Zhipu AI — trading internationally as Z.ai, and legally renamed Knowledge Atlas Technology — is the Chinese frontier lab that turned itself into a stock market phenomenon. Founded in 2019 out of Tsinghua University’s Knowledge Engineering Group, it is the oldest of China’s “AI tiger” startups, and on 8 January 2026 it became the first foundation-model lab anywhere to go public, raising HK$4.3 billion (US$558 million) on the Hong Kong Stock Exchange (Bloomberg).[1][2] The retail tranche was oversubscribed 1,159 times (CGTN).[3]
What happened next was extraordinary. After the mid-June 2026 launch of GLM-5.2 — an open-weight model that beats OpenAI’s GPT-5.5 on SWE-bench Pro at roughly one-sixth the per-token cost (VentureBeat) — Zhipu’s market capitalisation touched HK$1 trillion (~US$128 billion) intraday on 22 June 2026, with the stock up more than 2,400% from its HK$116.20 IPO price at its peak (SCMP, Caixin).[4][5][6] Few listed AI pure-plays, Chinese or Western, had ever reached that scale.
The shares have since given back much of that, but Z.ai used the rally to raise about $9 billion more: a roughly $4 billion follow-on share sale in July and $5 billion from shares and convertible bonds in September.[7][8]
The catch: Zhipu remains deeply unprofitable, booking a RMB 4.72 billion loss on just RMB 724 million of revenue in 2025 (SCMP).[9] First-half 2026 revenue jumped to RMB 954 million, more than all of 2025, but the period loss was still RMB 2.07 billion (Reuters, August 2026).[10] It is the purest live experiment in what a frontier lab is actually worth — and the market is grading it in public, every quarter.
From Tsinghua lab to Z.ai: the oldest of China’s AI tigers
Zhipu was spun out of Tsinghua University in June 2019 by a team from the university’s Knowledge Engineering Group (KEG) including Tang Jie, Zhang Peng and Liu Debing, with Zhang as CEO — making it three years older than OpenAI’s ChatGPT moment and the longest-running of the Chinese frontier labs, ahead of Moonshot AI (2023) and DeepSeek (2023).[11][12][13] Its GLM architecture predates the current boom: the original General Language Model paper was presented at ACL in 2022, and the ChatGLM series made Zhipu the default open-source Chinese LLM before Qwen and DeepSeek overtook it in mindshare during 2024-25.[14]
The company raised roughly US$1.5 billion pre-IPO from a who’s-who of Chinese tech — Alibaba, Tencent, Ant Group, Meituan and Xiaomi — plus state-linked funds and Saudi Aramco’s Prosperity7 Ventures, which joined a US$400 million round in May 2024 at a ~US$3 billion valuation.[15][16][17] In July 2025 it rebranded internationally as Z.ai, a shorter badge for its overseas API, chatbot and coding products, while keeping the Zhipu name at home.[17]
The product line now spans the MIT-licensed GLM open-weight model family, the Z.ai assistant, the AutoGLM device-control agent, and the GLM Coding Plan — an agentic coding subscription that, at the time of the IPO, was priced from around 20 yuan (under US$3) a month in China, roughly one-seventh the cost of Anthropic’s equivalent (CGTN).[3] For where Zhipu sits in the broader ecosystem, see our China AI market overview.
