Author | Hu Min, Editor | Bao Yonggang
"The cloud computing industry has truly changed over the past year," said Liu Qi, a cloud sales representative.
In previous years, whenever he gathered with peers, the conversation always revolved around a heavy question: "Is it time to leave?"
The seemingly endless downturn has dragged everyone into an endless internal friction: product and R&D teams working late to revise proposals, salespeople exhausted by socializing, partners struggling to survive in the cracks, and executives burdened with growth pressure—all are shrouded in the same anxiety and fatigue.
Worse still, large-scale integration projects and bottomless price wars have collectively plunged cloud vendors into a "profit black hole." When investors realized that this model of sacrificing profits for scale was unsustainable, they voted with their feet, causing the valuation of the cloud computing sector to hit rock bottom. This golden track, which once represented the future, was once abandoned by the capital market, and its halo has gradually dimmed.
"This industry might really be over," Liu Qi and his friends often say.
But has the story of cloud computing really ended here?
If by 2025, investors still view this industry with past perspectives, they are highly likely to miss a critical turning point. Because AI (Artificial Intelligence) is pushing cloud computing back into the spotlight—it not only holds the promise of repairing the industry's revenue and profits but is also dramatically reshaping the market landscape.
01 In the First Half, Why Were Cloud Giants Abandoned by the Capital Market?
China's cloud computing industry was once gradually abandoned by capital and valued at an extremely low Price-to-Sales (PS) ratio. Cloud businesses were even excluded from the market capitalization estimation systems of tech giants, all stemming from a widening "growth rift."
When the dividend of internet traffic peaked and the backbone migration of enterprises to the cloud was basically completed, the natural growth curve of the market inevitably flattened. However, based on inertial paths and capital market expectations, the high-growth targets set by major cloud vendors were not adjusted accordingly.
On one side is the real market gradually hitting the ceiling; on the other side are the unyielding high-growth mandates. When incremental growth cannot be obtained externally, the pressure shifts entirely inward, evolving into a cruel "stock slaughter," where orders are constantly "moved back and forth" among different vendors and different teams. Today you snatch a client from a competitor, and tomorrow your backyard might be raided by another.
Some time ago, T3 Mobility fully migrated to Tencent Cloud, but many people do not know that before fully moving to Tencent Cloud, T3 Mobility had already shifted among several cloud vendors.
T3 Mobility was initially on Alibaba Cloud, but 2020 was precisely when Huawei Cloud was aggressively expanding the market, and they quickly snatched this client. For the migration, Huawei Cloud once dispatched over 40 people to be stationed on-site. However, despite going to great lengths, Huawei Cloud failed to keep T3 Mobility. Through group investments, traffic allocation, and other methods, the Tencent Cloud Smart Mobility team directly moved T3 Mobility's business from Huawei Cloud to Tencent Cloud by 2025. It is understood that the volume of this migration was at least tens of millions.
This war of mutual slaughter in the existing market has spawned three systemic distortions:
The First Distortion: The Poisonous Price War, Where Technology and Services Are Reduced to Discount Figures.
The competition among cloud vendors has been directly simplified to the most primitive dimension: price. Over the past few years, almost all cloud sales representatives have felt that this price war has been intensifying.
"Initially, it was just a price war among Alibaba Cloud, Tencent Cloud, and Huawei Cloud, but the discounts offered to many clients were basically above 50%. Later, the entry of carrier clouds such as Tianyi Cloud directly dragged the price war down to 10%-20%."
In many cloud industry chat groups, it is often seen that someone's note reads: Tianyi Cloud at 10% of the original price, indicating extremely severe price involution.
"Now when meeting clients, the first sentence is 'How many percentage points can you offer?'" a sales representative with years of experience frankly admitted.
The evil consequence of the price slaughter is that the entire industry has collectively fallen into a profit black hole. The profits of several domestic cloud vendors are basically scraping the bottom. Alibaba Cloud achieved profitability first, but its latest annual report shows that its profit margin is still less than 9%. Tencent Cloud has been shouting for profits for several years, and it was not until 2025 that it surfaced and achieved break-even. As for Huawei Cloud, its pace is even slower; it is currently undergoing various major adjustments, striving to recover its losses.
