智东西（公众号：zhidxcom）
作者 | 王涵
编辑 | 李水青
智东西10月29日报道，今日凌晨，谷歌发布了其最强的旗舰模型Gemini 4 Carbon，该模型专为持续运行的Agentic工作流而构建，重点提升了编程、智能体和科学任务，以及全栈Web开发和多模态理解等能力。
该模型在最近更新的v5.0版Artificial Analysis Intelligence Index上以59分排名第一，超越Claude opus 5.5、GPT-6 Astra（max）和GPT-6.1 Sol（max）等一众模型。
价格方面，Gemini 4 Carbon的首发优惠价格为每百万个输入Token收费2美元，每百万个输出Token收费10美元，与Argon首发优惠价一致。并且，缓存输入Token的价格还在输入Token单价基础上可享95%减免，也就是说只需支付原价的5%。
同时，谷歌母公司Alphabet公布2026年第三季度财报。本季度，Alphabet营收达1275.77亿美元（约合人民币8504.35亿元），同比增长约25%，实现连续13个季度两位数增长；净利润达261.7亿美元（约合人民币1744.51亿元），同比下跌约25%；摊薄后每股收益为2.12美元（约合人民币14.13元），同比下跌约26%。
其中，谷歌云的营收同比增长103%至307.09亿美元（约合人民币2047.08亿元），成为本季度最亮眼的业务板块。同期，微软Azure云计算业务营收约332亿美元，同比增长48%。相比之下，Azure的营收依旧领先，但谷歌云增速超过Azure两倍。
在随后举行的财报电话会上，谷歌进一步披露了AI业务的商业化进展：谷歌云积压订单达到5610亿美元（约合人民币3.74万亿），超过一半预计将在未来24个月转化为收入。
Alphabet第三季度资本开支达到513.55亿美元（约合人民币3423.35亿元），较上年同期增长约114%。财务负责人阿什肯纳齐称，当季资本开支的绝大部分用于支持AI投资所需的技术基础设施，其中约65%用于服务器和网络设备，其余约35%用于数据中心。
截至美东时间10月28日收盘时，谷歌A类股报324.89美元/股（约合人民币2165.73元/股），较前一交易日上涨约1.34%，总市值约3.97万亿美元（约合人民币26.48万亿元）。财报发布后，截至发稿前，谷歌A的盘后股价下跌约9.97%，至292.50美元/股（约合人民币1949.82元/股）。
一、Gemini 4 Carbon发布，智能指数排名第一
财报发布当天，谷歌宣布推出Gemini 4 Carbon。这是继9月底发布Gemini 4 Argon、10月中旬向开发者开放Argon之后，Gemini 4系列的又一次迭代。
根据谷歌官方介绍，Carbon是其目前最强的前沿模型，重点强化了复杂编程、Agent任务执行和网络安全防御能力。
与侧重单次回答质量的模型应用相比，谷歌更强调Carbon处理长周期任务的能力：模型能够在数天内持续推进复杂工程任务，即使遇到死胡同，也能尝试调整方案，最终完成多步骤工作。
谷歌透露，内部团队已经使用Carbon处理大型代码库中的复杂问题，包括定位和修复隐蔽Bug、优化底层系统。
该模型还针对全栈Web开发、科学任务、多模态理解等领域进行了改进。在Web开发评测WebDev Arena中，Carbon的Elo评分提升89分，达到1769分，排名第四，超过GPT-6.1 Sol和Claude Fable 5.1，仅次于Claude Opus 5.5、GPT-6 Astra和Claude Sonnet 5.5。
基准测试方面，编程能力上，Carbon在SWE-sweep和Terminal-Bench 4.0上排名第一。但在FrontierSWE 2上，GPT-6 Astra以65.5%排名第一，Carbon的61.2%低于 GPT-6 Astra和Claude Opus 5.5。
在计算机操作测试中，Carbon的Agents’ Last Exam v1.1得分为77.4%，高于GPT-6 Astra、Claude Opus 5.5。在实际计算机环境操作OSWorld-2.1离线子集中，Claude Opus 5.5以87.2%领先，Carbon为81.5%，排名第二。
知识工作测试则是Carbon表现相当突出的领域。在AutomationBench上，它得到53.9%，明显高于GPT-6 Astra和Claude Opus 5.5。在Vending-Bench 2模拟经营测试中，Carbon最终获得16484美元的模拟账户余额，高于GPT-6 Astra。
