Kingsoft Cloud Holdings - Earnings Call - Q1 2025

Transcript

Operator (participant)

Thank you for standing by. Welcome to the Kingsoft Cloud's first quarter 2025 earnings conference call. At this time, all participants are in a listen-only mode. After the speaker's presentation, there will be a question-and-answer session. To ask a question during the session, you will need to press star 1, 1 on your telephone. You will then hear an automatic message advising your hand is raised. To whisper your question, please press star 1, and 1 again. Please be advised that today's conference is being recorded. I would now like to hand the conference over to your speaker today, Nicole Shan, IR Director of Kingsoft Cloud. Please go ahead.

Nicole Shan (IR Director)

Thank you, Arpreter. Hello everyone, and thank you for joining us today. Kingsoft Cloud's first quarter 2025 earnings release was distributed earlier today and is available on our IR website at ir.ksyu.com, as well as on the PI Newswire services. On the call today from Kingsoft Cloud, we have our Western CEO, Mr. Zou, and the CFO, Mr. Henry He. Mr. Zou will review our business strategies, operations, and company highlights, followed by Mr. He, who will discuss the financial performance. They will be available to answer your questions during the Q&A session that follows. There will be consecutive interpretations. Our interpretations are for your convenience and reference purposes only. In case of any discrepancy, management statements in the original language will prevail.

Before we begin, I'd like to remind you that this conference call contains forward-looking statements within the meaning of Section 21E of the Securities Exchange Act of 1934, as demanded and as defined in the U.S. Private Securities Litigation Reform Act of 1995. These forward-looking statements are based upon management's current expectations and current market and operating conditions, and relate to events that involve known or unknown risks, uncertainties, and other factors, all of which are difficult to predict, and many of which are beyond the company's control, which may cause the company's actual results, performance, or achievements to differ majorly from those in the forward-looking statements. Further information regarding these and other risks, uncertainties, or factors are included in the company's filings with the U.S.

SEC, the company does not undertake any obligation to update any forward-looking statements as a result of new information, future events, or otherwise, except as required under applicable law. Finally, please note that unless otherwise stated, all financial figures mentioned during this conference call are denominated in RMB. It's now my pleasure to introduce our CEO, Mr. Zou. Please go ahead. Thank you.

Zou Tao (CEO)

大家好，欢迎参加金山云2025年第一季度业绩电话会。我是金山云CEO，周涛。本季度我们继续扎实推进公司业务，以高质量可持续发展为目标，聚焦AI关键领域。首先，我们的收入实现同比增长11%，达到19.7亿元。工业云和行业云均实现了同比增长，其中工业云同比增长14%，实现了收入13.5亿元人民币。其次，我们继续以AI引领进步。本季度AI业务账单收入达5.3亿元，同比增长超过200%，环比增长11%，占工业云收入比例继续提升至39%。此外，本季度我们正在以更灵活的资金投入方式加紧建设算力集群，预计在二季度上线进行服务，进一步加速AI收入的提升。第三，作为小米金山生态的唯一战略云平台，我们与生态的业务合作顺利推进。本季度小米和金山生态收入达5亿元，同比增长50%，占收入比例进一步提升至25%。我们与小米金山在AI领域的合作持续推进且日益深入，充分整合双方的优势资源，与小米金山共同扩张AI时代的云基础设施。最后，从利润水平来看，本季度调整后毛利润为3.3亿元，同比增长9.6%，调整后EBITDA率达到16.2%，同比提升14.3个百分点，主要得益于AI收入比重的持续提升。但我们也意识到利润环比出现了波动，经调整毛利率环比下降2.6个百分点，为16.6%。毛利率的下降主要由于行业云收入比重下降导致的贡献利润减少，以及算力成本前置的变动影响。调整后经营利润受毛利润下降影响，本季度亏损5,581万元人民币，调整后经营利润率为负的2.8%。相较去年同期的亏损7.2个百分点，提升了4.4个百分点，环比由盈利转为微亏。尽管本季度的财务业绩短期内出现环比波动，市场变化和供应链因素交织，但从长远视角洞察，我们正坚定地沿着长期向好的战略布局稳步迈进。在小米金山生态合作算力布局、AI应用部署方面筑牢根基，推动公司整体AI云服务的战略布局。

Hello everyone, thank you and welcome all for joining Kingsoft Cloud's first quarter 2025 earnings call. I am Zou Tao, CEO of Kingsoft Cloud. This quarter, we continued to steadily advance our business with a target on high-quality and sustainable development, centering on key areas of AI. First, we recorded year-over-year revenue growth of 11%, reaching RMB 1.97 billion. Both public cloud and enterprise cloud achieved year-over-year growth, among which public cloud increased by 14%, reaching RMB 1.35 billion. Second, we continued to drive progress with AI. This quarter, AI gross billing reached RMB 525 million, representing a year-over-year increase of over 200% and a quarter-over-quarter growth of 11%, further contributing 39% of public cloud revenue.