The Second Distortion: The Farmer and the Viper Ecosystem, Where Partnerships Shift from Symbiosis to Devouring.
A healthy cloud computing ecosystem should be a "tropical rainforest," where the platform nurtures partners such as ISVs (Independent Software Vendors), channel distributors, and SIs (System Integrators), and the latter help the platform prosper. However, under the pressure of stock slaughter, this rainforest is degrading into a "hunting ground" at an astonishing rate. The relationship between the platform and partners has undergone a cruel trilogy from "utilization" to "plunder," and finally "abandonment":
Step 1: At the beginning of cooperation, partners are reduced to "sales tools."
It is justifiable for reseller partners to help cloud vendors sell products, but what is bizarre is that ISV and SI partners have all become sales tools for cloud vendors. ISV partner Zhang Yu mentioned that for SaaS (Software as a Service) companies to enter the cloud marketplace, if they want the cloud vendor to issue consolidated invoices, they must achieve a certain performance target; for service providers to enter the directory, they need to sell cloud resources worth at least several million, or even tens of millions, first.
Step 2: During the cooperation, partners become "objects of replication."
When the partner's products and solutions have successfully validated the market, the platform's "learning" begins.
Zhang Zhe, the founder of an ISV focusing on financial risk control, has a deep understanding of this. In 2021, his company jointly launched a solution with a certain cloud giant to explore the market together. After the solution matured and the scenarios were validated, in 2023, the cloud giant launched its self-developed competing product, leveraging its platform advantages for bundled sales and price suppression, rapidly marginalizing Zhang Zhe's product.
"We taught the platform how to hunt, and then we became the prey ourselves," Zhang Zhe said helplessly. This "Farmer and the Viper" script is repeatedly staged, leading to a complete bankruptcy of trust.
Step 3: In the later stage of cooperation, partners face "purging and elimination."
When market exploration enters the deep-water zone and incremental growth becomes difficult, the value of "allies" who relied on partners to expand territories in the early stages begins to depreciate. Vendors frequently adjust channel policies, taking back the docking rights of core clients to direct sales, causing the revenue of several provincial top channel distributors who once made significant contributions to be halved overnight. From "joint exploration" to "kicking away the ladder," the fragility of partnerships is fully exposed.
The Third Distortion: The Collapse of the Bottom Line, Where Corruption and Chaos Erode the Foundation.
When the pressure of growth distorts all actions, the organization's bottom line begins to collapse comprehensively. In order to achieve those impossible numbers, "invigorating" ultimately slides into "foul play," resulting in systemic chaos.
First, "inflating transaction volumes" has become a common practice. This is already an unspoken rule among cloud practitioners. Some practitioners claim that some teams have inflated transaction volumes to the level of tens of billions. Even so, every year there are people testing the edge of danger and still being penalized for fake orders.
Second, direct sales and channels "collude," conspiring together to jointly extract high rebates and rewards from cloud vendors.
A typical case is that a top cloud vendor once discovered large-scale fake performance in its South China region. Upon investigation, it was found that direct sales personnel colluded with multiple agents to fabricate a large number of IoT (Internet of Things) device orders in the name of helping to "complete performance targets," jointly arbitraging. Ultimately, this led to the vendor deducting and penalizing huge commissions from the involved agents, while the core operators had already absconded with the funds.
Such incidents are not isolated cases; they expose how the cooperation chain, which should have co-created value, has mutated into a hotbed of collusive fraud under the distorted assessment mechanism.
02 In 2025, Cloud Giants Collectively See Surging Performance Due to GPUs
Just when everyone was extremely disappointed with the involution of the cloud market and wanted to escape, the cloud computing industry ushered in a turning point in 2025 because of AI. AI has brought new revenue increments to cloud giants, from GPU (Graphics Processing Unit) resources to MaaS (Model as a Service) calls, and then to AI SaaS tools.
In 2025 Alone, GPU Computing Power Has Brought "Timely Rain After a Long Drought" to Cloud Vendors.
Cloud sales representatives feel this most deeply.