Carbon在图表理解和视觉相似度评测中均排名第一，上下文处理的成页也远高于GPT-6 Astra和Claude Opus 5.5。科学与数学领域同样如此，在RiemannBench上，Carbon得分80.0%，比Argon高4%；在Terminal-Bench-Science 0.1上，Carbon得分64.8%，领先于GPT-6 Astra。
网络安全是此次升级的另一重点。谷歌公布的测试数据显示，Carbon在SEC-bench Pro漏洞发现评测中的成绩为88.2%，在ExploitGym漏洞利用评测中达到54.8%，在CWE-bench漏洞修复评测中达到71%。
谷歌称，在内部测试中，Carbon生成安全代码的成功率超过Argon的2倍；Chrome团队也报告称，在高质量补丁生成任务上，Carbon的表现超过另一家AI实验室的一款前沿模型10倍以上。
除了能力提升，Carbon也针对模型可靠性和Agent执行边界进行了调整。谷歌称，在内部测试中，Carbon的高严重性失准警报减少近80%，无依据陈述约减少一半；模型越权执行的情况减少26%，执行不足的情况减少29%。在间接提示注入攻击评测中，Carbon的攻击成功率为0%，即在该测试所覆盖的攻击尝试中均成功抵御。
Carbon此次采取的是分阶段开放策略。谷歌首先向部分网络安全防御人员开放模型，并通过Fairwind计划逐步扩大可信合作伙伴的使用范围，之后再向开发者、企业和消费者开放。谷歌称，公司仍在参与美国政府的发布前模型测试流程，并将根据早期用户反馈持续调整安全措施。
在今日举办的财报会上，谷歌CEO桑达尔·皮查伊（Sundar Pichai）宣布，谷歌模型API目前每分钟处理近320亿个Token，较一年前增长超过4倍，较上一季度增长40%。与此同时，Gemini应用用户数达到10亿，成为谷歌历史上增长最快的产品。
谷歌还在推动AI Agent从独立工具走向企业业务流程。谷歌称，Gemini Enterprise过去12个月已被超过2600家客户用于处理至少1000亿个Token，近75家客户的使用量达到10万亿Token；财富100强企业中，近90%已使用Gemini Enterprise。
二、谷歌云营收翻倍，5610亿美元积压订单
谷歌云成为Alphabet第三季度最值得关注的业务。
财报显示，谷歌云第三季度营收达到307.09亿美元（约合人民币2047.08亿元），较上年同期增长103%。
同期，该业务营业利润达到111.21亿美元（约合人民币741.33亿元），较上年同期增长约209%；营业利润率从去年同期的23.7%提升至36.2%。
在财报会上，谷歌云CEO托马斯·库里安（Thomas Kurian）称，云业务的增长来自覆盖芯片、模型、数据、安全和智能体平台的整合式AI产品组合，同时受益于不同产品线、行业和客户群体的需求。谷歌云目前已有18条产品线的年收入分别超过10亿美元。
从具体业务来看，Google Cloud Platform（GCP）的AI基础设施、AI解决方案以及传统核心云服务均实现增长，TPU系统销售也成为重要的收入贡献来源。
库里安透露，谷歌正在将TPU的销售范围从自有云平台拓展至客户自有数据中心和其他服务商，以扩大产品覆盖面。
谷歌披露，超过100家重要客户已经在使用谷歌TPU系统。除Anthropic等现有合作方外，公司还在与Mistral、Physical Intelligence等模型公司，以及金融机构、大学和公共部门客户合作。谷歌还宣布与Oracle、CoreWeave AI建立TPU系统交付合作，并与Meta合作推进TorchTPU，以降低开发者将PyTorch模型迁移至TPU平台的门槛。
TPU业务的意义在于，谷歌可以向客户销售包含硬件、计算、存储、网络及系统软件的完整计算系统，而不只通过云服务按使用量收费。
与此同时，TPU系统销售也会带来不同于传统云服务的现金流周期：谷歌需要先采购硬件、积累库存，再向客户交付并确认收入。财务负责人阿纳特·阿什肯纳齐（Anat Ashkenazi）在会上解释，库存建设会先占用现金，随着交付和客户付款推进，相关现金才会逐步回流。
比单季度营收更能反映未来业务规模的，是云业务积压订单。阿什肯纳齐透露，谷歌云第三季度积压订单达到5610亿美元，其中大部分来自常规GCP合同。公司预计，超过55%的积压订单将在未来24个月确认为收入。
这一数字意味着，谷歌云已经积累了规模庞大的未来合同需求，但积压订单并不等同于已实现收入，实际确认节奏仍取决于合同执行、客户使用情况和交付能力。
管理层同时强调，市场需求仍受到计算资源供给的制约。谷歌正在增加可用容量，以满足AI基础设施和解决方案的长期需求。
由于供给紧张，公司已在第三季度末开始使用第三方计算资源，并预计第四季度扩大使用规模。这些额外资源短期内可能对云业务利润率形成压力。
三、月视觉搜索量已经超过500亿次，AI Mode变现水平相对较低
AI对Alphabet的影响也正在深入其最成熟的业务——搜索和广告。
