文 | 崔鹏CP
谷歌刚刚交出了一份增长亮眼、却让华尔街感到不安的财报。
10月28日母公司Alphabet公布第三季度财报，其中营收同比增长25%，达到1276亿美元。
从业绩数据上看，谷歌的AI生意越做越大，云服务收入实现翻倍，自研芯片开始对外销售，似乎一切都在向好。
但亮眼的增长没有安抚市场情绪，财报后Alphabet的股价连跌两天，降幅超过了10%，投资者很担心搜索广告增长放缓和自由现金流转负的情况。
对AI的巨额投入在持续吞噬谷歌的现金流，华尔街非常关心，它还需要多久才能获得与庞大资本开支相匹配的盈利能力。
AI搜索的变现效率不及传统搜索
谷歌的AI转型已经触及最核心的搜索业务。
三季度搜索及其他收入为633.5亿美元，同比增长12%，低于此前两个季度的19%和17%。
如果单看搜索的业务数据，查询量的同比增速达到过去五年来的最高水平，说明用户正在更多使用AI搜索。
可惜这些新增查询量还没有带来对应的收入增长。谷歌管理层在财报电话会上公开承认，AI搜索的变现效率低于传统搜索。
传统搜索的商业模式相当成熟。用户输入关键词，搜索引擎展示网页链接和广告，广告主根据关键词竞争排名，用户点击后形成收入。
在这套体系中，搜索结果页既是信息入口，也是广告交易场所。
AI搜索改变了整个过程。用户可以直接向AI提出复杂问题，它会总结信息甚至帮助完成决策。用户不需要逐个点击网页链接，广告展示和点击的机会也随之变化。
虽然更长的对话、更复杂的需求可能带来新的广告场景，但从现阶段看，用户查询量增长，并不能直接等同于广告收入同比例增长。谷歌需要重新摸索广告展示方式、用户体验和广告主付费意愿之间的平衡。
如果因为广告加载影响AI搜索的体验，用户可能转向其他AI助手。如果AI搜索吸引了更多用户，却只能产生较低的单位收入，公司额外支出计算成本将无以回报。
这对谷歌来说是两难境地，如何兼顾用户体验与广告收入，是搜索业务转型的关键。
谷歌不一定要赢下模型竞争
与AI搜索面临的变现压力不同，谷歌云已经从AI需求增长中获得了更直接的收入。
三季度谷歌云营收为307亿美元，同比增长103%，远高于Alphabet整体营收的25%增速；经营利润达到111亿美元，远高于去年同期的36亿美元。
云业务积压订单达到5610亿美元，也反映出企业客户对计算基础设施和AI服务的强劲需求。
同时经营着模型、云平台、芯片和应用，谷歌似乎不必在每一场模型竞争中获胜也能从AI产业扩张中获益。
它的优势在于并不依赖Gemini等自有模型赚钱，企业训练和运行AI模型需要大量计算资源，谷歌可以售卖算力，也可以销售自研TPU芯片。
也就是说，即使客户不选择Gemini，也可能成为谷歌的客户。
Anthropic就是典型案例，它在周一发布的S-1文件凸显了其与谷歌的密切关系：Anthropic承诺未来七年向谷歌支付约1145亿美元的计算服务费用，并计划在2031年前花费约超过1250亿美元购买来自谷歌等供应商的芯片；谷歌同时持有Anthropic超过5%的股权，并承诺在特定情况下追加200亿美元投资。
谷歌与Anthropic即是大模型领域的竞争者，前者也是后者的投资人，以及其云服务和芯片的供应商。
这种相互竞争与合作的关系说明，AI产业链的商业利益并不完全取决于大模型竞争的结果。
财报数据证明谷歌的自研TPU业务同样拥有巨大外部需求。过去TPU主要用于谷歌自身的AI训练，如今谷歌开始向客户交付TPU系统，希望从芯片销售和云服务中同时获得收入。公司管理层预计，TPU相关收入将在2027年进一步增长。
云业务的高增长已经证明市场需求存在，能否在持续扩张中取得足够回报，则要看未来投入与收入的变化。
收入增速远低于成本增速
AI业务带来收入增长的同时，也在迅速推高谷歌的资金投入。
三季度Alphabet营收同比增长25%，资本开支增长却高达114%，来到了514亿美元，两者的增速差距非常明显。
其它核心数据表现也不太好，经营现金流同比下降12.6%，自由现金流从去年同期的245亿美元转为负91亿美元，远低于市场此前盈亏平衡的预期。
AI搜索需要更多计算资源，云业务需要花钱搞基建，自研芯片也需要持续研发和部署，这些都需要大量开支。
不仅如此，随着数据中心和服务器陆续投入使用，设备折旧、运维和能源成本也是谷歌需要长期负担的资本支出。
此前谷歌已将2026年资本开支预测上调至1950亿至2050亿美元，管理层甚至表示，2027年这个数字将大幅增加。
投资者需要判断，这些投入最终能否产生足够回报。即便收入继续增长，如果新增收入不足以覆盖新增成本，利润表现和现金流仍然会承受压力。
这种压力并非谷歌独有，三季度Meta同样出现了2012年以来的首次负自由现金流，凸显人工智能热潮的代价。
很多大型科技公司在为AI基础设施投入巨额资金，担心投入不足会失去竞争优势，将潜在市场拱手相让。
但如果所有公司都持续扩大建设规模，行业最终能够消化多少算力、这些算力又能产生多少利润，就成为每一份财报都无法回避的问题。
虽然谷歌的云服务和芯片业务增长迅速，但这并不能保证所有投入都能获得理想回报。
今年市场上出现过不少质疑AI叙事的声音，最重要的理由就是商业化进展缓慢，科技公司越来越依赖外部融资，资本开支成为肉眼可见的经营负担。
对谷歌而言，AI竞赛进入了一个更加现实的阶段。模型能力和算力储备固然重要，投资者最终还要看这些投入能否转化为健康的营收表现。
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。
Man, Choi PengCP
Google has just handed over a good-looking, unsettled financial report on Wall Street.