Zhang Yi, a cloud sales representative at Volcengine, frankly admitted that at the beginning of the year, he was still worried about doubling his performance of over 20 million. But after the Spring Festival, the wind direction changed suddenly. "The market set off a wave of DeepSeek deployment boom, and inference resources such as H20 became hot commodities overnight."
Zhang Yi stated that in the first half of the year, his team successively secured orders from multiple giants including Tencent, Alibaba, Weibo, and Xiaohongshu. "Up to now, we basically have no inventory left to sell."
Industry insider Zhang Yu told Leifeng.com that throughout 2025, Volcengine's GPU sales were the most fierce, for several reasons:
First, they indeed have a lot of cards in hand. According to a rough estimate by analyst Liu Yu's research, ByteDance alone has 480,000 H20 cards.
Second, having cards alone is not enough; they also need to solve the problem of being willing to sell resources externally.
At present, the large model competition is in full swing. All the cards hoarded by internet vendors are prioritized for their own businesses, whether it is Tencent, ByteDance, or Alibaba. For example, within Tencent, its core businesses are WeChat, gaming, and real-time audio and video. These businesses have extremely high requirements for stability and continuity. Each business line must reserve baseline computing power for its own traffic fluctuations in the long term. Even if resources are idling during certain periods, each business division is unwilling to release them for external rental.
But ByteDance is completely different. In 2025, ByteDance proactively released a portion of its cards for external rental, and the core reason is still to attract traffic. Originally, Volcengine was indeed a latecomer, lacking good leverage to break through clients in the cloud market. When the market demand for GPUs was so strong last year, Volcengine's senior management decisively seized the opportunity and released some resources to occupy the market.
Moreover, to win this battle, at the beginning of 2025, Volcengine adjusted its sales assessment, requiring that this year they must focus on selling GPUs and the three major public cloud products, and in terms of price, they can also be lower than competitors.
Of course, it is not just Volcengine that has good GPU sales; the other several vendors are the same.
Analyst Zhang Fan's research estimated that in the first half of 2025 alone, Volcengine's GPU revenue might have reached nearly 3.5 billion, with a synchronized growth rate exceeding 200%. In the first half of 2025, Alibaba Cloud's GPU revenue was estimated to be about 5.5 billion, with a synchronized growth rate also exceeding 100%. Additionally, because Tencent Cloud's internal card resources are relatively tight, it sold relatively less than Volcengine and Alibaba, but its overall growth rate is almost the highest among all products.
Additionally, Huawei Cloud sales representatives told Leifeng.com that although Ascend computing power is highly priced, due to strong market demand, their external rental of Ascend computing power this year has also been very successful, with its main buyers being carrier customers. Furthermore, Baidu Cloud, Kingsoft Cloud, UCloud, etc., have also secured many business orders in this wave of computing power demand.
In 2025, almost all cloud vendors have seen a significant increase in performance due to GPUs. And in 2026, with the accelerated implementation and application of large models, the GPU growth rate of cloud vendors will only increase, not decrease.
03 The MaaS and AI SaaS Battle Will Be Fought in 2026
But GPU computing power is just the first act of AI cloud. The real battlefield in 2026 will shift to MaaS and the upper-layer AI SaaS tools.
Of course, some still question the profitability of MaaS. Some practitioners believe that the current low unit price of tokens and fierce competition are difficult to support scaled profitability for cloud vendors.
But by the end of 2025, cloud vendors have become increasingly aware that what is being fought for here is not only current revenue but also core clients in the future AI cloud era.
"If we don't hurry to find these clients now, we are highly likely to miss the dividend of the next 3-5 years," several cloud sales representatives frankly admitted.
At the end of 2025, many cloud sales representatives have already started to feel anxious. Not only cloud sales, but also cloud vendors, such as Volcengine and Alibaba Cloud, have been making intensive layouts at the end of the year. Multiple industry insiders predict that in 2026, the cloud computing industry is highly likely to set off a wave of MaaS commercialization battles.
