Q2 2026 Kingsoft Cloud Holdings Ltd Earnings Call

August 19, 2026

Speaker #2: Good morning, ladies and gentlemen, and thank you for standing by for Kingsoft Cloud's second quarter 2026 earnings conference call. All participants are currently in listen-only mode.

Speaker #2: Following management's prepared remarks, we will open the call for questions. Please note that today's call is being recorded. I will now turn the call over to Mr. Jackie Zor, Senior Director of Capital Markets at Kingsoft Cloud.

Speaker #2: Jackie, please go ahead.

Speaker #3: Thank you, operator. Hello, everyone, and thank you for joining us today. Kingsoft Cloud's second quarter 2026 earnings release was issued earlier today and is available on our IR website and through PR Newswire.

Speaker #3: Joining us today are Ms. Zou Tao, Chairman and CEO; Ms. Li Yi, CFO; Mr. Liu Tao, Senior Vice President; Mr. Tian Kaiyan, Senior Vice President; Ms. Yu June, Vice President; Mr. Zhou Ruilong, Associate Vice President; and Mr. Kwak Chen, Board Secretary and Associate Vice President.

Speaker #3: Mr. Zou will discuss our business performance and key developments, followed by Ms. Li with a review of our financial results. Management will then take your questions.

Speaker #3: Consecutive interpretation will be provided for your convenience, and for reference only. In the event of any discrepancy, the speakers' statements in the original language will prevail.

Speaker #3: Before we begin, I would like to remind you that today's call contains forward-looking statements made under the Safe Harbor provisions of the U.S. Private Securities Litigation Reform Act of 1995.

Speaker #3: These statements involve risks and uncertainties, and actual results may differ materially from those expressed or implied by the forward-looking statements. Additional information concerning factors that could cause actual results to differ materially is included in the company's filings with the U.S.

Speaker #3: The company undertakes no obligation to update any forward-looking statements, except as required by applicable law. Unless otherwise stated, all financial figures discussed on today's call are denominated in renminbi.

Speaker #3: With that, it is my pleasure to turn the call over to our Chairman and CEO, Mr. Zou. Mr. Zou, please go ahead.

Speaker #4: 大家好，欢迎参加今天与2026年第二季度业绩电话会。我是今天与CEO周涛。本季度我们看到了AI云业务形态的进一步演进，开源大模型生态的空前繁荣，给综艺云厂商带来巨大的发展空间。而我们几年来一直坚信的AI落地天花板页正以模型即服务、智能体即服务和FDE相结合的形式向我们走来。面对这一深刻的行业变革，今天云继续坚持技术立业与高质量可持续发展战略，全面推动AI云服务MaaS业务FDE业务的发展，取得了可喜的成绩。首先，AI业务继续驱动公司收入高速增长。本季度实现总收入30.7亿元，创单季度历史新高，同比增长31%。其中AI云账单收入达13.3亿元，同比增长82%，占公用云收入比例进一步提升至56%。MaaS业务收入增长强劲，二季度MaaS收入较一季度增长12倍。其次，盈利水平显著提升，本季度经调整毛利率提升至15.4%，环比上升2.4个百分点。经营利润首次转正，经调整经营利润率达4%，创历史新高。这是我们抓住AI浪潮、提升收入质量、落实降本增效等多项举措并行所取得的可喜成果。第三，客户结构持续优化，生态内外商机加速兑现，生态内本季度来自小米和金山生态的收入达8.1亿元，同比增长28%，占总收入比例为26%。生态外前五大客户收入同比增长51%。AI云业务已覆盖互联网前沿AI实验室，具身智能、自动驾驶、AI for Science、金融科技、游戏、音视频等广泛行业，多元化的客户结构与业务布局，不仅带来收入规模的持续提升，也使我们能够更灵活地调配算力资源，提升议价能力和抗风险能力。

Speaker #5: Hello, everyone, and welcome to Kingsoft Cloud's second quarter 2026 earnings call. I am Zou Tao, CEO of Kingsoft Cloud. This quarter, we saw further evolution in the AI cloud market.

Speaker #5: The rapid growth of the open-source model ecosystem is creating significant opportunities for neutral cloud providers. At the same time, our long-held vision of bringing AI to every industry is becoming a reality through a combination of model as a service agent as a service and FDE services.

Speaker #5: Against this backdrop, Kingsoft Cloud remains committed to technology leadership and high-quality, sustainable growth. We are accelerating the development of our AI cloud, MaaS, and FDE businesses, with encouraging progress.

Speaker #5: First, AI continues to drive strong revenue growth. Total revenue reached a record of RMB 3.07 billion, up 31% year over year. AI cloud gross billings increased 82% to RMB 1.33 billion.

Speaker #5: It accounted for 56% of public cloud revenue. MaaS revenue also grew strongly, with Q2 revenue up more than 12 times from the Q1 level.

Speaker #5: Second, profitability improved significantly. Adjusted gross margin rose to 15.4%, up 2.4 percentage points quarter over quarter. Operating profit turned positive for the first time, with adjusted operating margin reaching a record high of 4.0%.

