Authors | Hu Min, Zhang Jiamin | Editors | Bao Yonggang On March 18, a price hike announcement from Alibaba Cloud sent ripples through the cloud computing industry. However, what matters more is not the "price hike" itself, but the signal it sends: after years of price wars, cloud providers are systematically passing on cost pressures to the market for the first time. That afternoon, Baidu AI Cloud quickly followed suit; earlier, AWS, Google Cloud, UCloud, and Tencent Cloud had also successively raised their prices. Thus, a global wave of AI-driven computing power price hikes has emerged. "If we don't raise prices now, we really can't sustain the costs," a practitioner who has long tracked the cloud industry told Leiphone directly. Over the past year, the explosion of AI applications has pushed computing power demand to unprecedented heights; meanwhile, the supply of GPU capacity, data centers, and electricity has struggled to expand synchronously. When demand grows exponentially while supply remains physically constrained, price becomes a variable that is bound to be triggered sooner or later. However, this is not a "synchronized" price hike. Under the same cost pressures, different domestic cloud providers are making vastly different choices: some are only adjusting AI-related prices without touching their core business, some are raising prices across the board, and others are choosing to hold their ground, attempting to seize market share through price differentials.
01 "If We Don't Raise Prices Now, Everyone's Costs Will Become Unsustainable"
"Everyone actually knew this wave of price hikes was coming sooner or later," Zhang Xiangyu, a cloud sales representative, told Leiphone directly. In his view, this is not a proactive choice by cloud providers, but rather a passive adjustment. Changes on the demand side are the most direct driving force. "Last year, everyone was still calculating the call volume of chatbots; this year, no one calculates it that way anymore," said an AI application entrepreneur. As large models iterate rapidly in the directions of coding, agents, and multimodality, the computing power consumption of new-generation applications is multiplying. Industry practitioners estimate that the token consumption of AI agents is generally 100 to 1,000 times that of traditional chatbots. Generating a 15-second multimodal video can consume as many as 308,800 tokens. "We actually had no concept of token consumption before," Liu Xing, an OpenClaw user, told Leiphone. He gave a very intuitive example: the Qwen model gives away 1 million tokens to new users, but he just connected to the model and used OpenClaw to perform a very simple task—checking the weather in Shenzhen and writing it into a Feishu document—and consumed about 100,000 tokens. "I just asked one question, and 100,000 tokens were gone." After using it on the first day, Liu Xing quickly realized the severity of the problem. "Just by asking a few random questions, hundreds of thousands of tokens were quickly consumed." It was only after running a round with "Lobster" that he truly established a perception of token consumption and costs: he used to think tokens were a very abstract unit, but after using it once, he realized how fast tokens are consumed. To cope with the explosion in computing power demand, cloud providers have significantly increased their capital expenditures over the past year. However, "computing power is not something you can just add whenever you want," an insider at a cloud provider admitted. GPU capacity is constrained, electricity and data center resources face physical limitations, and cluster loss is generally between 30% and 60%. These investments cannot be immediately and fully converted into available computing power. Meanwhile, upstream supply chain costs have also begun to climb continuously. "Since the third quarter of last year, upstream prices have already been rising," a hardware channel distributor stated. Beyond GPUs, the prices of optical modules and PCBs have been successively raised, subsequently transmitting to storage and advanced processes, and even affecting CPUs in January this year. Now, this round of cost pressure has finally begun to transmit to the cloud service layer. Overseas cloud providers were the first to propose price hikes; as early as January this year, AWS had already taken action. In January this year, AWS raised the prices of GPU cloud services such as H100/H200/B100 by 15% (with some instances increasing from $34.6/hour to $39/hour). Google Cloud raised the prices of global data transfer services. The price per GB of data transfer in North America increased from $0.04 to $0.08, in Europe from $0.05 to $0.08, and in Asia from $0.06 to $0.085. Overseas cloud providers fired the first shot in cloud price hikes, followed