Leifeng.com learned from multiple industry insiders that in 2026, Volcengine's strategy on MaaS has shifted from "exploration" to "strong attack." Volcengine's revenue target for MaaS in 2026 is very aggressive, aiming to achieve at least 5-6 times growth. In 2025, Volcengine's MaaS and AI tool revenue was approximately 2-3 billion, which means that in 2026, they need to achieve 10-18 billion.
Because they need to generate revenue from MaaS and AI tools, Volcengine has also adjusted its sales assessment this year. Last year, sales could still sprint using GPUs, but this year it is no longer possible; all sales must sell large models and related AI tools.
"The large model performance task for each sales representative is at least 10 million as a baseline. For some sales representatives who have existing large model clients in hand, it is not surprising that their large model tasks are 20-30 million," Volcengine sales representative Liu Yang told Leifeng.com.
Alibaba Cloud is the same. Last year, Alibaba Cloud's sales basically all carried a daily call target of 100 million tokens, and this target is estimated to be more aggressive in 2026. According to Leifeng.com, at the end of this year, Alibaba Cloud's sales line has already made some adjustments to prepare for this battle in 2026. To learn more about Volcengine and Alibaba Cloud, feel free to add the author's WeChat mindy 1857 for communication.
How exactly is MaaS commercialized? Is it still about promoting large language models?
In fact, everyone has long discovered that this path does not work, and MaaS revenue grows very slowly.
"In 2024, after promoting it for a while, everyone quickly found that the revenue ceiling was very low. The value of a large model client consuming 10,000 RMB per month is only equivalent to the volume of a client consuming tens of millions annually in the traditional public cloud era," said cloud sales representative Wang Jie. "At that time, whether it was direct sales or channels, no one had enough motivation to promote large models."
Cloud vendors also discovered the commercialization difficulties of large language models, so they began to shift to promoting voice models.
Many sales representatives have started to implement them in standardized interactive scenarios such as intelligent collection and customer service outbound calls. For example, some consumer finance companies need to handle millions of collection outbound calls every day. The traditional manual customer service model is highly costly, and traditional collection robots can only recite by rote, resulting in poor collection effects. And this is the landing scenario for voice large models.
Leifeng.com learned that some cloud sales representatives have indeed found many such clients, and MaaS revenue has started to see some breakthroughs. However, these intelligent collection scenarios have one major risk: regulation. "Once a client complains, the line is blocked, and you get fined."
Therefore, the vast majority of cloud sales representatives have not pinned their hopes on voice models, but have placed heavy bets on "high-consumption" scenarios such as text-to-image and text-to-video. Because they are rapidly landing in multiple industries with clear demand and strong willingness to pay, such as: AI comic dramas, educational content generation, smart hardware, creative design applications, etc.
Liu Yang revealed that Volcengine's internet sales teams are now almost fully "sweeping" these markets. "As long as a new short drama company is registered, they immediately associate it to their name, and from time to time, there are clients whose monthly consumption surges to hundreds of thousands."
It is understood that Volcengine sales have an incomparable advantage when targeting AI short drama clients: they have the support of Hongguo resources. If clients choose the Doubao model, they can also simultaneously receive traffic support and even investment from its affiliated Hongguo short drama platform.
Undoubtedly, these scenarios will accelerate the birth of clients with annual consumption of millions in 2026. At the same time, these tracks will become increasingly competitive, and cloud sales representatives all want to squeeze their heads in to enter.
04 How to Fight the "Profit Turnaround Battle" for Cloud Vendors?
GPU and MaaS are fiercely contested, but all wars ultimately have to calculate the economic account.
Over the past decade, the core narrative of the global cloud computing industry has been "enclosing land by riding horses, scale first." Behind the growth myths of giants such as AWS and Alibaba Cloud are continuous price wars and constantly thinning profit margins. Cloud computing has been turned into a "bitter business with astonishing scale but meager profits," and the capital market's valuation logic has long revolved around revenue growth rather than profitability.
However, this decade-long model is expected to usher in structural improvements in 2026. The alleviation of profitability pressure will benefit from the combined effect of short-term and long-term forces, forming a rare "double click."
In the Short Term, AI Computing Power Price Increases Are Highly Likely to Bring Immediate Profit "Hemostasis" to Cloud Giants.
At the end of 2025, global top cloud vendors have initiated price increases for AI-related services.