In addition, this quarter, we are accelerating the construction of our computing clusters with more flexible capital deployment, which is expected to launch service officially in the second quarter, further boosting our AI business revenue. Third, as the only strategic cloud platform of the Xiaomi and Kingsoft ecosystem, our business cooperation with the ecosystem progressed smoothly. This quarter, revenue from Xiaomi and Kingsoft ecosystem reached RMB 500 million, up 50% year-over-year, with its contribution to total revenue further increasing to 25%. By fully integrating the strengths of both parties and jointly expanding cloud infrastructure for the AI era, our collaboration with Xiaomi and Kingsoft in the AI space continued to progress and deepen. Finally, in terms of profitability, this quarter, our non-GAAP gross profit was RMB 327 million, representing a year-over-year increase of 9.6%.

Non-GAAP EBITDA margin reached 16.2%, an increase of 14.3 percentage points year-over-year, mainly attributable to the continued increase in the proportion of AI business revenue. However, we also witnessed that our profit has experienced some fluctuation quarter over quarter, while non-GAAP gross margin declined by 2.6 percentage points quarter over quarter to 16.6%. The decline in gross margin was mainly due to declined profit contributions from a lower proportion of enterprise cloud revenue, as well as the impact of front-loaded investment of computing resources. The non-GAAP operating profit was impacted by the decline in gross profit, resulting in the loss of RMB 55.8 million this quarter. Non-GAAP operating margin was negative 2.8%, representing an improvement of 4.4 percentage points compared with a loss of 7.2 percentage points in the same period last year, and turned from a profit to a slight loss quarter over quarter.

Despite quarter-over-quarter fluctuations in financial performance this quarter and the headwinds of both market pressure and supply chain uncertainties, we remained firmly on track with our long-term strategy and continued to move forward with confidence. We have strengthened our foundation in ecosystem cooperation, computing infrastructure deployment, and AI application, advancing the strategic layout of our company's overall AI cloud services.

下面我向大家具体介绍2025年第一季度的业绩情况。工业云方面，本季度实现收入13.5亿元，同比增长14%。AI业务作为重点的增长驱动力，本季度账单收入大幅增长至5.3亿元，同比增长超200%，环比增长11%，占工业云收入比例达39%。继续领军的行业，我们在生态客户、AI大模型客户用量稳定提升的基础上，互联网客户业务场景AI应用，如在线教育、在线旅游等行业的算力需求也有所突破。在集群建设方面，我们针对重点客户的需求节奏高效协同、快速响应，打造了季度内完成大型集群全品类云服务交付的标杆案例。此外，我们也通过灵活的资金合作模式，确保底层算力供应支持AI业务的快速发展。

Now let me walk you through the key business highlights for the first quarter of 2025. In public cloud space, revenue reached RMB 1.35 billion this quarter, representing a year-over-year increase of 14%. AI business, as a key growth driver, recorded a significant increase in gross billing to RMB 525 million, up over 200% year-over-year and 11% quarter over quarter, accounting for 39% of public cloud revenue, continuing to lead the industry. Based on steady usage growth in ecosystem customers and foundation model customers, computing demands for AI applications in internet customer business scenarios, such as online education and online travel, also made breakthroughs. In the construction of clusters, we efficiently coordinated and responded quickly to the demand schedule of key customers, created benchmark case of delivering full-spectrum cloud services for large-scale clusters within the quarter.

In addition, through flexible capital cooperation models, we ensured sufficient underlying computing power supply to support the rapid growth of our AI business.

行业云方面，本季度实现收入6.2亿元，同比增长5%。本季度我们受到季节性验收节奏放缓的影响，行业云收入环比下降。分行业来看，公共服务领域推进AI在政务云和国资云的应用，积极拥抱AI成为趋势。金山云通过开源模型市场构建丰富的模型资源，同时匹配包含数据加工、模型精调、模型评估、模型量化等关键环节在内的一站式模型工具链，始终致力于为用户提供全流程一站式的AI服务，助力客户在实际业务场景中深度优化模型性能。数字健康领域，本季度开始武汉市检验检查结果互认共享平台建设，金山云在江苏、重庆等地沉淀的影像云能力再次得到验证，从影像场景扩展到检验检查场景，并在湖北市场复制拓展。

In enterprise cloud space, revenue reached RMB 616 million this quarter, representing a year-over-year increase of 5%. Affected by seasonal slowdowns in project delivery and acceptance process, enterprise cloud revenue declined quarter over quarter. By industry, in the public service sector, we are advancing the application of AI in public service cloud and state-owned asset cloud, actively embracing the AI-driven trend. Kingsoft Cloud has built a rich set of model resources through open-source model marketplace, while providing a one-stop model toolchain that covers key processes, including data processing, model fine-tuning, model evaluation, and model quantization. We remain committed to delivering full-process one-stop AI services to help customers deeply optimize model performance in their business scenarios. In the healthcare sector, we initiated the construction of a platform for mutual recognition and sharing of test and examination results in Wuhan.

