For years the cloud business sold compute the way a utility sells electricity — by the socket, by the card, by the hour. A Beijing company wants to sell it the way a factory sells finished goods: by the degree, by the token, by the result.
That is the pitch behind JiuZhang Yunji, also known as DataCanvas (九章云极), which filed for a Hong Kong IPO and is bidding to become the city's first listed "NeoCloud (智算新云)" provider.
JiuZhang Yunji (九章云极) submitted its main-board prospectus to the Hong Kong Stock Exchange on 29 September 2026, with China Galaxy International (中国银河国际) as exclusive sponsor. The company is not yet listed; the deal size and price range will be set later in the offering process.
Key takeaways
- JiuZhang Yunji / DataCanvas (九章云极) filed a HKEX main-board prospectus on 29 September 2026, with China Galaxy International (中国银河国际) as exclusive sponsor.
- Revenue climbed from RMB 106.2M (≈ US$15.0M / HK$117M) in 2023 to RMB 1,098.2M (≈ US$154.7M / HK$1.21B) in 2025, a compound annual growth rate of 221.5%; the first half of 2026 added RMB 588.4M (≈ US$82.9M / HK$647M), up 92.0% year on year.
- The firm brands itself a NeoCloud (智算新云) provider — it architects and operates AI data-center (AIDC) projects and sells compute via subscription, pay-per-use and dedicated clusters, rather than simply renting GPUs.
- It invented a metering unit, 1 Degree of Compute (1度算力, or 1 DCU), defined as 312 TFLOPS at FP16 running for one hour, which it says was selected as a Ministry of Industry and Information Technology (MIIT) exemplary case.
- By 2025 revenue it ranked ninth among China's intelligent-computing service providers and was the only NeoCloud (智算新云) vendor in the top ten, per CIC (灼识咨询); it remains unprofitable on a reported basis, though adjusted losses are narrowing.
What "NeoCloud" actually means
Traditional vendors sell capacity: you lease a cluster, you manage it. JiuZhang Yunji sells outcomes. Its two business lines are Intelligent Computing Infrastructure Solutions — designing, procuring hardware for and integrating AI data centers (AIDC) — and Intelligent Computing Cloud (智算云服务) services, where compute, network and storage are pooled and delivered on subscription, pay-per-use or dedicated-cluster terms.
The shift is visible in the revenue mix. Legacy AI foundational-software (AIFS) work fell from 76.5% of revenue in 2023 to 0.1% in the first half of 2026, while intelligent-computing lines rose to 99.9%. The company deliberately does not own most data-center real estate; it partners with facility owners and compute-infrastructure funds, then handles architecture, hardware procurement, deployment and operations.
The growth curve
The numbers are steep. Revenue moved from RMB 106.2M (≈ US$15.0M / HK$117M) in 2023 to RMB 395.5M (≈ US$55.7M / HK$435M) in 2024 and RMB 1,098.2M (≈ US$154.7M / HK$1.21B) in 2025, a two-year rise of roughly tenfold. In the first half of 2026 it recorded RMB 588.4M (≈ US$82.9M / HK$647M), up 92.0% from a year earlier. Cloud customers grew from 2 in 2024 to 75 in 2025 and 84 in the first half of 2026, and the company says it serves more than 150 enterprise clients in total.
Profitability is a work in progress. Reported net losses were RMB 394M in 2023, RMB 458M in 2024 and RMB 336M in 2025, with a first-half 2026 loss of RMB 1,095M that included about RMB 1,068M from fair-value changes on financial liabilities. Stripping out those non-cash items, adjusted net losses narrowed from RMB 194M in 2023 to RMB 91.27M in 2025 and RMB 14.884M in the first half of 2026 — close to break-even on that basis.
Why a metering standard matters
JiuZhang Yunji's quiet innovation is billing. Its "1 Degree of Compute (1度算力)" standard anchors a unit of useful work — 312 TFLOPS at FP16 for one hour — and its "AI Factory (AI工厂)" system separates a training factory from a high-concurrency Token inference factory. The point is to turn a messy pile of GPUs into a measurable, schedulable product, the way kilowatt-hours turned a power plant into a billable utility.
That framing resonates with Chinese policymakers worried about wasted, idle compute. CIC (灼识咨询) estimates China's intelligent-computing service market reached about RMB 68.4B (≈ US$9.6B / HK$75.2B) in 2025 and could exceed RMB 277.5B (≈ US$39.1B / HK$305.3B) by 2030, a 32.3% compound rate, with inference cloud eventually overtaking training cloud.
The strategic tailwind
State-linked capital has already taken a seat. Beijing Information Industry Investment Fund and Beijing Artificial Intelligence Industry Investment Fund are among the shareholders, a signal of official interest in an independent, non-hyperscaler compute layer. As Chinese enterprises move from "do we have a model" to "can we run it cheaply at scale," a vendor that is not bound to one chip route or one big-tech ecosystem has a clear story.
Honest limitations
This is a prospectus filing, not a completed listing; no offer size, price range or trading date has been set, so any valuation discussion is premature. All financial figures are company-stated in the prospectus and not independently audited by a third party here. The "largest NeoCloud (智算新云) provider" and "ninth-ranked intelligent-computing service provider" claims come from CIC (灼识咨询), a paid industry research firm commissioned for the prospectus, not a regulator. The near-break-even adjusted result depends on excluding fair-value movements that are real accounting items. Revenue remains concentrated in AIDC integration projects, which carry hardware and execution risk, and the broader intelligent-computing sector is racing to add capacity that could outrun demand.
What readers can do now
- If you run workloads in China, benchmark JiuZhang Yunji's pay-per-use cloud against a hyperscaler and a self-built cluster on total cost per finished task, not list price per GPU-hour.
- For investors, track the eventual offer size and whether cornerstone investors signal confidence; a filing alone says little about demand.
- If you follow China's AI build-out, watch the "degree of compute" metering idea — if it spreads, it changes how utilization and waste get measured across the sector.