by domestic cloud providers joining the trend. On February 11 this year, UCloud issued an announcement stating that due to the continuous intensification of global supply chain fluctuations, infrastructure costs such as core hardware procurement have seen significant and structural increases. They decided to adjust the prices of all product lines and services for all renewing and new users upward starting from March 1, 2026. This price adjustment covers major service categories including computing, storage, network, and security. Subsequently, Tencent Cloud announced that starting March 13, GLM 5, MiniMax 2.5, and Kimi 2.5 models would end their public beta and officially start billing, while prices for the Hunyuan series of models would be increased. On March 18, Alibaba Cloud released a price hike announcement stating that starting April 18, 2026, the prices of AI computing power and storage-related products and services would be raised. Among them, the prices of services related to computing cards such as T-Head (Alibaba’s T-Head Semiconductor) Zhenwu 810E would increase by 5%-34%, and the price of CPFS (Cloud Parallel File Storage) would increase by 30%. That afternoon, Baidu AI Cloud also announced that starting April 18, 2026, the prices of AI computing power-related products and services would be raised by about 5%-30%; prices for parallel file storage and related products would rise by roughly 30%. Thus, cloud providers such as Alibaba Cloud, Tencent Cloud, Baidu AI Cloud, and UCloud have all issued price hike announcements.
02 Under the Same Cost Pressures, Cloud Providers Take Three Different Paths
However, the market environments at home and abroad are not the same. Compared to overseas cloud providers who can more directly pass cost pressures downstream, domestic cloud providers always need to weigh repeatedly when it comes to "price hikes." They must consider both profit recovery and market share; once the pace gets out of control, they are very likely to be "intercepted" by competitors. For this very reason, "whether to raise prices, how to raise them, and by how much" has never been a simple cost problem in China, but a complex strategic choice. Under the same cost pressures, different providers have given different answers: behind this seemingly synchronized price hike action is actually a "business war" where each has their own calculations.
1. The Precision Pricing Faction
Represented by Alibaba Cloud, Baidu AI Cloud, and Tencent Cloud, the characteristic of this faction is: only raising prices where necessary, without touching the core business. From the adjustments already announced by Alibaba Cloud, the price hikes are mainly concentrated on AI training-related products, including GPU-based Elastic Compute Service (ECS), AI training-oriented Relational Database Service (RDS), and CPFS for AI Computing, etc. Most basic cloud products used daily by customers are not included in the price hike scope, such as standard ECS, standard RDS, and Object Storage Service (OSS). "This price hike actually has little impact on ordinary customers," cloud sales representative Zhang Xiangyu stated directly. Take CPFS (Cloud Parallel File Storage) as an example; it is a typical high-performance file storage product oriented towards AI training scenarios. Zhang Xiangyu explained to Leiphone: "You can understand it as a 'shared hard drive' specifically for large model training. Dozens or even hundreds of GPU cards need to read and write data simultaneously; ordinary storage simply cannot handle it. But a parallel file system like CPFS can ensure high throughput and low latency, preventing training from being slowed down by I/O." Therefore, such products are mainly used in large model training, high-concurrency inference caching, and scientific research and computing scenarios with high I/O throughput. Ordinary enterprise website hosting and business system development will almost never touch this level of capability. Only raising prices where necessary and not touching the core business—this consideration is not only about costs but also about competition. On the one hand, we are currently at a critical juncture, and the secondary market still has high expectations for Alibaba Cloud's growth; on the other hand, competition in the domestic cloud market remains fierce. Once core products are comprehensively priced up, it is very likely to directly impact the existing customer structure. "If we raise prices across the board and competitors take the opportunity to poach customers, the loss would outweigh the gain," an industry insider noted. Especially against the backdrop of players like Volcengine accelerating their catch-up, any fluctuation in price could translate into a shift in market share. What is even more intriguing is the timing