AWS has increased the prices of GPU cloud services such as H100, H200, and B100 by 15% (some instances increased from $34.6/hour to $39/hour). Alibaba Cloud is also following suit with AI computing power price increases, but reducing the prices of basic cloud services (such as ECS instances) by 10%-12% to attract small and medium-sized clients. Subsequent price increases will extend to the PaaS (Platform as a Service) layer, with businesses such as AI Agents and sandboxes expected to see price increases of 5%-8%.
Why dare to raise prices? It is still because the demand is too rigid.
This is not just an individual choice by vendors; behind it is a clear "AI inflation" transmission chain.
According to research reports from institutions such as Guolian Securities, starting from the first half of 2025 when memory chips took the lead in price increases, to the follow-up of CPUs (Central Processing Units) in early 2026, inflationary pressure is being transmitted top-down along the AI industry chain. Now, the pressure has finally reached the cloud. Marked by AWS taking the lead in raising prices, the industry iron law of "cloud service prices only drop and never rise" that has lasted for twenty years has been thoroughly broken.
Perhaps, the price increase of this high-growth, high-gross-margin business will directly "transfuse blood" to the 2026 profit statements of cloud vendors, becoming the most direct lever to improve profitability in 2026.
Of course, this is only short-term. From a long-term perspective, the reason why the profits of cloud giants may be repaired is that AI is driving the industry to irreversibly return to the public cloud, which will fundamentally repair the biggest "profit black hole."
In the past, in order to snatch a large order from a bank or a large state-owned enterprise, cloud vendors had to act as "general contractors": advancing funds, custom development, and dispatching personnel to be stationed on-site, with project cycles often lasting one or two years. "For that kind of order, the contract amount looks scary, but in the end, the profit left might not be as much as a medium-sized internet company's annual public cloud consumption," frankly admitted a solution architect who has experienced many such projects.
AI is ending this kind of "hard and tiring work." The speed of large model iteration on a monthly basis makes private deployment neither economical nor smart. "You spend a lot of money to buy and deploy it, and the version is already outdated. Smart clients now choose to call it on the public cloud." This irreversibly locks high-value AI workloads on the public cloud.
In the future, what cloud vendors sell will no longer be the manpower and time for custom development, but "software" such as model calls and intelligent tools with extremely low marginal costs. The health of the business model has made a qualitative leap.
Therefore, the profitability story for cloud vendors in 2026 is very clear: in the short term, rely on AI computing power price increases to quickly "recover blood"; in the long term, rely on the public cloud product model to achieve a complete "rebirth."
With the superposition of these two forces, the time has also come for the capital market's logic in viewing cloud computing to inevitably change.
05 The AI Cloud Battle Will Bring Huge Shocks to the Cloud Market Landscape
AI will completely change the existing revenue and profit dilemmas of cloud vendors, but at the same time, it is also reshaping the cloud market landscape.
Cloud practitioner Zhang Zhen believes that the rise of Volcengine will be the biggest highlight of this AI cloud competition.
The original Volcengine was always called a latecomer, and many practitioners were not optimistic about it, but AI has given Volcengine a game-changing trump card.
Originally, clients basically would not think of Volcengine when bidding, but now when doing AI projects, "some clients specifically ask to test Doubao," a sales representative said. Now Volcengine is using its highly cost-effective model capabilities, especially in image and video generation, as a "sharp knife" to cut into new scenarios such as AI short dramas and creative applications. Once successful, it can leverage the momentum to drive clients' basic cloud consumption. And this is also why Alibaba Cloud is so nervous about Volcengine.
According to Leifeng.com, now in the AI cloud, Alibaba Cloud's number one target opponent is no longer Tencent Cloud or Huawei Cloud, but Volcengine.
Volcengine is very aggressive, but many practitioners still believe that Alibaba Cloud's position remains the most stable because it holds almost all the key cards.
First, it sits on about 4 million enterprise clients, which allows its sales to perceive the market temperature earliest. "When snatching AI projects from large clients, they are often one step ahead," a competing sales representative frankly admitted. "As soon as the client has an idea, their proposal might already be on the table."