第三季度，谷歌搜索及其他业务营收为633.50亿美元（约合人民币4222.97亿元），同比增长12%。皮查伊称，搜索查询量同比增长率达到过去5年来的最高水平，AI投资正在推动整体查询量增长，包括商业相关查询。
在产品层面，谷歌正在整合AI Overviews、AI Mode和传统搜索体验。皮查伊称，AI Mode中的查询平均长度约为传统搜索的3倍，全球约1/5的AI Mode查询使用语音、图片或文档等多模态输入；谷歌每月视觉搜索量已经超过500亿次。
这些变化意味着，用户正在通过更长、更复杂的提问完成信息检索和决策。但对于依赖广告收入的搜索业务而言，查询方式变化也带来了新的变现问题。
在财报会上，谷歌首席商务官菲利普·辛德勒（Philipp Schindler）解释称，AI Mode中的多数查询属于增量查询，能够带来新的商业意图和未来的变现机会。
谷歌正在将AI Overviews和AI Mode中的广告扩展至更多用户和查询，并首次将AI Mode广告的覆盖范围拓展到国际市场，同时在美国推出多轮交互广告。
谷歌也披露了AI对广告主工具的影响。第三季度，Gemini模型帮助广告主生成超过15亿个广告活动素材，同比增长超过7倍。AI Max等工具正在帮助广告主处理更长、更复杂以及多语言的搜索查询。
不过，AI搜索的商业化仍处于调整阶段。辛德勒在问答环节承认，AI Mode的变现水平相对较低，对搜索业务增速形成一定影响。管理层表态，正在逐步扩大广告覆盖面，并优先关注用户体验和广告主投资回报。
四、自由现金流转负，资本开支暴涨114%
与业务增长相伴的，是持续上升的基础设施投入。
Alphabet第三季度资本开支达到513.55亿美元（约合人民币3423.35亿元），较上年同期增长约114%。财务负责人阿什肯纳齐称，当季资本开支的绝大部分用于支持AI投资所需的技术基础设施，其中约65%用于服务器和网络设备，其余约35%用于数据中心。
谷歌维持2026年全年资本开支1950亿至2050亿美元的指引，并预计2027年资本开支将显著增加。
资本开支增加并非唯一的资金压力来源。阿什肯纳齐指出，AI基础设施扩张还将通过更高的折旧费用和数据中心运营成本持续影响利润表，其中包括能源等相关支出。谷歌同时需要为TPU系统销售提前建设库存，这也会占用现金。
第三季度，Alphabet经营活动产生的净现金为422.90亿美元，低于513.55亿美元的资本开支。按照公司披露的口径，当季自由现金流为负90.65亿美元；过去12个月的自由现金流为197.47亿美元。
阿什肯纳齐称，谷歌拥有通过搜索、云、订阅服务等多类业务实现AI投资变现的条件，并预计本轮投资周期能够取得高于资本成本的回报。但她也承认，技术基础设施投入和TPU库存建设将继续给自由现金流带来压力。
结语：云需求持续上涨，谷歌要把AI算力投入变成利润
从财报和财报会，我们可以看出谷歌正在进入一个需要同时管理需求、产能和资金回报的阶段。谷歌云积压订单不断增加，模型调用量和企业Agent应用持续扩大，都需要相应的计算资源支撑。与此同时，基础设施建设需要提前投入，相关收入和现金回流却分布在不同时间段。
对于谷歌而言，下一阶段的关键不仅是扩大AI服务的市场规模，还包括确保产能能够及时交付，并将基础设施投入转化为足够的利润和现金流。
本文由 AI宝库自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Intellectual (WeChat Official Account:zhidxcom)
Author Wang Xin
The editor, Li Shuqing
In the early morning hours of this morning, Genini 4 Carbon, the largest Google publish, reported on 29 October, was built for the continuing Agentic Z52Z, focusing on programming, AI agent and scientific missions, as well as all Web development and multi-modular understanding.
The model is ranked first in 59 on the recently updated V5.0 version of Artificial Analysis Inteligence Index, exceeding the various models Claude opus 5.5, GPT-6 Astra (max) and GPT-6.1 Sol (max).
