WillItSurvive

Chinese foundation-model lab making GLM AI models, listed in Hong Kong since Jan 2026.

Losing its protection
Assessed 20 July 2026 · Evidence as of 20 July 2026
The structural protection is eroding.
On the recordSince 20 July 2026 · Checkable from Q4 2026 · Every call on the record Checkpoint: Lock-up expiry and dilution from the $4B secondary placement starts weighing on shares
The call

Zhipu has government ties and fresh IPO cash, but it burns seven times its revenue and rivals like Kimi K3 are already matching or beating it on the same benchmarks.

Zhipu AI builds large language models and sells access to them, mostly to Chinese government companies and state-owned businesses. It is not hard to copy or replace because Alibaba, ByteDance, Moonshot, and DeepSeek are already doing similar work in China with more money, and they are winning on speed and performance (Moonshot's Kimi K3 already beats Zhipu's GLM-5.2 at coding tasks). The one thing that could change this: if Zhipu's model stays top-tier against its Chinese rivals and holds onto government contracts while cutting costs, it survives; if competitors keep winning benchmarks or steal customers, the company burns through cash and needs constant government support to stay alive.

Why
01
AI pressure: Moonshot Kimi K3, DeepSeek, Alibaba Qwen. Kimi K3 already benchmarks ahead of GLM-5.2 on coding, and DeepSeek plus Qwen fight for the same government buyers with no lock-in stopping a switch. Could AI do the core job? Only partly.
02
Bigger players: ByteDance. ByteDance can bundle its own models across a huge distribution base and undercut Zhipu on price whenever it chooses to push harder into enterprise AI. Could a giant copy it? Probably.
03
What protects it. State backing, Hong Kong listing cash, and existing on-premise contracts inside government and SOE clients that are slow to rebid.
What would change this call
Zhipu's gross margin trend in upcoming quarterly filings after 2025's drop from 56% to 41%. Shows whether low-margin SOE deployments keep eroding profitability or stabilize.
Independent benchmark comparisons of GLM-5.2 versus Kimi K3 on coding tasks over the next two quarters. Indicates whether Zhipu is losing technical leadership to a domestic rival.
Any public disclosure of SOE or government contracts shifting from Zhipu to Alibaba, ByteDance, or DeepSeek. Would show direct customer defection to a same-market competitor.
Key numbers
  • 2025 revenue ~$105M, net loss ~$695M (2025-12-31)
  • Gross margin fell from 56.3% to 41% (2025-12-31)
  • Free float only ~4% of shares (2026-07-15)
Named risks
  • Free float near 4% amplifies stock volatility, single-day swings over 15% seen in July 2026. Severity: high
  • Gross margin fell from 56.3% to 41% due to low-margin on-prem SOE deployments (73.7% of 2025 revenue). Severity: high
  • Net loss of ~$695M against $105M 2025 revenue, roughly 7x burn multiple. Severity: high
  • Rival benchmark leapfrogging: Kimi K3 (July 2026) already ahead of GLM-5.2 on coding metrics. Severity: medium
  • US Entity List and export-control enforcement could tighten chip access despite Huawei Ascend/Cambricon co-design. Severity: medium
Rivals, as read in July 2026
RivalDirectionEvidence
Moonshot AI (Kimi K3)gaining onKimi K3, July 2026, already ahead of GLM-5.2 on some coding benchmarks
DeepSeekcontestingUndercutting Zhipu on price amid domestic model war, mid-2026
Alibaba (Qwen)contestingNo structural bar to same SOE/government deals as Zhipu, also a Zhipu investor
ByteDancecontestingNamed alongside Zhipu on Time's 10 Most Influential AI Companies of 2026
OpenAI and Googlegaining onReach Chinese-linked corporates via Singapore entities, July 2026, eroding sovereign moat
Questions people ask
Is Zhipu AI profitable?
No. Zhipu posted 2025 revenue of about RMB 724 million (~$105 million) but a net loss of roughly RMB 4.72 billion (~$695 million), a burn rate near 7 times revenue, as of December 31, 2025.
Who owns Zhipu AI, is it public now?
Zhipu AI, trading as Knowledge Atlas Technology, listed on the Hong Kong Stock Exchange under ticker 2513.HK on January 8, 2026, raising about HK$4.35 billion (~$558 million). Pre-IPO backers included Alibaba, Tencent, Ant Group and Xiaomi.
Who is Zhipu AI's biggest competitor?
Domestically, Moonshot AI's Kimi K3, released July 2026, was already benchmarking ahead of Zhipu's GLM-5.2 on some coding metrics. DeepSeek, Alibaba's Qwen and ByteDance also compete directly for the same government and enterprise customers.
Could AI replace Zhipu AI?
No, Zhipu AI is itself a frontier model lab. The real threat is peer commoditization from rivals like DeepSeek and Moonshot's Kimi K3 undercutting on price and benchmarks, not displacement by AI, as of July 2026.
What is Zhipu AI's revenue?
Zhipu reported 2025 revenue of about RMB 724 million (~$105 million), up 131.9% year on year, with its model-as-a-service business reaching an annualized run rate of RMB 1.7 billion (~$251 million) by March 2026.
How did Zhipu AI's stock perform after its IPO?
Zhipu's Hong Kong-listed shares surged roughly 1,500 to 2,000% between its January 8, 2026 IPO and mid-July 2026, briefly pushing its market capitalization above HK$1 trillion.
Timeline
  • 2026-07-15Stock up ~1,500-2,000% since IPO, market cap briefly tops HK$1 trillion
  • 2026-07-09Placed 19.78M H-shares at HK$1,588 to raise ~$4.0B secondary placement
  • 2026-06-16GLM-5.2 released with 1M-token context window
  • 2026-02-28Share price fell 23% amid compute shortages and user complaints
  • 2026-02-12GLM-5 released: 745B-param MoE, 77.8% SWE-Bench Verified
  • 2026-01-08IPO on HKEX as 2513.HK, raised ~HK$4.35B (~$558M)
  • 2025-07-01Rebranded internationally as Z.ai, GLM-4.5 launch, open-weight MIT license
  • 2019-01-01Founded as Tsinghua University research spinout
Revision history
20 July 2026
First assessed. Losing its protection. The text above is fixed as of this date and is not edited afterwards.
In the same band
Sources