And in terms of technology, Alibaba is the only domestic giant that has truly completed the full-stack self-research of "chip-model-cloud." From reducing costs and increasing efficiency at the underlying Hanguang chip level, to making up for application shortcomings with the Tongyi Qianwen model, to binding large computing power consumers such as Moonshot AI and Zhipu Al mostthrough strategic investments, Alibaba Cloud is building a technology and ecosystem closed loop that is difficult to replicate. The invested enterprises alone contributed over 4.5 billion in revenue for it in fiscal year 2025. This gives it the deepest trump card in a protracted war.
Besides Alibaba Cloud and Volcengine, some practitioners believe that in this wave of AI cloud battles, Baidu Cloud may have the opportunity to further narrow the gap with the top-ranked players.
Baidu Cloud and Volcengine have completely different strategies in AI. Baidu Cloud focuses on large-scale AI data centers and industry large model projects. These projects often have high amounts, long cycles, and deep customization, which is exactly what other vendors consider "heavy" and are unwilling to fully invest in, but it is where Baidu can best leverage its technological accumulation.
In contrast, the pressure on Tencent Cloud and Huawei Cloud is more obvious, as they are walking on two different paths of defense and counterattack.
(The author has long tracked Alibaba Cloud, Tencent Cloud, Huawei Cloud, and Volcengine. Feel free to add the author's WeChat mindy1857 for communication.)
In order to hold onto profitability targets, Tencent Cloud has basically given up the price war on traditional cloud products in recent years. Although this makes the financial statements look better, it also makes it less fierce than Alibaba and Volcengine when snatching clients, resulting in the loss of some market share. In the most critical "large model" track, its Hunyuan large model has failed to lead in general capabilities. However, the biggest card in Tencent Cloud's hand is actually not the model itself, but its unparalleled "product ecosystem." Creating good AI SaaS products might be the key battlefield for its future turnaround.
As for Huawei Cloud, in the short term, it is in an adjustment period full of contradictions. "Profitability" is the overriding iron law at present. This has directly led to two things: first, being unable to go all out in the brutal price war, and second, the contraction of marketing expenses. According to sales feedback, even entertainment expenses are now strictly limited. This "tightening the belt" approach, while maintaining profits, also brings the real risk of client churn.
In terms of AI, the problem Huawei Cloud faces now is, first, the overall strength of Huawei's large model still has a relatively large gap with ByteDance and Alibaba; second, in AI commercialization, Huawei Cloud currently still faces a situation of "fighting with its left hand against its right hand" in some businesses. Cloud sales representative Wang Feng said that it forms a potential internal competitive and cooperative relationship with its sister department Ascend. When facing the same client, the people from the two teams may end up in a situation of competing against each other.
Huawei Cloud's current situation may be the most difficult, but this does not mean that Huawei Cloud will fall behind. Not long ago, Huawei Cloud's product and R&D line was merged into ICT. Perhaps this is also the solution given by Huawei's senior management to the current market competition. In the future, by selling cloud products together with hardware and creating joint solutions, this might be a way out for Huawei Cloud in the future.
06 The AI Cloud Battle Has Just Begun
For Chinese cloud computing practitioners, 2025 is the year when the cloud computing industry has finally endured to the end and seen hope.
The dilemmas that have plagued the industry in the past few years, such as growth deceleration and profit black holes, have finally shown signs of improvement in 2025.
Over the past year, GPU computing power has been like timely rain after a long drought, allowing sales representatives to get real orders again; MaaS services, from the questioned "small business," are showing the potential to impact tens of billions in annual revenue; and more profoundly, AI is irreversibly "locking" high-value workloads on the public cloud at a monthly iteration speed, which fundamentally shakes the old model that led to meager profits for cloud vendors.
The turning point has arrived, but the battle is just beginning. 2025 is not the endgame, but the starting point of a brand-new competition. Alibaba, Tencent, Huawei, Baidu, and Volcengine will engage in a new round of wrestling on the new arena of AI cloud.
*Zhang Yi, Liu Yu, Zhang Fan, Wang Jie, Zhang Zhen, and Wang Feng are all pseudonyms.