Kingsoft Cloud's medical imaging cloud capabilities, which have been deployed in regions such as Jiangsu and Chongqing provinces, were once again validated. Our capabilities also expanded from imaging scenarios to test and examination scenarios, and have been further replicated and expanded to the entire Hubei province market.

产品技术方面，我们继续秉持技术立业，聚焦于打造核心产品的一流客户体验。本季度我们持续提升自算云的产品能力。金山云新流迅推平台作为一站式AI开发与部署平台，始终致力于为企业提供高效、弹性、低成本的模型训练与推理服务，包括小米模型在内的优质模型接入，将进一步扩展平台的生态能力，助力客户在自然语言处理、多模态交互、智能决策等场景中实现技术应用。

In terms of product and technology, we uphold the principle of building success based on technology and innovation, focusing on delivering best-in-class customer experiences across our core product offerings. This quarter, we continue to enhance the product capabilities of our intelligent cloud computing services. Our Xingliu training and inference platform, as a one-stop AI development and deployment platform, remains committed to providing enterprises with efficient, elastic, and cost-effective model training and inference services. The integration of high-quality models, including the one from Xiaomi, will further expand the platform's ecosystem capabilities and help customers apply AI technologies in scenarios such as natural language processing, multimodal interaction, and intelligent decision-making.

总体而言，我们的AI业务持续快速发展，云服务进入新的发展周期。在工业云领域，我们夯实基础设施、迅推平台、AI工具的能力建设，以帮助客户降低训练成本，提升在模型开发、调优、使用过程中的稳定、便捷、高效。行业云领域以AI数据、办公为抓手，在企业AI+场景的发展趋势中，提供一站式模型解决方案与服务。展望未来，我们将保持与小米金山生态深度合作，全面理解和探索新的AI机遇，为我们的客户、股东、员工和其他立方持续创造价值。接下来有请CFO Kerry为大家介绍一季度财务业绩，谢谢。

Overall, our AI business continued to grow rapidly, and our cloud services have entered a new development cycle. In the public cloud space, we strengthened our capabilities in infrastructure, training and inference platforms, and AI tools to help customers reduce training costs and improve stability, convenience, and efficiency throughout model development, fine-tuning, and usage. In the enterprise cloud space, with a focus on AI, data, and office productivity, we provide one-stop model solutions and services under the trend of enterprise AI+ scenario development. Looking ahead, we will maintain deep cooperation with the Xiaomi and Kingsoft ecosystem, fully understand and explore new AI opportunities, and continue to create value for our customers, shareholders, employees, and other stakeholders. I will now pass the call to our CFO, Henry, to go over our financials for the first quarter of 2025. Thank you.

Henry He (CFO)

Thank you, Mr. Zou, and thank you all for joining the call today. I will now walk you through the financial results for the first quarter of 2025. This quarter, our AI strategy continued to drive our growth and laid a foundation for future development. Total revenues for this quarter were RMB 1,970.0 million, reflecting an 11% year-over-year increase. Out of this, revenues from public cloud services were RMB 1,353.5 million, up 14% from RMB 1,187.4 million in the same quarter last year. This growth was mainly fueled by a surge in AI-related business, with billing reaching RMB 525 million. This quarter, our capital expenditure reached RMB 605 million. Revenues from enterprise cloud services reached RMB 616.5 million, up 5% from RMB 588.2 million in the same quarter last year, primarily driven by increased demand in industry solutions.

However, we have witnessed a 25% sequential decrease of enterprise cloud revenues, which was mainly due to the seasonality impact. Total cost of revenue was RMB 1,651.7 million, up 11% year-over-year, which was in line with our revenue expansion. IDC costs dropped by 6% year-over-year from RMB 768.5 million to RMB 722.8 million this quarter, reflecting our execution on cost control and better reps utilization. Depreciation and amortization costs increased from RMB 183.5 million in the same period of last year to RMB 378.5 million this quarter, mainly due to the depreciation of newly acquired high-performing servers to expand our AI business. Solution development and the service cost rose by 13.3% year-over-year from RMB 446.0 million to RMB 505.2 million, driven by expansion in Kaimelot personnel to support revenue growth. Fulfillment costs and other costs were RMB 3.1 million and RMB 42.1 million this quarter, respectively.