On the price side, Gemini 4 carbon's initial preferential price is $2 per million input token and $10 per million output token, in line with Argon's initial offer. Furthermore, cache entered Token ' s price at a 95 per cent reduction on the Token unit price, i.e. only 5 per cent of the original price.
Meanwhile, Alphabet, the Google parent company, published the third quarter of 2026. This quarter, Alphabet received $127,577 million (approximately RMB 85,435 million), an increase of about 25 per cent over the same period and a double-digit increase over 13 consecutive quarters; net profits of $26,170 million (approximately RMB 17,44,451 million), a decline of about 25 per cent over the same period; and a spread of $2.12 (approximately RMB 14.13) per share, a decline of about 26 per cent over the same period.
Of this, Google Clouds earned between $103 per cent and $30,709 million over the same year (approximately RMB 20,470.8 million), making it the brightest business block of the quarter. During the same period, Microsoft Azure cloud computing operations received approximately $33.2 billion, an increase of 48 per cent over the same period. By contrast, Azure still leads the way, but Google clouds are growing twice as fast as Azure.
At the subsequent press conference, Google further revealed progress in the commercialization of AI ' s operations: Google cloud backlog orders amounted to $561 billion (approximately RMB 3.74 trillion), with more than half expected to translate into revenue over the next 24 months.
Capital expenditure in Alphabet in the third quarter amounted to $5,355 million (approximately RMB 34,235 million), an increase of approximately 114 per cent over the same period of the previous year. The Chief Finance Officer, Ashkenazi, stated that the vast majority of the current-period capital expenditure was spent on supporting the technical infrastructure required for AI investments, about 65 per cent of which was for servers and network equipment and the remaining approximately 35 per cent for data centres.