Speaker #5: This reflects our continued efforts to capture AI opportunities, improve revenue quality, and drive greater operating efficiency. Third, our customer mix continued to improve. With stronger momentum both within and outside our ecosystem, revenue from the Xiaomian Kingsoft ecosystem reached RMB 810 million, up 28% year over year.

Speaker #5: It accounted for 26% of total revenue. Revenue from our top five non-ecosystem customers grew 51%. Our AI cloud business now serves a broad range of sectors, including internet services, frontier AI labs, embodied AI, autonomous driving, AI for science, fintech, gaming, and online video, to name a few.

Speaker #5: This diversified customer base supports continued growth, while allowing us to allocate computing resources more flexibly and strengthen our pricing power and business resilience.

Speaker #4: 下面我向大家具体介绍2026年第二季度的业务进展。公用云方面，本季度实现收入23.6亿元，同比大幅增长45%。首先，小米AI全面赋能人车家全生态，金山打PC AI业务推进，作为小米金山生态的唯一战略云平台，AI时代所带来的云服务业务增长潜力空前膨胀。2026年6月，股东大会正式批准了我们再次提升来自小米关联收入的上限金额。2026年和2027年来自小米关联收入上限合计达100亿元，较上调前提升了39%。今年上半年来，自小米和金山的公用云收入同比增速达54%。其次，新流平台MaaS服务能力进一步强化，新流模型API服务持续完善，多模型服务与企业级接入能力。目前，新流平台已部署上线120款模型，主流新模型推出后做到同步上线，以接入230多家客户。第三，我们在新兴赛道深化与头部客户合作，面向头部具身智能客户和头部自动驾驶客户，完成大规模算力集群交付，稳定支撑其模型高效迭代与AI for Science行业头部客户达成深度合作，保障其新业务快速拓展。

Speaker #5: Now let me walk you through our business progress in the second quarter of 2026. In public cloud, revenue reached RMB 2.36 billion, up 45% year over year.

Speaker #5: First, Xiaomi continues to expand AI across its human-car-home ecosystem, while WPS AI continues to advance. As the only strategic cloud platform for the Xiaomi and Kingsoft ecosystem, we see substantial AI-driven growth opportunities.

Speaker #5: In June, our shareholders approved a further increase in the annual caps for connected transactions with Xiaomi. The combined caps for 2026 and 2027 now total RMB 10 billion.

Speaker #5: 39% higher than before the adjustment. In the first half, public cloud revenue from Xiaomi and Kingsoft grew 54% year over year. Second, we saw this strengthen the MaaS capabilities of our Staff Flow platform.

Speaker #5: Staff Flow now supports 120 models, with major new models launched on the platform in sync with their market release, and serves more than 230 enterprise customers.

Speaker #5: Third, we deepened cooperation with leading customers in emerging sectors. We delivered large-scale computing clusters to leading embodied AI and autonomous driving customers, supporting rapid model iteration, and expanded our cooperation with the leading AI-for-science customer to support the growth of its new business.

Speaker #4: 还有一个方面，本季度实现收入7.1亿元。公共服务领域，我们与长江、南京通信管理局签约打造降海云，共同打造适配长江航运需求的专属数字化底座。我们还与武汉市数据局、武汉云达成战略合作，在算力并网、数字政府、智能算力应用产业生态打造等领域展开合作。夯实武汉数字政府与产业生态底座。数字健康领域，我们牵头国家重点研发计划生物与信息融合专项，打造云端一体化医学虚拟手术平台，成果已在全国30多家医院应用，树立了云厂商牵头医工信交叉国家级科研公关的行业标杆。企业服务领域，我们与云上甘肃达成深度合作，以投建运一体化模式共同建设并运营甘肃省级政务平台。

Speaker #5: In enterprise cloud, revenue reached RMB 710 million. In public services, we signed an agreement with the Nanjing Communications Administration of the Yangtze River to build Jianghai Cloud, a dedicated digital infrastructure platform for Yangtze River shipping.

Speaker #5: We also formed a strategic partnership with the Wuhan Municipal Data Bureau and Wuhan Cloud, focusing on computing resources interconnection, digital governance, intelligent computing applications, and ecosystem development.

Speaker #5: In digital health, we are leading a project under the national key R&D program on biology and information integration to develop a cloud-based virtual surgery platform, which has been deployed in more than 30 hospitals nationwide.

Speaker #5: In enterprise services, we deepened our cooperation with Yunshan Gansu to jointly build and operate the Gansu Provincial Public Services Cloud under an integrated investment, construction, and operations model.

Speaker #4: 在产品技术方面，我们紧密围绕试算落地与AI应用的需求，升级全栈AI能力。本季度，新流MaaS平台围绕高并发推理需求持续进行模型部署优化，多个核心模型吞吐能力显著提升，并支持按权限、用量与模型维度进行精细化管理。我们正式发布新源AgentKit平台，提供涵盖安全沙箱、知识与记忆及评测治理的全套底座，帮助企业快速搭建生产级AI Agent应用。同时，我们将云产品周边AI化，将数据库存储等组件变成为Agent可便捷调用的平台型产品。我们优化了新流迅推平台，提升资源调度灵活性，围绕训练与微调场景，新增了物理队列、资源借用与回收，以及资源算力的灵活配置能力，极大地提升了多任务并发下的资源利用率，降低了研发团队的训练环境门槛与运营成本。面向私有化与国产化试算需求，银河平台完成了多款主流国产算力芯片的深度适配及全生命周期的可视化管理。

Speaker #5: In products and technology, we continued to upgrade our full-stack AI capabilities for intelligent computing and AI application deployment. This quarter, we further optimized the model deployment on Staff Flow MaaS for high-concurrency inference, significantly improving throughput for several core models.