Our adjusted gross profit for the quarter was RMB 327.7 million, a 9.6% increase year-over-year, while a decrease of 23.4% quarter over quarter. Adjusted gross margin was 16.6% in this quarter, compared with 16.8% in the first quarter of 2024 and 19.2% in the fourth quarter last year. Our adjusted gross margin has been negatively impacted by the seasonality for enterprise cloud services and the higher upfront investments into servers and reps for AI business. On the expenses side, excluding share-based compensation, our total adjusted operating expenses were RMB 427.3 million, a decrease of 9% year-over-year and a 4.3% quarter over quarter, of which our adjusted R&D expenses were RMB 200.8 million, increased by 4% from the same quarter last year. Adjusted selling and marketing expenses were RMB 107.8 million, increased by 10.1% year-over-year.

Adjusted G&A expenses were RMB 118.7 million, decreased significantly by 13.6% year-over-year due to the decline of credit loss. Our adjusted operating loss was RMB 55.8 million, narrowed by 56% from RMB 127.0 million in the same period of last year. The improvement was mainly due to the increase of gross profit and a decrease of credit loss expenses. However, the adjusted operating profit turned loss from last quarter, which was mainly due to the decrease of gross profit in this quarter. Our non-GAAP EBITDA profit was RMB 318.5 million, increased by 8.6% from RMB 33.2 million in the same quarter of last year. Our non-GAAP EBITDA margin achieved 16.2%, compared with 1.9% in the same quarter last year. It was mainly due to our strong commitment to AI cloud computing development, strategic adjustment of business structure, and our strict control over cost and expenses.

As of March 31, 2025, our cash and cash equivalents total RMB 2,322.7 million, providing a strong liquidity position to support operations and AI investments. Looking ahead, our cloud infrastructure is ready to serve in a short time. The demand from ecosystems and other AI application scenarios not only fueled our business growth but also reinforced our confidence in this trajectory. With demand for AI cloud services continuing to grow, we are well-positioned to capture and capitalize on AI capital opportunities. Thank you. Thank you, Liz Kang, who is now preparing for your attention. We are now happy to take your questions. Please ask your question in both Mandarin, Chinese, and English if possible. Operator, please go ahead. Thank you.

Operator (participant)

Thank you. Thank you. As a reminder to ask a question, please press star 1 on your telephone and wait for your name to be announced. To withdraw your question, please press star 1 again. We will now take the first question from the line of Brian Gong from Citi. Please go ahead.

Brian Gong (Analyst)

谢谢观众，请问我有两个问题。第一个问题就是我们一季度看到公有云和这个行业云的这个增速好像都比之前想的要弱一点点，这个背后的原因是怎么样？然后我们现在该怎么看全年的这个收入情况？另外一个问题就是想问一下小米这边也发布了自己的这个模型，能否请管理层介绍一下这个最新小米对我们这边的需求，我们看到有怎么样的变化？谢谢。 I will translate myself. Thanks, management. So one is in the fourth quarter for both public cloud and enterprise cloud, the growth seems a little bit weaker than our previous expectation. What are the reasons behind this, and how should we see the full year growth right now? Secondly, recently Xiaomi released its own large language model, and management elaborate what's the latest demand from Xiaomi? Thank you.

Zou Tao (CEO)

我先回答，然后等一下看看Henry和刘涛的那个补充。你的第一个问题，看到Q1的公有云的增速和行业云，行业云我们其实在电话会上也谈过，其实它确实是受季节交付的影响，这还是有比较大的影响的，无论是我们本部的还是克莱特的。对，因为Q1总有个春节，而且很多行业客户也好，政企客户也好，在整个Q1的相对来讲的项目会少点，基本上是做预算的一个时候。公有云这块我先说，反正刘涛可以补充。因为本身我们主要是以大客户为主，基本上就是说本身建设有个周期，然后收费它基本上会递延到下个Q才逐渐能够看到。好比我们Q1，我们其实交付了小米的一个502集群，这基本上跟你第二个问题也有关联，但实际上我们基本是在Q1季度结束，即将基本上是在3月底才交付完。所以整个的收入它只能体现在二季度。所以我把这个也，我以前讲游戏，我经常讲我说楼梯型的曲线，就是因为我们以大客为主，就是基本上它建设可能要一个Q，所以在本Q反而是你看到投入居多，它的收入和利润的影响会反映在下个Q。所以你到下个Q你就看，你就能看到公有云这块就会比较快，好不好？所以我觉得这个这是关于你第一个问题。第二个问题就是关于小米模型这个对我们的影响，其实刚才已经谈到了，实际上它是在5月份发布的吧，在4月份发，它那个GB模型基本上就是我们交付之后在我们的集群上跑出来的。对，而且我们刚才也说了，就是我们小米今天生态体系Q1的收入增速相比去年来讲是有50%。其实我们还没有单独披露AI，其实AI这个是百分之一两百吧。对，所以总体来讲这个也反映出小米跟小米本身在AI方面的加速是数据是相匹配的，好吧？从长远看，想象空间就更广阔，无论是在模型的训练，还是随着小米的模型的逐渐成熟，应用到它的广泛的生态链产品当中去带来的巨大的推理的需求，我想未来这个还是有更大的一个想象空间。当然这一切我们都是因为我们毕竟是为小米服务，总体这个节奏还是跟随着小米本身AI的发展节奏而走的，好不好？你先翻译一下，等一下看有没有补充。