The first frontier lab you can buy shares in
Zhipu listed on the Hong Kong Stock Exchange (ticker 2513) on 8 January 2026 at HK$116.20 per share, valuing it at roughly US$7.1 billion — modest next to OpenAI’s private valuation, and that gap is precisely what the market spent the next six months closing.[1][18] Shares rose 34% on the day GLM-5 launched in February, fell nearly 23% later that month on reports of compute shortages, jumped 11.5% on GLM-5.1’s open-sourcing in April, and surged as much as 42% intraday when GLM-5.2 landed in June (SCMP).[5] By 22 June 2026 the market cap had crossed HK$1 trillion — briefly making a seven-year-old, loss-making Beijing startup worth more than most of China’s listed internet incumbents.[6]
Public status cuts both ways. Zhipu’s first annual results as a listed company, for the year to December 2025, showed revenue of RMB 724.33 million (US$104.8 million), up 131.9% — but below the RMB 756 million Bloomberg consensus (SCMP, March 2026).[9] Few frontier labs have ever missed an earnings estimate in public, because few have earnings estimates at all. JPMorgan nonetheless raised its 2026-2030 forecasts after GLM-5.2, projecting revenue growth of 534% in 2026 — roughly RMB 4.6 billion — and first profitability in 2028 (SCMP, June 2026).[5] Hong Kong has become the default listing venue for this cohort: our guide to Hong Kong as Chinese AI’s listing capital tracks who is queued up behind Zhipu.
The first half of 2026 showed what the shift to the cloud looks like in numbers. Revenue rose about 400% to RMB 954 million, with API and open-platform sales up more than 27-fold to around RMB 825 million, 86.5% of the total, while on-premises deployment shrank.[10][19] The period loss narrowed to RMB 2.07 billion, while R&D spending rose to RMB 2.13 billion.[20][21] JPMorgan now expects about RMB 5 billion of revenue for 2026 and an adjusted profit by 2028.[10]
GLM-5.2: the cost-performance play
Zhipu’s 2026 model cadence has been relentless. GLM-5, released 11 February 2026, is a 744-billion-parameter mixture-of-experts model (40 billion active per token) under an MIT licence, reportedly trained entirely on Huawei Ascend chips with the MindSpore framework — zero Nvidia dependency — and scored 77.8% on SWE-bench Verified, ahead of GPT-5.2’s 75.4% (Hugging Face analysis, February 2026).[22] API pricing launched at US$1.00 per million input tokens and US$3.20 per million output tokens.[22]
GLM-5.2, released as a hosted model on 13 June 2026 and open-sourced on 16 June, pushed to 753 billion parameters and a 1-million-token context window.[4][6] It scored 62.1 on SWE-bench Pro against GPT-5.5’s 58.6, and 74.4% versus 72.6% on FrontierSWE (VentureBeat, June 2026) — while pricing at US$1.40/US$4.40 per million input/output tokens, roughly one-sixth of GPT-5.5’s per-token cost.[4] On Code Arena’s front-end leaderboard it ranks second globally, trailing only Anthropic’s frontier Claude models (Caixin).[6] Independent reviewers put it just behind Claude Opus 4.8 on FrontierSWE (74.4% versus 75.1%) and further back on longer-running tasks — behind the absolute frontier, but at a fraction of the price, which is exactly the wedge.[23] Our AI model cost tracker logs the per-token maths, and our US vs China capability tracker follows the benchmark gap.
The rumoured GLM-5.5 never arrived. Zhipu shipped GLM-5.3 instead in mid-August, an extended post-training pass on the same base model rather than a new pretraining run, followed by the multimodal GLM-5.3-Flash.[24][25] At its interim results briefing on 31 August, founder Tang Jie named GLM-6.0 as the next-generation base model, without giving a date.[26]
Who uses GLM?
Per its IPO prospectus figures, Zhipu counted around 12,000 enterprise clients, 150,000 paying developers across 184 countries, and 2.9 million users on its model-as-a-service platform as of mid-2025, including half of China’s ten largest internet companies (CGTN, January 2026).[3] Its heritage is enterprise and government on-premises deployment — though CEO Zhang Peng said after the 2025 results that clients are “gradually shifting” to cloud API consumption, a higher-margin model the listing story depends on (SCMP).[9]
The more geopolitically interesting customer base is abroad. OpenAI publicly warned in June 2025 that Zhipu had made “notable progress” selling sovereign-AI infrastructure — often bundled with Huawei hardware — to governments and state-owned enterprises in Malaysia, Singapore, the UAE, Saudi Arabia and Kenya.[27] At its IPO, Zhipu said it was leading an “International Alliance for Independent Large Model Co-construction” spanning ten ASEAN and ten Belt and Road countries (CGTN).[3] Overseas revenue was still only 11.6% of the total in H1 2025, almost all from Southeast Asia — small money, big strategic footprint.[3]
The Entity List problem
Zhipu was added to the US Commerce Department’s Entity List in January 2025, cutting off access to US-origin technology, including Nvidia’s export-grade chips (Bloomberg, January 2025).[28] Zhipu said it “strongly opposed” the designation and would be unaffected, since it did not rely on American large-model technology.