Speaker #5: And enabling more granular access, usage, and model-level management. We also launched AgentKit, providing secure sandbox knowledge and memory management, as well as evaluation and governance tools, to help enterprises build production-grade AI agents.

Speaker #5: At the same time, we are making general-purpose cloud products, such as database and storage, easier for agents to access and use. We enhanced the staff flow training and inference platform with more flexible resource scheduling, sharing, and allocation for training and fine-tuning workloads, improving utilization and reducing development and operating costs.

Speaker #5: For private deployment of the MaaSic AI infrastructure, our Galaxy Stack platform has completed deep integration and full lifecycle visual management for multiple mainstream domestic AI chips.

Speaker #4: 台湾未来，我们将继续深耕生态内外巨大的商业机遇，不断优化算力资产运营效率。在技术演进与应用落地的浪潮中，持续增强自身的盈利与造血能力，以坚实的业绩表现为客户、股东和社会创造长期可持续的价值。接下来有请公司CFO李毅为大家介绍二季度财务业绩。谢谢

Speaker #5: Looking ahead, we will continue to capture opportunities both within and outside our ecosystem, improve the operating efficiency of our computing assets, and strengthen our profitability and cash generation capabilities, amid AI industry tailwinds.

Speaker #5: We remain committed to creating long-term, sustainable value for customers, shareholders, and society. With that, I will hand the call over to our CFO, Li Yi, who will review our second quarter financial results.

Speaker #5: Thank you.

Speaker #3: Thank you, Mr. Zhou and Mr. Tian. And thank you all for joining the call today. I will now discuss the second quarter financial results, using USD as the currency.

Speaker #3: Before we walk through the details of the financial results for the second quarter, I would like to highlight the following aspects. First, our quarterly revenue reached over $3 billion for the first time in our company history.

Speaker #3: Our year-over-year for the last consecutive quarter in particular, our AI cloud gross billing increased 82% year-over-year to $1.33 billion, accounting for over 43% of our total revenue versus 31% a year ago.

Speaker #3: This reflects a continued structural shift in our business mix towards AI. Second, our profitability has improved. Our adjusted gross margin was 15.4%, up 2.4 percentage points quarter over quarter and 0.5 percentage points year over year.

Speaker #3: Our adjusted EBITDA margin reached 36%, up from 17% in the same quarter last year and 82% last quarter. Notably, we returned to break-even at the operating income level this quarter and recorded an adjusted operating profit margin of 4%.

Speaker #3: This outcome validates our ability to turn strong AI business demand into healthy profit growth. Third, we continue to invest to accelerate the build-out of our AI compute capacity.

Speaker #3: Capital expenditures, together with right-of-use assets obtained through third-party financing and finance leases, reached $3.3 billion this quarter, compared to $2.9 billion last quarter and $2.8 billion in the same quarter last year.

Speaker #3: Now, let me walk you through our financial results for the second quarter of 2026. This quarter, total revenue was $3,072 million, up 31% year-over-year, or 40% quarter-over-quarter. Of this, revenues from public cloud services were $2,358 million, up 45% from $1,625 million in the same quarter last year.

Speaker #3: Revenues from enterprise cloud services reached $740 million, compared with $724 million in the same quarter last year, down slightly by 1% year-over-year. Total cost of revenues was $2,606 million, representing a 30% year-over-year increase.

Speaker #3: Mainly due to continued investment in AI cloud infrastructure, IDC costs increased by 23% year-over-year, from $803 million to $990 million this quarter. The increase was mainly due to the growth in rack services.

Speaker #3: Depreciation and amortization cost increased by 75% year-over-year, from $552 million in the same quarter of 2025 to $964 million this quarter. This was largely due to the depreciation of newly acquired and raised AI infrastructure.

Speaker #3: Including servers and network equipment. Solution development and services cost increased by 4% year-over-year, from $564 million in the same quarter of 2025 to $586 million this quarter.

Speaker #3: The mortgage increase was mainly due to higher costs incurred in AI transformation in solution development and delivery. Forming costs and other costs were approximately $66 million in total this quarter, compared to $92 million in the same quarter last year.

Speaker #3: Our adjusted gross profit for the quarter was $472 million, an increase of 35% year-over-year and 34% quarter-over-quarter. Adjusted gross margin was 15.4%, up from 14.9% in the same quarter last year and from 13% last quarter.

Speaker #3: The increase was driven by higher gross margin in the public cloud business, thanks to strong AI demand tailwinds. On the expense side, excluding share-based compensation expenses, our total adjusted operating expense was $391 million, a decrease from $561 million in the same quarter last year and from $455 million last quarter, mainly reflecting our disciplined cost and expense control.