Thank you very much for your questions. In relation to the first question about the client's speed in terms of both public cloud and enterprise cloud, as we noted in the prepared remarks, especially for the enterprise cloud, the seasonality is quite evident. That includes impact for both our own enterprise cloud services as well as the business from Kaimelot. Because obviously Q1, there is the factor of the Chinese New Year, and a lot of customers for enterprise cloud business are still doing their budgeting. In terms of the public cloud, as we mentioned before, the public cloud business of Kingsoft Cloud focuses mainly on key customers, which typically are large customers, and they have their own cycle of the construction of the clusters before they can actually get online and generate revenue.

In fact, in the first quarter, we have delivered a 512-node cluster to our key customers, Xiaomi. However, that delivery time point was only at the end of March, and therefore the revenue and profit reflection, as in the financial numbers, will only be shown in the next quarter, which is the second quarter, Q2. Now, in relation to your question about the model from Xiaomi, in fact, the Xiaomi 7 billion parameter model was actually trained from our cluster, and in fact, the tremendous growth of the KC's AI business has much to do with the demand coming from Xiaomi. In the long-term perspective, the tremendous demand for model inference coming from Xiaomi still has significant potential.

However, since we're serving Xiaomi, the pace of Xiaomi's demand has a lot to do with when and the amount the revenue and profit will show on our financial statements on a quarterly basis. Last question, please. Operator, thank you.

Operator (participant)

Thank you. We will now take the next question from the line of Xiaodong Su Cloud CICC. Please go ahead.

Xiaodan Zhang (Analyst)

哎，周总，Henry，Clark，Nick，晚上好。感谢接受我的提问。那我第一个问题是我注意到最近这个金山云和金山办公室联合推出了政务领域的AI一体机，那想请这个管理层分享一下就是怎么看待政务AI领域的这个业务机遇。那能否也介绍一下这个AI一体机的定价模式，以及我们和这个办公的一个利益分配的模式？那第二个问题是关于我们的这个利润率，那我们看到这个non-GAAP OP margin从上个季度转正之后，那这个季度的话其实在这个季节性的影响之下环比也有一些回落。那管理层如何看待后续季度的这个利润率的趋势？快速翻译一下我的问题。 So thanks, management, for taking my questions. And first of all, I noticed that the Kingsoft Cloud and Kingsoft Office have recently jointly launched an AI all-in-one model machine for government affairs sector. So how do you view the business opportunities in the government affairs AI field? And also, could you introduce the pricing models of this all-in-one machine? And secondly, could you please update on the non-GAAP OP margin outlook for the subsequent quarters? Thank you.

Zou Tao (CEO)

好，谢谢小田。第一个问题，因为今天电话会的我们政企负责人瑞荣正好在现场，所以我就请他来回答，我等一下稍微做补充。第二个问题等一下由Henry来解答。好，瑞荣。 好，您好。就这样，我把这个关于刚才您提问政务一体机的情况简单来做一下解释。是这样，政务办公一体机主要还是因为现在在政企在政府这个领域里面，大模型发展推理场景发展速度非常的快。所以我们和我们的兄弟公司金山办公一起也在国内一些重要的地市也正在落地政务办公大模型这样的解决方案。这个方案也经过阶段性的市场一个验证，它也是一个未来发展一个很重要的需求。那一体机是怎么回事？主要是我们把政务的解决方案和AI的这个制算的硬件设备放在一起，而且它是小巧灵活，适合各个不同类型的客户在使用。所以在上个月我们和办公一起为了针对这个行业办公领域大模型发展的一个推理场景，那么发布的软件和推理的硬件结合在一起的一体机的产品。这个市场是面向于主要是面向于中国的政府各个委办局的客户在使用。所以这个是一体机的大概的情况。然后关于您刚才提到的问题，关于定价的情况，因为它取决于不同的硬件，所以我们的一体机的产品涵盖了从基础配置的产品，包括与一些国内的一些硬件产品的组合，所以这个价格是不同。所以也正在逐渐地在市场推广的一个过程之中。基本上我的回答就是这些问题。

Let me quickly translate. As Riley pointed out, the market for AI in the office automation targeting the government space has been developing very fast, and therefore that's the reason why Kingsoft Cloud and Kingsoft Office have come together to jointly develop and offer a plan or a solution to such market, which has been actually validated by the market. That integrated solution is actually an integration of both software and hardware, and notably in the software portion, there's the AI component that helps them to become more efficient in office scenarios. This is targeted for Chinese government agencies in various layers of various tiers. In terms of the pricing question you asked, it actually varies based on the different configuration of hardware that we are providing to the customer. The pricing can actually vary by quite a lot.