Nearly two years on, that claim looks less like bravado than roadmap: GLM-5 was reportedly trained wholly on Huawei Ascend silicon, and it was built to run inference on domestic chips from Moore Threads, Cambricon and Kunlunxin as well as Ascend.[22][29] But the constraint is real. Zhipu’s February 2026 share slump was triggered by reported compute shortages, and domestic chips still trail Nvidia on training efficiency — an overhang on every future model, GLM-6.0 included.[5] The Entity List also complicates Western enterprise adoption of Z.ai’s hosted API, even where the MIT-licensed open weights themselves face no restriction.
What could go wrong?
Four risks stand out. First, valuation: at a ~US$128 billion peak against RMB 724 million of 2025 revenue, Zhipu traded at over 1,000 times trailing sales; even JPMorgan’s bullish RMB 5 billion 2026 estimate leaves it priced for flawless execution, with profits not expected before 2028.[9][10] Second, competition: DeepSeek, Alibaba’s Qwen and Moonshot’s Kimi ship comparable open-weight models at similarly aggressive prices, and Zhipu’s own chairman Liu Debing warned at the IPO of a global price war spreading from China’s “hyper-competition” (CGTN).[3] Third, compute: Entity List restrictions make every training run hostage to domestic chip supply. Fourth, the listed-lab dynamic itself — quarterly disclosure of losses, R&D spend (RMB 3.18 billion in 2025) and misses gives rivals and regulators a level of visibility no private lab endures, and makes the share price a volatility amplifier for every model launch.[9] Zhipu is still the market’s chosen proxy for Chinese AI. That is a crown with a margin call attached.
Frequently asked questions
What is the difference between Zhipu AI and Z.ai?
They are the same company. Zhipu AI adopted Z.ai as its international brand in July 2025, and its listed legal name is Knowledge Atlas Technology JSC Ltd (HKEX: 2513).[30] Zhipu remains the name used in China; Z.ai fronts the overseas API, assistant and coding products.
How much money does Zhipu AI make?
Zhipu reported revenue of RMB 724.33 million (US$104.8 million) for 2025, up 131.9% year on year, against a RMB 4.72 billion loss (SCMP, March 2026).[9] First-half 2026 revenue was RMB 954 million, up about 400%, with a RMB 2.07 billion loss (Reuters, August 2026), and JPMorgan expects about RMB 5 billion for the full year and an adjusted profit by 2028.[10]
Is GLM-5.2 really better than GPT-5.5?
On agentic coding benchmarks, yes: GLM-5.2 scored 62.1 on SWE-bench Pro versus GPT-5.5’s 58.6, at about one-sixth the per-token price (VentureBeat, June 2026).[4] On broader capability it still trails Anthropic’s Claude Opus 4.8 on several key benchmarks, so “cheaper near-frontier” is the accurate framing.[23]
Can Western companies use GLM models despite the Entity List?
The Entity List restricts exports of US technology to Zhipu, not downloads of Zhipu’s MIT-licensed open weights, which remain freely available on Hugging Face. Using Z.ai’s hosted API is legally distinct again, but many Western enterprises avoid it for compliance and data-governance reasons.
What is Zhipu’s latest model?