Speaker #3: Of which, our adjusted research and development expenses were $184 million, up 101% year-over-year. Adjusted selling and marketing expenses were $102 million, down 7% year-over-year. Adjusted general and administrative expenses were $105 million, down 61% year-over-year, largely due to lower credit loss expenses.

Speaker #3: Our adjusted operating profit was $124 million, turning a profit from an adjusted operating loss of $166 million in the same period last year. This improvement was primarily driven by the expansion of our revenue scale, higher gross margin, and enhanced operating efficiency.

Speaker #3: Adjusted operating profit margin was 4% this quarter, compared with minus 7.1% in the same period last year and minus 2.2% last quarter. Our adjusted net loss was $6 million, down from $300 million in the same quarter last year and $237 million in the previous quarter.

Speaker #3: Our long gap EBITDA profit was RMB 1,100 million, an increase of 171% from RMB 406 million in the same quarter last year. Our long gap EBITDA margin achieved 36%, compared with 70% in the same quarter last year and 82% last quarter.

Speaker #3: It was mainly due to our improving gross profit, as well as higher depreciation costs in our cost base, as we accelerate our AI computing capacity build-out.

Speaker #3: As of June 30, 2026, our cash and cash equivalents totaled $4,674 million, compared with $4,904 million as of March 31, 2026. The decrease was mainly due to our continued investment in AI infrastructure to support business growth.

Speaker #3: Looking ahead, we aim to capitalize on the explosive growth in AI demand by further investing in infrastructure, expanding our product and service offerings, managing credit and liquidity risk, and improving operating efficiency.

Speaker #3: We remain committed to our all-in AI strategy and continue to deliver high-quality growth to our shareholders. Thank you, all.

Speaker #1: This concludes our prepared remarks. We will now begin the Q&A session. If possible, please ask your questions in both Mandarin and English. Operator, please proceed.

Speaker #2: Thank you. We will now begin the question and answer session. If you wish to ask a question, you will need to press star 1 1 on your telephone and wait for your name to be announced.

Speaker #2: To withdraw your question, please press star 11 again. We will take our first question. Your first question comes from Lipping Zhao from CICC. Please go ahead, your line is open.

Speaker #4: First of all, Lizzo. 晚上好。感谢接受我的提问。首先恭喜公司这个季度获得了OP层面的一个转正。那我这边有两个问题，其实都是关于我们星流那个MAS平台的。最近就是关于开源模型能力提升的讨论也挺多的。那请问这个对公司的MAS业务的影响，那么基于对咱们自己平台的这个观察，目前用量增长的趋势是怎么样的？然后主要来自于哪些这个典型的用例？然后第二个问题是考虑到整体星流这个平台它的投资回收周期可能更短，不知道公司是否会像这块业务去倾斜更多的这个资源。 Let me translate by myself. So good evening. Ms. Zhao and Ms. Li, thanks for taking my questions. I got two questions on your MAS business.

Speaker #4: First, how will improvements in open-source model capabilities affect the company's MAS business? Based on your observations, what's the current usage growth trend, and which use cases are driving it most?

Speaker #4: And second, given that the payback period for the MAS business might be shorter, will the company allocate more resources to it? Thank you.

Speaker #1: 好，这个问题我来回答一下。首先来看就是模型能力提升带来的几个变化。一方面是说国内的vibe coding需求的这个整个市场盘子是非常大的，需求量非常多。那么传统意义上来看，我们看到大量的客户是在使用可能会使用海外的模型。随着像GRM K3这些模型的能力提升，我们可以看到在vibe coding领域存在这种国产模型对海外模型的替代效应。所以这是一个对国产模型的需求的增长。另一方面，agentic的这个渗透性在提高，越来越多的场景在使用agentic了，包括我们很多客户的agent开始上生产了，所以它引出了我们的agent key的产品。另一方面，在agent的使用上，我们可以看到两种分化的需求。一方面是在做复杂问题解析的时候，可能会倾向于用K3，包括GRM 5.3这样的新型模型。但是当它需要去完成日常任务的时候，我们可以看到客户会倾向于选择最低性价比的模型。就最近大家可能看到DeepSeek的斩杀线。所以像DeepSeek V Flash，包括像Mimo Flash这些模型，在日常工作场景中的调用量我们看到是有增长的。所以这是第一个问题，就是我们的这个增长趋势来源的主要是哪些东西。另一方面来说，MAS业务和算力业务来说，实际上两者是各自有特性的。对于算力业务来说，当我们把一台机器售卖出去的时候，它的售卖率就是100%，并且通常我们安全起见都是签长合约。所以这个业务的安全性是相对很高的，并且它的利润水平是相对高并且固定的。那么MAS业务它涉及到了这个市场的价格竞争的波动，然后新模型的替代，包括这个运营水平以及这个多方面因素的影响。所以它的利润率是一个变动的数值。那么所以说这两者其实是一个对我们来说希望是一个平衡性的选择。一方面我们希望两者都发展，另一方面我们也会在两者这个业务之间互相作为备份。比如当我们的算力遇到了特殊的情况的时候，我们可以拿算力业务去做推理，反过来说，当我们的推理掉量的时候，我们也可以把推理的业务去转向算力业务。所以整体来说是一个平衡发展的策略。谢谢。

Speaker #4: Okay. So just to quickly translate some. This answer comes from our SVP, Mr. Liu Tao. In relation to your first question, the development in open-source large language models has mainly three impacts.