Henry He (CFO)

Yeah, thank you, Clark. Xiaodong, this is Henry. Yeah, thank you for the question on the margin side. I think probably a few things I want to point out. Given the seasonality in Q1, as probably many of you understand, primarily OpEx is the human capital expenses. The salaries, compensation, bonuses, and so on and so forth. In Q1, basically, we not only need to incur the normal salary payment, but also there's a bonus and also certain kind of employee benefits also incurred in Q1 during the holiday seasons, as you understand. Given the revenue side was affected by the delay of certain projects and also the top-line growth has been affected by seasonality. The variation of the top-line growth largely affected the bottom line. The cost structure, especially on the salary and the compensation, are relatively fixed.

That's why it's kind of affecting on the OP margin Q1. It's kind of very straight line the reason. Going forward, I think given the better margin profile, for example, the AI projects and our relationship with our ecosystem partners, especially for Xiaomi as our CEO, Mr. Zou, just mentioned, the big projects were likely to deploy on the way, and it will be largely booked starting from Q2 going forward. Hopefully, our top-line growth will drive a better margin expansion in the following quarters. We are in the view that our OP margin will be better in the following quarters. It depends on the top-line growth, the pacing of that, and the scale of the revenue quality in the following quarters.

On the other end, we also want to point out, given the EBITDA margin will be likely better compared with the OP margin given the nature of the AI business. As the AI business penetration becomes higher in the public cloud services, hopefully the EBITDA margin can be recovered a little bit faster than the OP margin side. I think I will encourage the audience to probably closely track our growth margin profile and EBITDA margin profile as well as the R&D cost, which are primarily linked to the expenses and the compensation and the salaries of employees. I think these are factors that are going to be the leading indicator for the OP margin going forward for the following quarters.

While we do not give a formal guidance, I understand Brian also mentioned this for the full-year top-line revenue in the call today, but we are in the view that the margin profile in the following quarters likely, especially in the second half of this year, will be better than the first half of this year. At this moment, we do not give a kind of formal management guidance for the top line, but the margin profile will continue to be better, especially in the second half of this year. Thank you.

Xiaodan Zhang (Analyst)

Thanks. Question please. Thank you.

Operator (participant)

Thank you. We will now take the next question from the line of Thomas Chong from Jefferies. Please go ahead.

Thomas Chong (Analyst)

晚上好，谢谢管理层接受我的提问。我的问题是关于我们全年的AI的CapEx跟那个费用的一个人事可以好像上一个季度能差一下这个breakdown。另外的话，我想多问一下就是关于芯片，就是一个chip bank的那个问题，对我们Q2的CapEx versus Q1会不会有些影响。然后这个问题也是关于一些行业的问题，就是关于大家开始用一些小模型，其实对我们云的服务会不会有些影响。 Thanks, management, for taking my question. My first question is about the AI CapEx and the OPEX breakdown. Can we have an update like last quarter? My second question is about our quarterly CapEx like Q2. Is the chip bank issue affect the sequential momentum in CapEx? My third question is more about the industry landscape. Given that customers are now using more like a distilled model, smaller models rather than large models, would that affect the cloud revenue? Thank you.

Henry He (CFO)

Thank you, Thomas. I will take the first question, and I defer to CEO, Mr. Zou, for the second and third question. 我先回答第一个问题。第二，关于芯片限制和小模型影响，请周总再做反馈。 The first question regarding the CapEx expenditures. I have mentioned in this quarter, our total CapEx was RMB 600.05 million for the first quarter of this year. If the audience remembers, since last year, we actually diversified the way we are financing the investment into AI, especially the infrastructure construction. Our own cash to be used for the CapEx is one part, but also we are going to, for example, do the financial leasing and operation leasing with our partners, for example, Xiaomi as well.

But also we are going to get the financing, for example, the bank loans and a certain leasing agreement with the state-owned banks in China locally, as well as some leasing companies in China as well, to arrange certain of the balance sheet financing arrangements. I think these are the three ways that we're financing the total AI infrastructure demand. Out of that, I think the reason you probably point out the total CapEx for this quarter is kind of RMB 600 million for this quarter, but I think our total investment into AI, because we also start to leasing certain servers from third parties to reduce our burden to pay out of pocket from own cash. There is a certain leasing and a rental agreement we actually start to negotiate and set up with third-party staff from Q1.

Hopefully for next quarter, we're going to give some updates regarding our total spending, including both our own cash out of pocket, but also the leasing agreement we'll start to arrange with third parties. This is going to be the total investment of the AI infrastructure from our definition. We're going to likely have some updates for next quarter because we just start to arrange those agreements with third parties, ensure to reduce and the saving and to be more efficient in terms of our own cash management. I think right now we give the number on total CapEx, but the leasing plus the CapEx and plus our own cash will be another update for next quarter.