of Alibaba Cloud's price hike announcement. According to the announcement, services purchased before April 18 will not be affected by this adjustment within the current billing cycle. This is equivalent to artificially setting a "window period" to urge customers to place orders in a concentrated manner before the price hike. And Alibaba's fiscal year ends on March 31. "This will actually bring a clear wave of front-loaded orders," a channel agent told Leiphone. Against this backdrop, the price hike is not just a price adjustment; to some extent, it is also further sprinting towards the FY2026 (fiscal year ending March 31, 2026) performance, and even long-cycle renewals. However, some industry insiders also pointed out that even if customers choose to lock in long-term contracts, the discount intensity is highly likely to be less than before. After all, upstream costs have genuinely risen. On the afternoon of the day Alibaba Cloud issued the announcement, Baidu AI Cloud also quickly followed suit. However, as of now, Huawei Cloud and Volcengine have not yet made public statements. "It's not that they don't want to raise prices; they are calculating a more complex account," a practitioner stated. For example, with Huawei Cloud, if they follow up with a price hike now, although it would help profit targets, whether it would affect the pace of their ongoing market expansion still needs to be weighed. As for Volcengine, their investment in AI is still very aggressive, and their goal this year is to catch up with Tencent Cloud's revenue scale, so they are probably still weighing whether or not to raise prices.
2. The Comprehensive Pricing Faction
In contrast, UCloud's actions are much more direct. Core products such as GPUs, storage, and CDN have seen price hikes across almost the entire line, which means that both large customers and small-to-medium customers will find it hard to avoid rising costs. "This is a comprehensive price adjustment in the true sense," a practitioner evaluated. Alibaba Cloud doesn't even dare to raise prices on such a large scale, so why does UCloud dare to do this? Leiphone communicated with multiple industry insiders, and the general consensus is that there are none other than three considerations behind their bold price hikes: First, the customer base is small, so they can afford to raise prices. "UCloud's overall cloud customer scale is much smaller compared to top-tier providers. Even if they raise prices across the board, the impact scope is limited and the risk is controllable," analyzed Liu Yu, an industry insider. Second, as a listed company, they need to protect profits. As a listed cloud provider, profit margin is one of the core indicators focused on by the capital market, and price hikes are the most direct means of profit recovery. Third, and most importantly—boosting the stock price. "It has almost caught every wave of hotspots," investor Zhang Han stated directly. From being the first to integrate OpenClaw to following up with computing power price hikes, every step UCloud takes lands on the nodes of the main AI theme. The stock price surged from over 20 yuan in January to over 40 yuan. "This is the most direct feedback of this strategy."
3. The Price-Cut Faction
Of course, some providers choose to hold their ground. Shortly after Alibaba Cloud issued the price hike announcement, JD Cloud quickly stated: not only will they not follow the price hikes in the short term, but they will also increase discounts on multiple products. "When everyone else is raising prices, not raising them is in itself a strategy," an industry insider evaluated. On the one hand, compared to top-tier providers, JD Cloud's investment scale in AI infrastructure is smaller, and the cost pressure is relatively limited. Even without raising prices, the impact on overall operations is small. On the other hand, going against the trend by announcing no price hikes is also a strategy to attract customers. Against the backdrop of continuously rising computing power costs, once a significant price difference emerges among different cloud providers, some cost-sensitive customers are highly likely to re-evaluate their choices. It can be seen that this round of price hikes is not a synchronized action, but a strategic choice by different providers based on their respective positions. Some are cautiously testing the waters, some are going all out, and others are holding their ground. Under the same cost pressures and AI opportunities, paths have begun to diverge. The price hike is just the beginning; the real competition has just changed its form.
The author has long tracked Alibaba Cloud, Tencent Cloud, Huawei Cloud, and Volcengine. Welcome to add the author's WeChat mindy1857 for communication. *Zhang Xiangyu and Liu Xing are both pseudonyms.