Speaker #4: Number one is that we're seeing very big demand coming from Vibe coding. And traditionally, foreign models have been taking the lead in this area.

Speaker #4: However, once we have seen the launch of GRM and K3, these kinds of high-performance models, we're seeing increasingly more users from mainland China adopting and using this made-in-China large language model.

Speaker #4: And secondly, the increasing use of agentic scenarios also brought change to our business. With that, we have launched, as mentioned in the prepared remarks, the Agent Kit product to satisfy such a need.

Speaker #4: And thirdly, it is worth mentioning that, in terms of day-to-day routine tasks and workloads, the choice usually is the price-for-value kind of models, which are essentially the Chinese models.

Speaker #4: So that is why this development—open source large language model—is actually beneficial for our business. And your second question regarding the balance between MAS business and the computing power: We basically have different business models for these two businesses.

Speaker #4: For the computing power business, essentially, once we sell the computing power, the utilization is by nature 100%, and we usually have long-term contracts to secure the utilization throughout a prolonged period of time.

Speaker #4: And therefore, it's relatively safe, so to speak. But for the MAS business, it is subject to quite a few factors, including the fluctuation of the token price, the launching of new models—which the customers might prefer to use—and also the operating efficiency that we're able to achieve.

Speaker #4: In doing a MAS business. So therefore, we generally balance these two business models and hope to have each one of them complement the other.

Speaker #4: So we generally dynamically evaluate these two businesses and try to decide how many resources to allocate. Thank you. Thank you, Tao Zong. That's very helpful.

Speaker #5: Okay. We will take our next question. Next question: Confirm, when can VU from CISA, please go ahead. Your line is open.

Speaker #6: 周总、李总，各位管理层，晚上好。感谢给我这个提问的机会，也恭喜公司非常强劲的业绩。我这边有两个问题想请教一下。第一个问题是：从6月到目前的这个月份，我们看到芯片的采购进度是怎么样的？我们对于全年的CAPEX（资本开支）最新的预期是怎样？第二个问题是关于企业云这块。我们看到这一块收入在过去两个季度有一些降速，那我们应该如何预期今年全年企业云，也就是行业云这部分的增速？另外，公司对于这块业务的AI转型，以及中长期的定位是怎样？我翻译一下我的问题。The first question is, since June, how has the chip procurement progressed in recent months, and what's your latest full-year CAPEX guidance?

Speaker #6: And the second question is about the enterprise cloud. This segment of revenue has decelerated in the past two quarters. How should we think about full-year enterprise cloud growth, and what's the AI transformation and medium-term positioning for this segment?

Speaker #6: Thank you.

Speaker #1: 那个第一个问题我来回答一下。就是供应链的问题，其实我们每次做业绩发布的时候，大家都非常的关心。我相信大家其实也都观察到，就在三年以前这个AIGC这一波开始发展的时候，其实供应紧张就一直是伴随着业务在开展的。而且这种供应紧张，我们看起来它不会是一个短期的，应该是一个还会维持比较长的一个时间。所以我们应该把它看作一个新的一个常态，就是低点。第二点就是，那么我们这么几年实际发展下来，大家也能观察到整个中国的AI的产业以及制算云的发展，其实也一直是在以非常快的速度在推进。就包括我们的业绩其实也反映了这一点。那么从我们的角度来讲，最重要的一个策略，其实我们就是不断地去增加我们的合作伙伴，还有供应商的数量。同时的话，在技术上也加快我们适配的速度，用这种这样的方式来增加我们供应链的多元化。大家应该也发现，现在国产芯片这个公司也上市了，从他们的性能、产能到适配的优化，其实都在快速的提升，特别在推理领域还是得到了广泛的一些应用和认可。那么第三点的话，就是因为供应市场它其实存在一定的这个节奏性，所以它还存在这个一次性采购可能金额比较大，然后批量这个节奏可能不太持续的一个状态。所以如果我们单从每个月的情况去看，它并不是一个线性发展的。从采购的金额的波动，它会比较大，但是我们放到全年的这个规模上去看的话，它应该是离我们的这个预期其实是比较接近的。具体的这个CAPEX的数字的话，可以请我们的CFO这块来做个回答。

Speaker #4: So allow me to quickly translate. The answer comes from our SVP, Mr. Tiankai Yan. There are three points. Number one: Actually, since 2023, it's been three years, and the market has always been hearing voices about the limited supply.

Speaker #4: So, I would say this is actually a new norm. Such supply difficulty is actually a long-term kind of situation. But secondly, we should also be aware of the fact that despite those constraints—financial constraints—the Chinese cloud computing or AI industry development has not been restricted, or largely restricted, by that.