At the time of closing on 28 October in Eastern United States time, Google A reported $324.89 per unit (approximately $2165.73 per unit), an increase of approximately 1.34 per cent over the previous transaction date and a total market value of approximately $3.97 trillion (approximately $26.48 trillion per unit). After the financial paper publish, Google A's post-card share price fell by about 9.97 per cent to $292.50 per share (approximately RMB 1949.82/equity) as of the time of the submission.
I. Gemini 4 CarbonZ94Z, number one intelligence index
On the same day, Google announced the launch of Gemini 4 Carbon. This is another iterative version of the Gemini 4 series after publishGemini 4 Argon was opened to developers in mid-October.
According to the official Google presentation, Carbon is now its strongest front-line model, with a focus on enhancing complex programming, Agent mission implementation and cybersecurity defence capabilities.
Google places more emphasis on the ability of carbon to handle long-cycle tasks than on the application of models that focus on single-response quality: The model is able to carry forward complex engineering tasks on an ongoing basis for a period of days and, even in the event of a dead end, can attempt to adjust the programme and eventually complete the multi-step exercise.
Google revealed that the internal team had used carbon to address complex issues in the large code repository, including locating and repairing the hidden Bug and optimizing the bottom system.
The model has also been improved in areas such as the whole of Web development, scientific missions, multi-modular understanding. In WebDev Arena, the Elo rating of Carbon rose by 89 points to 1769, ranking fourth, exceeding GPT-6.1 Sol and Claude Fable 5.1, after Claude Opus 5.5, GPT-6 Astra and Claude Sonet 5.5.
With regard to baseline testing and programming capacity, Carbon ranked first on SWE-sweep and Terminal-Bench 4.0. However, in FrontierSWE 2, GPT-6 Astra ranked first at 65.5 per cent, and 61.2 per cent of Carbon was lower than GPT-6 Astra and Claude Opus 5.5.
In the computer operations test, Agents ' Last Exam v. 1.1 of Carbon was divided into 77.4 per cent, which was higher than GPT-6 Astra, Claude Opus 5.5. In the operational computer environment OSWorld-2.1offline, Claude Opus 5.5 leads by 87.2% and Carbon by 81.5%, ranked second.
The knowledge work test is an area where Carbon's performance is remarkable. On Automation Bench, it received 53.9 per cent, significantly higher than GPT-6 Astra and Claude Opus 5.5. In the Vending-Bench 2 simulations, Carbon eventually obtained a mock account balance of $16,484, higher than GPT-6 Astra.
Carbon ranked first in the chart understanding and visual similarity assessment, and context processed pages much higher than GPT-6 Astra and Claude Opus 5.5. The same is true for science and mathematics, where in Riemann Bench, Carbon scored 80.0 per cent, 4 per cent higher than Argon, and in Terminal-Bench-Science 0.1, Carbon scored 64.8 per cent, ahead of GPT-6 Astra.
Cybersecurity is another focus of this upgrade. The test data published by Google show that carbon ' s performance in the assessment of the SEC-bench Pro gap was 88.2 per cent, 54.8 per cent in the evaluation of the Expluit Gym loophole and 71 per cent in the assessment of the CWE-bench gap repair.
Google stated that, in internal testing, the success rate from Carbon generated the safety code was twice as high as Argon; the Chrome team also reported that Carbon performed more than 10 times more than a forward model from another AI laboratory on a high-quality patch-generation mission.
In addition to capacity enhancement, Carbon has also made adjustments to the model ' s reliability and Agent ' s operational boundary. According to Google, in internal tests, the high-level criticality misalignment alert of carbon has been reduced by nearly 80 per cent and the unfounded statement has been reduced by about half; the over-implementation of the model has been reduced by 26 per cent and the under-implementation by 29 per cent. Carbon ' s attack success rate was 0 per cent in the indirect infusion assessment of the attack, which was successfully resisted in the attack attempt covered by the test.