Also, given the OpEx, as I mentioned in earning costs, our R&D expenses for this quarter is around CNY 200 million, and the sales expenses is around CNY 108 million, and the management SG&A is around CNY 119 million. If you are putting the three numbers together, it is roughly about CNY 400 million on the OpEx side, including R&D, sales marketing, and SG&A. I think this probably can give you a kind of a color regarding the total OpEx plus CapEx on a quarterly basis. Thank you.

Zou Tao (CEO)

好，那个可能有请CEO周涛总反馈一下关于芯片限制和小模型的问题。 第二个问题我来回答一下，不用反映了。第二个问题我来回答一下，就是确实关于H10的限制肯定是有影响的。具体的影响我想展开来说一下。第一，其实对于这种芯片的那个限制，它不是第一次，也不是第一天。所以我们尤其是对于这一波，我们之前还是有一定的准备，所以也做了适当的储备。所以从短期来看，尤其是我们主要是服务几个大客，从短期来看影响不大，但是从中长期不仅仅对我们对整个行业都还会有比较大的影响。这是第一个结论。第二个，其实从2023年开始，为了应对这种中美这种关系的起起落落，我们其实已经加强了跟国内的厂商的紧密的合作。我记得以前有个QA谈过，我说其实金山办公我们自身的模型，包括金山云自己训的模型都是基于国产算力去做的。所以围绕这种合作我们也一直在持续。所以从中长期看，必须要做好国产替代的这种准备。我们为此也在积极努力和准备。好吧，这是第二个。简单讲，短期影响不大，中长期可能会受比较大的影响，但我们已经做好了用国产算力逐步替代的这种方案和可行性的准备。

Okay. Okay, very quickly to translate from Mr. Zou's response to the chip bank question. Honestly and very frankly, there surely will be impact, right? However, this is actually not the first time for similar kind of restrictions to be imposed on chips. We have experienced similar things before.

Before this round of restriction actually hit us, we have had some inventory and storage. Therefore, the short-term impact is not that material. However, from a mid to long-term perspective, there will be meaningful impact for both Kingsoft Cloud as a company and for the industry as a whole as well. We mentioned before, since 2023, we have already strengthened our cooperation with Made in China Computing Resources. That kind of cooperation, we have managed to continue that cooperation since then. In light of similar restrictions, we have made and we have been well prepared for the gradual substitution of Made in China Computing Resources should restrictions continue to be more restrictive.

Henry He (CFO)

第三个问题，正好我们这个业务负责人刘涛，金山云的SVP在场，所以我请他回答一下。 对，先说结论。这个小模型，就是客户使用小模型的推理对我们的业务的增长和收入是没有影响的。原因是这样的，第一，我们的传统大模型的客户，他们的模型都是天衣量级的。所以他们基本上都是在训练和训推的组合上。所以这部分的业务其实没有什么影响。另外对于我们的互联网的客户来说，他们目前主力使用从一开始的模型就是偏小一点的模型，或者中型的，比如说像这种30多B的、10几B的这种模型。所以这个业务对我们来说始终都是一个纯增的量，所以它对我们来说也没有影响。对，然后包括我们未来的，包括小米生态的业务增长，其实可能对我们来说是一个正向的，因为随着小米的资源模型和WPS、GPT-6或等等其他这种开源和资源模型的使用的量的增长，它会有更多的推理量回到金山的生态内。因为可能它原来在调别人的大模型API，但现在这些量都回来变成自己的推理了。所以它对我们来说反而是个增长的刺激。就是这个结论。

The basic conclusion is that there is no negative impact for revenue and profitability for Kingsoft Cloud. The reasons being followed. The first one is for the traditional large language model customers, their models have been 100 billion parameter large language models, and they have historically been using our computing power for training and inferences, and this part is actually not impacted. Secondly, for the more traditional internet space customers, they have actually always been using relatively smaller models, including the 30B and 10B kind of parameter amounts for our business. This for us is actually an incremental amount for our revenue development.

Thirdly, in terms of the in the future for the Xiaomi and Kingsoft ecosystem, we also expect to have actually a positive impact coming from the more prevailing usage of relatively small models, because more of the model inference would originally that coming from outside of the Kingsoft Cloud computing power infrastructure from other companies. Actually, when the Xiaomi and Kingsoft ecosystem started to adopt more medium-sized or smaller-sized models, they were actually starting to use more of Kingsoft Cloud's computing resources. That is actually another part of potential incremental revenue and profit for us.

Nicole Shan (IR Director)

Thank you. I'll bring our next question.

Xiaodan Zhang (Analyst)

Thank you. We will now take the next question from the line of Wenting Yu from CLSA. Please go ahead.