Speaker #4: And the way that we actually tackle such situations is that we try to increase the number of business partners that we work with.

Speaker #4: We try to increase the number of suppliers we work with, and we are also working on increasing the compatibility of Made in China chips.

Speaker #4: You are all very well very much aware of the recently many of the Made in China chips are becoming public. And they are particularly good in use cases, such as modeling model inference.

Speaker #4: Now, number three, I would like to say that when you look at the CAPEX number on a month-to-month basis, it is usually quite volatile.

Speaker #4: And I have to say that the purchasing number, because it's usually a large chunk of money in a relatively small number of purchases, so the purchasing number, if you look at it on a monthly basis, is actually not a linear number.

Speaker #4: So I would say that for our whole-year CAPEX estimate, it should still be in line with what we have been expecting, and our CFO, Li Yi, should be able to give you more details in that regard.

Speaker #6: Hi, Tingting. CAPEX venture includes Capital Bank assets through listed arrangements, reaching $6.2 billion in the first half of 2026, accounting for over 75% of all four-year CAPEX last year.

Speaker #6: Whereas July status cannot fully represent the third quarters of all trade, it clearly shows tangible growth acceleration. Accordingly, we maintain our four-year CAPEX base case unchanged at $15 billion.

Speaker #6: Thank you, Tingting. Thanks, Tingtong and Yizong.

Speaker #5: 就是。

Speaker #2: Thank you. We will take our next question.

Speaker #1: Oh, sorry. We need to continue with another question.

Speaker #2: Apologies.

Speaker #5: Okay. 好，关于这个行业云过去两个季度收入增速有所放缓的问题，确实我们也不回避。但是我们也觉得不能简单地用两个季度的收入的增速来线性地外推全年的这个经营情况。那这一轮整个增速变化，我们觉得包含三类因素。第一类的话是众所周知的，从去年下半年开始，整个内存和存储价格的这个上涨。那这个上涨会导致政府和企业客户，他在做预算的时候需要反复调整。那这样的一个反复调整的话，导致我们的这个合同签约周期和预算周期会比以往要长。这是一个非常这个非常非常明显的这个上半年的市场的一个波动。那另外一个就是第二类，就是政府和国央企业，他本身就存在的这种签约交付验收，他的收入确认的一些这个季度性的波动。因为大家也知道，尤其是这个大型项目，下半年就是正企客户一般来说下半年营收确认相对会比较集中。所以的话，我们判断全年的收入可能不能只看这个上半年的这个同比。那从我们来说，更看就是以前的未确认的合同，还有项目的这个验收节点，还有一些这个高确定性商机的一个覆盖程度。那么第三类因素，就是由于这一波的这个AI的话，其实金三银行业云我们也主动进行这个收入结构的调整。过去行业云它主要依靠这个云基础设施的建设，系统集成和项目交付，收入规模比较大，但是它的项目制的这个特征比较明显。然后也会出现这个季度性的这个波动。那么今年随着这个，特别是小龙虾这一波在中国市场那个火了之后，今年我们主动去减少这种低毛利、低费用回款质量较差的项目，然后把这个核心资源能够集中到就是AI能够帮着客户进入核心业务流程，然后形成这个AI的自有产品能力和长期的运营收入的这个项目上。所以的话，就是短期收入的增速和合同质量，还有合同期限，以及未来的可持续收入之间，它可能因为这样的一个这个我刚说的这三类因素，它出现一些这个阶段性的背离。但这并不是说我们这个放弃增长，而是说我们实际上在重构这个增长的基础。好。

Speaker #1: Okay, so this answer comes from our VP, Ms. Yujun. Generally, I don't think, although we're seeing relatively slow growth in the enterprise cloud segment, that it is the right way to look at it from a linear extrapolation perspective.

Speaker #1: I will give you three reasons. I think number one, just to explain why we're speaking of looking at relative weakness in this regard, is that the upstream supply pricing hike, which changed quite significantly in recent quarters, has affected our prospective customers, essentially the SOE companies and also the government agencies. They have had to frequently adjust their budgeting quota, which delayed their decision-making process.

Speaker #1: So that's number one. And number two, you're all quite aware that the seasonality in the enterprise cloud business is quite strong. Usually, the delivery and the revenue recognition are concentrated in the second half of every year.

Speaker #1: So we actually have quite a strong pipeline to deliver in the second half of the year. And thirdly, this is a result of a proactive adjustment of our business structure.

Speaker #1: Namely, proactively shifting from the project-based business model to the operating-based business model, where the operating business model, from a financial reporting perspective, is automatically classified into public cloud.

Speaker #1: So, this is not simply as it was seized, as a weakening of the enterprise cloud business. Those are the three points I would like to offer.

Speaker #1: Thank you.

Speaker #6: That's helpful. Thank you.

Speaker #2: Thank you. We will take our next question. Your question comes from Timothy Zhao from Goldman Sachs. Please go ahead, your line is open.

