Zhipu eyes multi-billion-dollar share sale as Chinese AI firm rides 2,000% stock surge

Zhipu Share Sale Could Raise Billions After Stock Surge

Zhipu is considering a multi-billion-dollar Hong Kong stock sale as soon as next month, capitalizing on a remarkable 2,000% rally that has pushed the Chinese AI firm's market value above HK$1 trillion, or approximately $128 billion 1

1

2

2

. The company, which trades as Knowledge Atlas Technology, priced its shares at HK$116.2 during its $558 million Hong Kong listing in January. This week they touched HK$2,980 before closing at HK$2,174 1

1

. A six-month lock-up period from the IPO expires on July 8, opening the door for the placement 2

2

. The Beijing-based lab is also pursuing a Shanghai A-share listing as part of its broader fundraising efforts to support AI model development 2

2

.

Geopolitical Opening Created by Fable 5 Export Controls

Investor interest in Zhipu accelerated after an unexpected geopolitical shift handed the company a strategic advantage. In early June, Anthropic released Fable 5 as the public-facing version of its powerful Mythos model, which the company had described as advanced enough at writing code to pose a global cybersecurity risk. Three days later, US export controls barred all foreign nationals from access, including Anthropic's own employees 1

1

. Unable to guarantee compliance, Anthropic pulled both Fable 5 and Mythos 5 worldwide on June 12 1

1

. The sudden shutdown raised questions among developers, enterprises, and governments about depending on systems Washington could switch off. Lizzi Lee of the Asia Society Policy Institute noted the market "immediately understood the opening that Anthropic just created," while French politician Bruno Retailleau called the episode a "wake-up call" for Europe 1

1

.

Open-Source GLM-5.2 Model Climbs Global Rankings

Zhipu launched its answer the day after Anthropic's withdrawal. The company released GLM-5.2, a 744-billion-parameter open-source model that anyone can build on for free 1

1

. The open-source GLM-5.2 model climbed to second place on Arena's front-end coding leaderboard, behind only Fable 5. No Chinese AI model had reached the global top three before on the Artificial Analysis Intelligence Index, where only Anthropic and OpenAI rank higher 1

1

. Jefferies called it "a milestone for Chinese AI," while JPMorgan projected a revenue jump of more than 534% this year and a swing to profit by 2028 1

1

. The confidence surrounding these coding benchmarks prompted Zhipu's founder Tang Jie to engage in a public exchange with Elon Musk on X, with Tang asserting that matching Fable 5 "won't take that long" and that "focus is all we need, especially focusing on what intelligence truly is" 1

1

.

Profitability Concerns Shadow Market Enthusiasm

Despite the stock surge and investor enthusiasm, Bloomberg Intelligence analysts Robert Lea and Jasmine Lyu warned that comparisons with Anthropic "don't stand up to scrutiny." They noted that Zhipu "will likely remain deeply unprofitable," with rising agentic demand driving losses higher over the next 24 months 1

1

. The analysts added that the cash-hungry business will need continual fundraising for three years as agentic AI use widens its operating losses, making the multi-billion-dollar placement look less like a victory lap and more like a refuelling stop 1

1

. Jonathan Zhun Qiu, a partner at Meridian Capital Asia, rates GLM-5.2 as comparable to Anthropic's previous-generation Opus 4.7, and still behind Opus 4.8 on coding, though the gap continues to narrow 1

1

. Jefferies argues that a Fable 5 ban will not translate into meaningful revenue for Chinese labs because most enterprise AI still runs through direct interface access rather than self-hosted open models 1

1

. What has fundamentally changed, however, are the geopolitical implications. The Fable 5 shutdown demonstrated that access to leading American models can hinge on US domestic decisions, a reality that is driving the rally and reshaping calculations about AI dependency worldwide 1

1

.