Wenting Yu (Analyst)

quickly translate my questions. The first question is, can management share the recent gross margin trend for AI cloud leasing services? Previously, we observed the aggressive price competition in the industry, mainly focused on model API pricing. Are we seeing the pricing pressure extend to the AI server leasing business as well? The second question is, following the open sourcing of R1 by DeepSeek, how do we see the latest dynamics of model training demand? While there is incremental demand for post-training models, the open source release of R1 may also lead to some model vendors abandoning further iteration. How should we evaluate these two factors and their overall impact in the industry? Thank you.

Henry He (CFO)

好，我先回答第一个问题。第二个刘涛来回答。第一个问题是关于国内这种竞争加剧，AI这个毛利的趋势。我觉得从内外两个维度讲，外部的确实就是诚如你所说，就是整个市场随着AI玩家的集中化，大家确实存在着一定的竞争，这个有一定的影响，尤其是主要影响的是我们的一些新项目。因为过去旧的，这都是一年前、两年前的，这个长合约基本上影响不大，主要是新客过去存在着一定的竞争。所以这个在一定程度，从总体上来看，这个会有一定的影响。当然影响的程度，这个得看本身这些客户的体量了。如果体量小，这个分子小，分母大，这影响就小点。如果体量大，这个分子大，这个整个就会影响大一点。这是一个。第二个还有就是我们总体来讲，我们内部把它叫做资源池的这种供应方式，我们从去年下半年陆续开始，我们会尤其是除了头部核心客户，我们会有自主的KPS来支持服务。然后我们对于外部的一些非核心客户，我们也会采取一些外部资源池的这种方式。其实类似于融资租赁的方式。所以这个某种程度也会给我们的这个毛利有一定的影响。简单说，其实就好比是分润模式了。对，我们帮助去运营这些资源，有合作方提供这些资源的采买，然后我们两家一起分润的这种方式，也会对我们今年的这个AI总体的这个毛利会有一定的影响。好吧。

In relation to the AI cloud service margin or pricing pressure question, let me respond to you from two dimensions, which is internal and external. Externally, as you rightly pointed out, we're actually seeing sort of market concentration for AI players as our customers. Therefore, they do have some impact to our new projects, which we entered into more recently. The impact would largely depend on the project size. However, for the old long-term contracts that we have signed in the past and for those projects that we entered into in the past, there's no meaningful or no material impact. Another point I want to point out is that starting from the second half of 2024, we're increasingly leveraging the so-called resource pool supply chain to expand our infrastructure.

The strategy basically is for top KA customers, we use our own cash and dry powder and CapEx to build the infrastructure and to provide cloud service to them. However, for non-top customers, we're increasingly leveraging the partners who will supply us the servers and computing resources, and we jointly together provide cloud service to these customers. Because these are partners that provide resources, we actually also have to share some of the profit with them. This is like a partnership in profit sharing model. For that reason, that new or emerging supply chain model has also, to some extent, negatively impacted the gross margin level of the company.

Zou Tao (CEO)

第二个问题我来回答一下。就是从训练需求角度来看，确实我们可以看到，因为DeepSeek的出现，让国内的大模型的客户的训练的需求可能会有一定的减少。但是因为我们过往的这些合同都是长合约保护，所以目前在这方面我们看到的影响是非常少的。那么另一方面，DeepSeek的成功其实刺激到了国内的大型玩家。其实大家可以看到，包括小米在内，就是认为中国人用适量的资源就可以训出一个SOTA的模型。所以我们可以看到在国内的大型客户，包括互联网的大型客户，以及甚至一些新的行业里的大型客户开始有了自己去训练大模型的需求。所以从这个上面来讲，我们认为整体大模型的训练需求对于这个是比较平稳的。有一方面可能，比如说大模型的创业公司可能会有一定的收敛，但是对于这个新进的玩家可能认为，对吧，我们有对等的资源就可以训出一个自己控制的SOTA。所以我们认为整体需求是平稳，然后有略增。

In response to your question regarding the impact of open source of DeepSeek R1 to the demand of model training, my answer to you is this. This is coming from our SVP, Mr. Liu Tao. The old batch of large language model companies, the so-called independent large language model companies, we do have seen their demand shrinking to some extent due to the success of DeepSeek. However, most of the contracts that we entered with them are long-term contracts, which we have entered in the past. As we responded in the first question, the impact is very limited. However, there is also a group of other companies other than those large language model companies, some of them coming from the internet space, some from other emerging industries, they are inspired by the success of DeepSeek.

I also think that by using relatively manageable resources, they will also potentially be able to train an old state-of-the-art model. Therefore, we are seeing increasing demand coming from these customers as well. All in all, we do not see negative impact from the success of the DeepSeek model.

Wenting Yu (Analyst)

Okay.