Speaker #3: 好的，周总，您晚上好。感谢您接受我的提问。我这边有两条问题想请教。第一个问题还是回到我们的星流Mass业务，想请问一下金山云平台相对于市场上的其他竞争对手，我们的定位、应用场景或者竞争优势具体有哪些？关于Mass服务的收入确认和利润率的情况，能不能再详细分享一下？这是第一条问题。第二个问题是关于AI云服务的价格变化。能否请您分享一下过去几个月AI云服务价格变化的趋势，以及行业整体的价格情况？另外我们也宣布了一些提价或者减少折扣的动作，现在客户反馈如何？能不能量化一下价格因素对AI云收入增速的影响？我很快翻一下英文：Thank you, my friend, for taking my question. My first question is regarding the Mass. Just wondering, compared to the peers in the market, how do you think about the Kingsoft Cloud competitor advantage in mass services in terms of application scenarios, etc.?

Speaker #3: And could you share more about the revenue recognition and the profitability profile of the Mass service business? And my second question is regarding the overall pricing trend in the AI cloud business.

Speaker #3: Just wondering if you can share what the latest trend is over the past couple of months, and what have you heard from customers after you announced certain price hikes or discount reductions over the past few months?

Speaker #3: And whether you are able to quantify the impact from the price hike to your overall AI cloud revenue growth? Thank you.

Speaker #1: 那么Timothy，我来回答一下这两个问题。第一个是说关于我们这个星流Mass平台的定位优势问题。首先，这里面定位，我们在市场上算是一种比较独特的定位。就是第一，我们并没有自研的大模型。第二，就是我们又是一家云厂商。我想咱们在市场上看到的做这个Mass的场景有几种客户。一种是这个做大模型的云厂商，比如说像阿里等等，包括百度。另一种是最近可能大家看到像什么硅基流动之类的这个所谓的纯的云端Token工厂。我想我们可能和他们都互相有一些区别。首先，我们没有自研的大模型。那么在售卖模型的政策指导上来说，我们始终指导着我们的销售和业务团队去销售最好的模型，客户最喜欢的模型。比如说GRM 5.3，比如说Kimi K3，包括可能在视频生成领域的像Cdance或者是包括Mini Max H3。所以这一方面我们的，我们会根据客户的需求来去售卖客户更喜欢的模型，而不会捆绑在说我们必须要卖某一个模型的政治任务上。我想这个在竞争策略上来说是一种灵活性。第二，对于这个云端Token工厂来说，其实Token的生产有个非常重要的生产要素是算力。那么掌握低成本算力的公司才可以在整个Mass的生态链上赚到钱。如果是纯粹做软件的这个生产去租赁算力的话，那么可能大量的这个利润会被这个算力的供给方占有。这一点我想大家可以关注一下。像海外的NCP的一张卡的卡池，然后再看看对比那个售价，就可以测算出来一个纯粹的不掌握资源的厂商去做推理，他能够赚多少钱。好吧，这是我觉得在Mass平台的定位优势上。另外，关于要不你先回，先翻译这个问题。

Speaker #4: So, in relation to your question about the positioning, we do have a unique positioning in the mass business. Namely, different from some of the full-stack cloud providers, which have their in-house or proprietary models, we do not have such models.

Speaker #4: And therefore, correspondingly, we do not have to sell those large language models that our affiliated companies have to offer. As a result, we're able to actually sell, and we actually encourage our sales team to sell, the models that our customers like the most.

Speaker #4: For example, the GRM, etc. So that’s number one. And secondly, is it quite important in today’s market to have your proprietary, or your own, computing power—which is the only way that you can actually secure significant profitability in this business.

Speaker #1: Gotcha. 关于价格变化，大家也都看到我们在存储和算力相关的领域其实都显著地提高了价格。在存储上面，我们的可能提高的比例也非常大。那么存储在这个计算场景下，它实际上是随着计算的增长而跟着增长。所以说客户们对于存储的涨价并不感觉很在意。所以我觉得目前我们看到都是很轻易地接受了我们的涨价。那么在这个涨价过程中，一方面我们追上了我们的成本上涨，另一方面我们也还是保持我们的利润，或者是提升了我们的利润率。另一方面，算力当然我们也是以同样的原理，就是算力的业务的机器的价格肯定是在增长的，但是我们因为我们的能力的独特性，那么我们一方面是我们会提供这个跟随着算力价格增长的指导的价格，但是我们会保持毛利率，甚至会提升我们的毛利率。另一方面就是我们同时也会接受客户托管的算力，由我们来组织和运营运维。那么在本期我们也见到了这样的场景落地。那么在这种场景下，我们这个业务的毛利润是显然是更高的。就这些。

Speaker #4: So, in relation to your question about the price hike, there are basically two products or solutions where we have implemented price increases: number one is storage, and number two is computing power.

Speaker #4: So I'll talk about them separately, respectively. So, in terms of storage, storage is usually—the incremental amount of storage actually comes with the intelligent computing demand.

Speaker #4: That is a relatively small portion of the intelligent computing overall ticket size. And therefore, in the vast majority of the customers that we negotiate with, they have relatively easily accepted such a price hike.

Speaker #4: In which case, as a result, we're actually able to, in some cases, not only pass through the increase in our cost, but also increase our profitability.