Carbon has adopted a phased approach to opening up. Google first opened the model to selected network security guards and, through the Fairwind plan, gradually expanded the use of credible partners before opening it to developers, businesses and consumers. Google stated that the company was still participating in the publish pre-model testing process in the United States Government and would continuously adapt security measures to early user feedback.
At today ' s press conference, the Google CEO Sundar Pichai announced that the Google model API currently handles nearly 32 billion Tokens per minute, more than four times more than a year ago and 40 per cent more than the previous quarter. At the same time, the number of users of Gemini applications reached 1 billion and became the fastest-growing product in Google history.
Google is also promoting AI Agent from independent tools to business processes. Google reports that Gemini Enterprise has been used by more than 2600 clients over the past 12 months to handle at least 100 billion Tokens, and that nearly 75 clients have used 10 trillion Tokens; nearly 90 per cent of the wealth 100 enterprises have used Gemini Enterprise.
Double Google Cloud Camp, $561 billion in backlog orders
Google Clouds became the most interesting business for Alphabet in the third quarter.
According to the press, the third quarter of Google Clouds received $30,709 million (approximately RMB 20,470.8 million), an increase of 103 per cent over the same period of the previous year.
During the same period, the business profits amounted to $11,121 million (approximately $7,413.3 million yuan), an increase of some 209 per cent over the same period of the previous year; the business profits increased from 23.7 per cent to 36.2 per cent over the same period last year.
At the press conference, the Google Clouds CEO Thomas Kurian stated that the growth of the cloud business came from an integrated set of AI products covering chips, models, data, security and AI agent platforms, while benefiting from the needs of different product lines, industries and client groups. Google Clouds now has 18 product lines with annual revenues of over $1 billion.
In terms of specific operations, Google Cloud Platform (GCP) has experienced growth in AI infrastructure, AI solutions and traditional core cloud services, and TPU sales have become important sources of revenue contributions.
Curian revealed that Google was expanding the distribution of TPU from its own cloud platform to client-owned data centres and other service providers to expand product coverage.
Google revealed that more than 100 important clients were already using GoogleTPU. In addition to existing partners such as Anthropic, the company is working with model companies such as Mistral, Physical Inteligence, as well as with financial institutions, universities and public sector clients. Google also announced cooperation with Oracle, CoreWeave AI to establish TPU delivery and with Meta to advance TorchTPU to lower the threshold for developers to migrate the PyTorch model to the TPU platform.
The significance of the TPU operation is that Google can sell to its customers complete computing systems containing hardware, computing, storage, network and system software, rather than charging for usage only through cloud services.
At the same time, TPU sales will lead to a cash flow cycle that differs from traditional cloud services: Google will need to procure hardware, accumulate inventory and then deliver to customers and recognize revenue. At the meeting, Anat Ashkenazi, the chief finance officer, explained that the build-up of the inventory would first take over cash and that, as delivery and client payments advanced, the related cash flow would be gradually reversed.
What is more reflective of the scale of future operations than a single-quartered battalion is the backlog of orders for cloud operations. Ashkenazi revealed that Google Clouds had a backlog of orders amounting to $561 billion in the third quarter, the bulk of which came from regular GCP contracts. The company expects that over 55 per cent of the backlog of orders will be recognized as revenue in the next 24 months.
This figure means that Google Clouds has accumulated a large amount of future contract demand, but the backlog of orders is not equal to the realized income, and the pace of actual confirmation remains dependent on contract implementation, client use and delivery capacity.
At the same time, management stressed that market demand remained constrained by the computation of resource availability. Google is increasing its available capacity to meet the long-term needs for AI infrastructure and solutions.
Owing to supply constraints, the company started using third-party computing resources at the end of the third quarter and is expected to expand in the fourth quarter. These additional resources may put pressure on the profitability of cloud operations in the short term.
The number of monthly visual searches has exceeded 50 billion, and AI Mode is relatively low.