GLM-5.3, released in mid-August 2026 in place of the rumoured GLM-5.5 and followed by the multimodal GLM-5.3-Flash.[24][25] Zhipu has named GLM-6.0 as its next base model but hasn’t given a date.[26]
Sources & Further Reading
- China’s OpenAI Rival Zhipu Debuts in Hong Kong After $558 Million IPO — Bloomberg, 2026-01-07
- Zhipu AI Becomes the World’s First Public Company Centered on AGI Foundation Models with Hong Kong Listing — TechNode, 2026-01-08
- 4 key takeaways: Zhipu becomes first Chinese AI firm to go public — CGTN, 2026-01-09
- Z.ai’s open-weights GLM-5.2 beats GPT-5.5 on multiple long-horizon coding benchmarks for 1/6th the cost — VentureBeat, 2026-06
- Zhipu AI market cap tops HK$1 trillion as shares of GLM-5.2 developer soar — South China Morning Post, 2026-06
- New Model Propels Zhipu AI’s Market Value to Record HK$1 Trillion — Caixin Global, 2026-06-23
- China’s Zhipu AI launches US$4 billion Hong Kong shares offering, term sheet shows — The Standard (Reuters), 2026-07
- China’s Z.AI raises US$5 billion from new share, convertible bond sales, filing shows — The Standard (Reuters), 2026-09-13
- Zhipu AI revenue jumps 132% in first post-IPO report, missing estimates — South China Morning Post, 2026-03
- China’s Zhipu AI revenue quintuples in first half, loss narrows — The Standard (Reuters), 2026-08-31
- In Profile: This Trio Is Charting the Course for China’s Leading AI Developer — Caixin Global, 2026-07-02
- Moonshot AI — Wikipedia
- DeepSeek — Wikipedia
- GLM: General Language Model Pretraining with Autoregressive Blank Infilling — ACL Anthology, 2022
- Alibaba, Tencent and other major Chinese backers invest US$342 million in start-up Zhipu AI — South China Morning Post, 2023-10
- Saudi Fund Joins $400 Million Financing for China AI Firm Zhipu — Bloomberg Law, 2024-05-31
- China’s Z.ai Scoffs at US Sanctions, Launches Supercharged AI With $1.5B War Chest — Decrypt, 2025-07-28
- China’s Zhipu AI Jumps in Hong Kong Debut — Caixin Global, 2026-01-08
- Zhipu AI’s H1 Revenue Jumps Nearly 400% as API Business Takes 86.5% of Sales — EqualOcean, 2026-09-02
- Zhipu Revenue Rises Nearly 400% as Cloud Business Gains Ground — Caixin Global, 2026-09-01
- Zhipu lost 2 billion yuan in the first half of the year. Can selling Tokens on Tmall solve its profitability dilemma? — 36Kr, 2026-09-03
- GLM-5: China’s First Public AI Company Ships a Frontier Model — Hugging Face, 2026-02
- GLM-5.2 Just Beat GPT-5.5 at a Sixth of the Cost — Labellerr, 2026-06-22
- Z.ai launches GLM-5.3 with claimed 50% gain on coding benchmark — TechNode, 2026-08-17
- Zhipu launches GLM-5.3-Flash, its first natively multimodal model built for Chinese chips — Crypto Briefing, 2026-08-26
- Zhipu responds to criticism of ‘post-training only’: next-generation foundational model GLM-6.0 targets ‘self-evolution’ — AICoin, 2026-09-01
- OpenAI says Zhipu AI accelerates China’s global efforts to challenge US tech dominance — South China Morning Post, 2025-06
- US Blacklists Tencent-Backed AI Startup as Tech Curbs Deepen — Bloomberg, 2025-01-16
- Chinese AI startup Zhipu releases new flagship model GLM-5 — Reuters (via AOL), 2026-02-11
- China ‘AI tiger’ Zhipu debuts in Hong Kong IPO — CNBC, 2026-01-08