Speaker #4: In that scenario, number two, in terms of computing power, because of our specific capabilities, including path capabilities as well as the operating maintenance and network capabilities, again, we are able to pass through that cost hike into our customers in some of the cases.

Speaker #4: We also increased our profitability. In this quarter, we also have some projects where we are providing managed services, which is an SLI business model.

Speaker #4: We look forward to seeing more of that coming through and being reflected in the financial statements. Thank you.

Speaker #2: Thank you. We will take the next question. Your next question comes from Wei Zhong from UBS. Please go ahead. Your line is open.

Speaker #3: 好的，谢谢。管理层晚上好，恭喜这个季度非常强劲的业绩，也感谢接受我的提问。我有一个问题想请教。刚才咱们谈到了咱们平台上Mass的定位和差异化。那么我想追问一下，考虑到其他云厂商的自研模型和用户生态，我们应该如何思考平台在云行业的长期定位，以及可持续的稳定利润率？我也很快自己翻译一下。Good evening, management. Congrats on a solid quarter, and thank you for taking my question. Considering the proprietary models and user ecosystems of other cloud providers, how should we think about our long-term positioning in the cloud market and the sustainable margin level down the road?

Speaker #3: Thank you.

Speaker #1: 好，我来回答一下这个问题。其实刚才我们也提到了，就是Mass的服务，它其实一方面要提供顶尖的模型，另一方面要提供稳定可靠的服务。那么作为一个中立厂商，我们始终会和各家原厂保持良好的关系，并且在第一时间推出这些模型的线上服务，并且我们自有的资源和原厂的资源组合在一起，能够给客户提供一个高可用、高可靠的这个云上的Mass服务。那我想这一点来说，就是我们在这方面是没有任何所谓的心理负担的。另一方面，其实我们也可以看到，因为我们手里所掌握的资源的总量来看，那么通常情况下，我们在SLA方面能够给客户提供更高的保障。我想这一点，云厂商大规模的基础设施是对这个Mass业务的一个更好的支撑。那从利润率来的角度讲，一方面我们会和原厂配合，比如说我们和原厂可以配合合作在，比如说模型推理框架，包括一些模型在发布时候的权重未披露的权重的一些合作，来提升我们的这个推理效率。另一方面，我们也始终长期的在投入推理优化本身的工作。目前我们的团队其实在多个开源模型上，我们都已经验证了我们能够达到和甚至或者接近原厂的这个推理的效率。所以整体来说，这个业务的毛利水平，不管模型怎么更新，我们认为我们的团队也有这个能力是跟住这个新模型的这个推理效率，并且能够保持一个良好的毛利率。谢谢。

Speaker #4: So, we believe that Mass or Cloud AI Cloud Service Provider is important to be able to offer the top models, which the customers like, and also stable services to our customers.

Speaker #4: So, as mentioned, as a neutral cloud player, we are able to maintain good relations with all of those top model providers and large language model labs, and be able to provide the best model according to our customers' demands.

Speaker #4: And also, we're able to, based on our technology capabilities, we're able to provide highly available and highly reliable services to them. Based off out of the SLAs that we signed with them.

Speaker #4: I think, thirdly, in relation to the profitability question you asked, it is important to work closely with the LLM firms' labs to, for example, optimize the inference of those models.

Speaker #4: And that would include, for example, working with them based on the undisclosed weighting of the models to increase our inference, model inference efficiency.

Speaker #4: In some cases, we're able to get to a very close level, or even reach the same level, of inference efficiency as the LLM companies themselves.

Speaker #4: Thank you.

Speaker #2: Thank you. We will take our final question. And your final question comes from Ying Liu from Morgan Stanley. Please go ahead. Your line is open.

Speaker #5: 非常感谢提问的机会。首先，恭喜公司取得亮丽的业绩。我这边的问题是想问一下，公司现在在算力和MaaS两种商业模式下的投资回报率（ROIC）水平，以及这些回报的最近边际变化是向上还是向下的。谢谢。 Let me translate my question. I would like to ask, under the two business models—computing power leasing and Model as a Service—what is the ROIC for these two business models, and what is the recent marginal change for the ROIC?

Speaker #5: Thank you.

Speaker #6: Thank you, Liu Yang. At this stage, we don't disclose separate ROIC on Mass and AI computing power services, because ROIC varies across projects, driven by payback cycles, gross margins, fixed assets, and depreciation policies. Overall, Mass delivers much better profitability than AI computing power services.

Speaker #6: At this stage, we have seen continuous improvement in operating leverage. As our AI business scales up, fixed costs are steadily diluted, and our training track models’ adjusted operating profit has trended positive during a gradual recovery in our overall ROIC.

Speaker #6: We adhere to a demand-driven and disciplined AI investment strategy, with a strong focus on capital efficiency. With continuous business structure optimization and the maturing of AI commercialization, I think our overall ROIC will keep improving steadily.

Speaker #5: Okay, thank you.

Speaker #2: Thank you. There are no further questions. That concludes the question and answer session. I will now hand back for closing remarks.

Speaker #1: Okay, thank you all for joining us today. If you have any further questions, please contact our IR team. Have a good evening. You may now disconnect.

Speaker #1: Thank you.