai’s H1 report is the financial counterpart to a strategy shift the company has been executing since late 2025. After falling behind in traditional enterprise AI deals — where rivals like Alibaba’s Qwen and ByteDance’s Doubao bundled models into cloud contracts — Z.ai pivoted hard to what it does best: open weights plus cheap, fast APIs. The GLM-5.3 series, released with open weights under a bespoke license, became the default “good enough and 10x cheaper” choice for a huge cohort of Chinese and increasingly Western developers. Coding agents, in particular, became Z.ai’s wedge: the GLM Coding Plan undercut comparable plans from US vendors by an order of magnitude while posting competitive benchmarks.

That pivot shows up cleanly in the segment mix. S&P Global research had forecast that Z.ai’s cloud/MaaS business would overtake its on-premise deployment business in 2026, accounting for roughly 63% of total revenue for the year — and the H1 API surge suggests that crossover is happening on schedule. The company was ranked first among independent general large-model developers in China by 2024 revenue with a 6.6% market share, and the 2026 numbers indicate it is extending that lead among independents even as the overall market commoditizes.

What it means

Three takeaways worth watching for the rest of 2026:

1. Open-source is a distribution strategy, not a charity. Z.ai’s fivefold revenue growth is the strongest public-market validation yet of the “open weights, monetize the API” playbook. Developers who downloaded GLM weights for free converted into paying MaaS customers at scale — something Meta has struggled to demonstrate with Llama.

2. The price war’s survivors will be decided by cost structure, not model quality. With Chinese open-source models 60-90% cheaper than Western frontier APIs and inference prices at yearly lows, the question is no longer “which model is best” but “who can serve a capable model at the lowest cost per token and still cover the GPU bill.” Z.ai’s narrowing loss suggests it is closer to that answer than many peers, but a 2.07 billion yuan half-year loss is still a long way from breakeven.

3. Expectations are now the enemy. An 800%-in-eight-months stock run means every earnings report is a referendum. Z.ai beat on growth and narrowed losses, and still got “sales miss” headlines. For China’s first listed LLM pure-play, the H1 2026 report is genuinely good news wrapped in a valuation that leaves no room for anything less.

The next checkpoint is the full-year report. If the guided 514% annual growth lands and the loss keeps narrowing, Z.ai will have proven that the open-source AI business model can work as a public company. If the price war intensifies instead, the US$71 billion question gets a lot harder to answer.