Zhipu AI’s Hong Kong‑listed shares erupted this week after the company open‑sourced its next‑generation foundation model GLM‑5 and raised prices on an AI coding subscription, briefly touching HK$492 and finishing at HK$485 on February 13. The stock has more than doubled since the start of February and climbed roughly 138% in the past week, a surge driven by renewed investor appetite for assets tied to large language models and monetisation signals from product pricing.
GLM‑5 is central to the market excitement: independent rankings placed it fourth globally and first among open‑source models, a plaudit that amplifies Zhipu’s technical credibility outside of China’s closed domestic ecosystem. Open‑sourcing a high‑performing model is a strategic bid to capture developer mindshare, accelerate ecosystem adoption and differentiate Zhipu from both domestic rivals and western incumbents that are less permissive about code release.
At the same time Zhipu has taken the commercial step of hiking prices for its GLM Coding Plan, cancelling first‑time purchase discounts while keeping quarterly and annual subscription incentives and raising list prices by at least 30% for new buyers. The company framed the move as a response to surging demand and heavier usage, and said it is simultaneously investing in additional compute capacity and model optimisation to preserve service quality under high load.
The market’s enthusiasm contrasts sharply with Zhipu’s underlying profitability picture. Revenues have grown rapidly — a compound annual growth rate north of 130% from 2022–24, and RMB 191 million (≈$27m) in the first half of 2025 — but net losses are enormous: RMB 2.958 billion (≈$410m) in 2024 and a further RMB 2.358 billion (≈$328m) in H1 2025. R&D is the dominant cost centre, at RMB 2.195 billion (≈$305m) in 2024 and RMB 1.595 billion (≈$221m) in H1 2025, with cloud and compute services alone accounting for roughly RMB 1.553 billion (≈$216m) last year.
That cost profile highlights a structural challenge for model builders everywhere: scale brings both revenue opportunity and rapidly escalating compute bills. As parameter counts and training iterations balloon, the price and availability of GPU hours — and the ability to secure stable, cost‑efficient compute supply — become determinative for margins and operating cadence.
Zhipu is also advancing its China listing plans. The firm has refiled IPO counselling paperwork for a mainland Shanghai STAR Market debut, swapping to a joint sponsor arrangement with Guotai Haitong and CICC. Market analysts point to a persistent A‑share valuation premium over H‑shares and say a mainland listing could widen the investor base and improve access to domestic capital, potentially allowing Zhipu to better underwrite its heavy R&D and infrastructure spending.
For international observers the episode offers a snapshot of the dynamics shaping China’s AI industry: technical credibility and open‑source signalling can move markets, price discipline and enterprise monetisation are becoming front‑of‑mind priorities, and the capital markets remain a crucial lever for firms burning cash to chase AI scale. The path from model accolades to sustainable profit, however, remains narrow and contingent on user monetisation, compute economics and regulatory clarity.