China is beginning to build its own version of the “chip-to-cloud” ecosystem, with a more visible distribution of power. In the first half of 2025, Baidu and Huawei together accounted for more than 70% of the Chinese GPU cloud services market, based on a Frost & Sullivan report that evaluates providers capable of controlling the entire value chain: from AI chips to computing clusters and cloud services.
In fact, the leadership is not evenly matched. Baidu is estimated to have reached a 40.4% market share and Huawei 30.1%, a snapshot confirming that competition is no longer just about designing accelerators, but about transforming thousands — or tens of thousands — of proprietary chips into a usable, stable, and scalable computing “pool”.
From standalone chips to the complete ecosystem: the “full stack” as a strategy
The key to this shift is that the market is rewarding those who go beyond just the silicon piece. In China, major tech conglomerates and their semiconductor divisions are pushing toward full stack: hardware, software, development platforms, and cloud products. The goal is clear: reduce external dependence and build an alternative to the dominant AI ecosystems where software and tools are as critical as raw chip performance.
The report highlights this transition: leaders are no longer competing solely on specifications but are integrating end-to-end infrastructure. In this context, “GPU cloud” doesn’t refer to any company renting GPUs, but rather to those that control the entire value chain and can ensure supply, operation, and technological evolution without relying on third parties for critical components.
The caveat: technical and market limits still exist
However, the outlook is not overly optimistic. Frost & Sullivan notes that China’s domestic GPU market still faces restrictions in hardware performance, software ecosystems, system integration, and commercial operation. Only a limited number of providers have achieved large-scale deployment.
This nuance is important because it describes the real state of China’s accelerator industry: it has moved from a “contingency plan” to a “serious plan,” but is still developing habits, compatibility, and tools that turn a GPU into an attractive platform for developers and companies.
Nonetheless, the report’s projection is ambitious: over the next 3 to 5 years, domestic GPUs could shift from being a “strategic backup” to a “core pillar” of the country’s AI industry, supported by an increasingly convincing value proposition for buyers: cost-performance ratio in real training and inference workloads.
Baidu and Kunlunxin: integration as an advantage
In Baidu’s case, leadership is attributed to the integration of its AI computing platform Baige with GPUs developed by its chip division Kunlunxin. This combination — platform plus proprietary chip — exemplifies the vertical “lock-in” that enables service scaling: the provider is not dependent on external accelerator availability and can optimize software, drivers, compilers, and deployment tailored to its own hardware.
A key figure is significant: Kunlunxin reportedly shipped nearly 70,000 units in 2024. This indicates that it’s no longer just prototypes or pilot projects but sufficient volume to build clusters and gain operational experience.
Additionally, Kunlunxin — originally established internally in 2012 — has been gaining independence over time. That process fuels speculation about the next step: entering the capital markets.
The IPO wave: chips, cloud, and national narrative
The interest extends beyond industry to finance. Amid efforts toward self-sufficiency and growing demand for computing power, several Chinese companies are accelerating their IPO plans. Baidu has announced that it has confidentially filed for an IPO of Kunlunxin in Hong Kong, although the final size and structure remain undefined.
This move follows a dual logic: on one hand, to raise capital and increase visibility for a strategic unit; on the other, to reinforce the message that China aims to develop a complete supply chain for AI — from chips to cloud services.
The competition is also active. The article notes that Enflame, another Shanghai-based Chinese designer, has completed its “tutoring” process ahead of a potential listing on the STAR Market, China’s main tech segment. This suggests that China’s domestic accelerator market is entering a more mature phase: more capital, greater pressure for actual sales, and a focus on execution.
Huawei: the second pillar of China’s “GPU cloud”
Huawei emerges as the second major player, with a 30.1% share. Its presence reinforces the idea that China’s “GPU cloud” is consolidating around large-cap companies capable of operating massive infrastructure and maintaining a complete product cycle from chip to service, with the strength to support deployments and attract big clients.
In this context, market share isn’t just a market statistic — it’s a reflection of who’s succeeding in transforming computing into an accessible service, a goal that requires much more than designing chips. It demands support, tools, documentation, integration into business workflows, and an experience that doesn’t penalize users for diverging from the dominant standard.
Implications for China’s AI ecosystem
For Chinese companies training or deploying models, the shift toward “chip-to-cloud” providers promises greater supply continuity and a domestic path to scaling AI. For the technical ecosystem, it shifts the focus to software, compatibility, and tools, because that’s where developers’ habits are won or lost.
For the global market, the message is clear: demand for AI computing is driving the creation of comprehensive alternatives, not just chips. China is betting on an industry that doesn’t rely on foreign imports for essential components, with Baidu and Huawei emerging as the primary beneficiaries of this strategy for now.
FAQs
What does “GPU cloud” mean in the context of China and domestic chips?
It refers to providers that control the entire “chip-to-cloud” value chain: proprietary accelerators, computing clusters, and ready-to-use cloud infrastructure for customers.
Why do Baidu and Huawei dominate this market compared to other players?
Because they combine proprietary chips with the capacity to operate large-scale data centers and offer cloud services at scale, requiring integration and operational muscle.
Can Chinese GPUs now compete directly with the world’s best?
There are still limitations in performance, software, and integration, but improvements in cost-performance in real workloads and progress toward larger deployments are evident.
What does it mean that Kunlunxin and other designers are seeking IPOs?
It typically signals a growth phase: a need for capital, pressure for growth, and the goal of establishing themselves as strategic players in AI infrastructure.
via: scmp