AI's impact on Alphabet is also reaching out to its most mature business — search and advertising.
In the third quarter, Google Search and Other Business Battalions received $63,350 million (approximately 42,297 million yuan), an increase of 12 per cent over the same period. According to Pichai, the search search search volume grew at the highest rate in the past five years on a year-to-year basis, and AI investments are driving the growth of the overall search volume, including business-related queries.
At the product level, Google is integrating AI Overviews, AI Mode and traditional search experiences. According to Pichai, the average length of queries in AI Mode is about three times that of traditional searches, and about one fifth of the AI Mode queries worldwide use multi-modular inputs such as voice, pictures or documents; Google has over 50 billion visual searches per month.
These changes mean that users are completing information retrieval and decision-making through longer and more complex questions. However, for search operations that rely on advertising revenues, changes in search patterns also pose new liquidity problems.
At the press conference, Philip Schindler, Chief Commercial Officer of Google, explained that most of the inquiries in AI Mode were incremental and could provide new business intentions and future liquidity opportunities.
Google is expanding advertising in AI Overviews and AI Mode to more users and queries and, for the first time, to the international market, with multiple rounds of interactive advertising in the United States.
Google also revealed the impact of AI on the main advertising tool. In the third quarter, the Gemini model helped advertisers to generate more than 1.5 billion advertising activities asset, a sevenfold increase over the same period. Tools such as AI Max are helping advertisers to handle longer, more complex and multilingual searches.
However, the commercialization of the AI search is still at the adjustment stage. In its question and answer session, Sindler acknowledged that the relatively low level of realization of AI Mode had an impact on the speed of the search operation. Management stated that it was gradually expanding the coverage of advertising, giving priority to user experience and investment returns for advertisers.
IV. Flows of free cash turned negative and capital expenditure surged by 114 per cent
Business growth is accompanied by rising infrastructure inputs.
Capital expenditure in Alphabet in the third quarter amounted to $5,355 million (approximately RMB 34,235 million), an increase of approximately 114 per cent over the same period of the previous year. The Chief Finance Officer, Ashkenazi, stated that the vast majority of the current-period capital expenditure was spent on supporting the technical infrastructure required for AI investments, about 65 per cent of which was for servers and network equipment and the remaining approximately 35 per cent for data centres.
Google maintains the guidelines for capital expenditure of between $5.5 billion and $20.5 billion for the whole year of 2026 and expects a significant increase in capital expenditure in 2027.
Increased capital expenditure is not the only source of financial pressure. Ashkenazi noted that the expansion of AI infrastructure would also continue to affect profit statements through higher depreciation costs and data centre operating costs, including related expenditures such as energy. Google also needs to build up its inventory for the early sale of the TPU system, which will also take up cash.
In the third quarter, the net cash generated by Alphabet ' s operations amounted to $42,290 million, less than $51,355 million in capital expenditure. According to the company's disclosed calibre, free cash flows for the current season were negative $9,065 million; and free cash flows for the past 12 months were $19,747 million.
According to Ashkenazi, Google has the conditions to realize AI investments through multiple types of operations, such as search, cloud, subscription services, and expects a return on the current investment cycle above capital cost. She also acknowledged, however, that technical infrastructure inputs and TPU inventory construction would continue to put pressure on free cash flows.
CONCLUSION: Cloud demand continues to rise, Google is going to turn AI ' s calculator into a profit.
From the press and the press, we can see that Google is entering a phase in which demand, capacity and financial returns are managed simultaneously. The growing backlog of Google Cloud orders, model calls and the continuous expansion of the Enterprise Agent application need to be supported by corresponding computing resources. At the same time, infrastructure needs to be built ahead of schedule, while the related revenues and cash flows are distributed over time.
For Google, the key in the next phase is not only to expand the market size of AI services, but also to ensure that capacity is delivered in a timely manner and that infrastructure inputs are translated into sufficient profits and cash flows.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.