On 28 October, the parent company Alphabet released the third quarter of its financial report, in which revenue grew by 25 per cent each year to $1276 billion.
In terms of performance data, the larger the Google AI business, the more the income from cloud services doubles, and it seems that everything is getting better since the chips began to be sold.
But the bright-eyed growth did not appease market sentiment, and after the financial statements Alphabet's stock price fell by more than 10 per cent for two days, and investors were concerned about the slow growth of search advertising and the negative impact of free cash flows.
Wall Street is very concerned about the huge investment in AI, which continues to devour Google’s cash flow, and how long will it take before it will be able to achieve profitability commensurate with its huge capital expenditure?
The AAI search is not as efficient as the traditional search.
Google's AI transformation has touched on the core search business.
Three quarters of search and other revenue amounted to $63.35 billion, an increase of 12 per cent over the same period, down from 19 per cent and 17 per cent in the previous two quarters.
When looking at the search's business data alone, the number of queries rises at the same rate as in the past five years, indicating that users are using AI more.
Unfortunately, these additional queries have not yet resulted in a corresponding increase in income. Google management publicly acknowledged at the financial conference that the AI search was less efficient than the traditional search.
The traditional search business model is quite mature. Users enter keywords, search engines display web links and advertisements, advertisers are ranked according to keywords, and users click to generate income.
In this system, the search result pages are both an entry point and an advertising place.
AI search changed the process. Users can ask complex questions directly to AI, which summarizes information and even helps to complete decision-making. Users do not need to click on web links on a case-by-case basis, and the opportunities for advertising and clicking change.
While a longer dialogue and more complex demand may lead to new advertising scenes, at this stage, the increase in user queries cannot be directly equated with an increase in advertising revenues on a proportional basis. Google needs to rediscover the balance between the way advertising is displayed, the user experience and the will of the advertiser paid.
If the ad loads affect the AI search experience, users may turn to other AI assistants. If the AI search attracts more users but generates only lower unit revenues, the calculation of the cost of the company ' s extra expenses will be in vain.
This is a dilemma for Google, and balancing user experience with advertising revenues is key to the transformation of search operations.
Google doesn't have to win model competition.
Unlike the liquidity pressure on AI searches, Google Clouds has generated more direct revenues from the growth of AI demand.
The third quarter of Google Clouds collected $30.7 billion, an increase of 103 per cent over the same period, which was much faster than the 25 per cent increase in Alphabet ' s overall collection; operating profits reached $11.1 billion, well above the $3.6 billion received during the same period last year.
The backlog of orders for cloud operations reached $561 billion, also reflecting the strong demand from business clients for computing infrastructure and AI services.
At the same time, with models, cloud platforms, chips and applications, Google does not seem to have to win in every model competition to benefit from the expansion of the AI industry.
Its advantage is that it does not depend on its own models, such as Gemini, for money, and that the training and operation of the AI model requires a large amount of computing resources, and Google can sell its own TPU chips.
In other words, even if the client does not choose Gemini, it can be Google customers.
Anthropic is a case in point, and its close relationship with Google was highlighted in the S-1 document of Monday publish: Anthropic committed approximately $114.5 billion in computing services to Google for the next seven years and planned to spend some $12.5 billion by 2031 on the purchase of chips from suppliers such as Google; Google also held more than 5 per cent of Anthropic's equity and committed to additional investments of $20 billion in specific circumstances.
Google and Anthropic are competitors in the field of LLM, who are also investors in the latter and suppliers of their cloud services and chips.
This competitive and cooperative relationship suggests that the commercial interests of the AI industrial chain do not depend entirely on the outcome of LLM competition.
Financial data confirm that Google's self-study TPU operations also have significant external needs. In the past, the TPU was used mainly for Google's own AI training, and now Google has started to deliver TPU to its clients in the hope of generating income from both chip sales and cloud services. The management of the company expects that TPU-related revenues will increase further in 2027.
The high growth of cloud operations has demonstrated the existence of market demand, and whether there will be sufficient returns from continued expansion will depend on changes in future inputs and revenues.
Income growth is much slower than cost growth
The growth in revenue generated by AI ' s operations is being accompanied by a rapid increase in Google funding.
In the third quarter, the Alphabet battalion increased by 25 per cent over the same period, while capital expenditure grew by 114 per cent, reaching $51.4 billion, with a significant increase gap.
Other core data also performed poorly, with operating cash flows falling by 12.6 per cent, and free cash flows reverting from $24.5 billion in the same period last year to negative $9.1 billion, well below previous market expectations of a balance of gains and losses.
Additional computing resources are required for AI searches, for cloud operations, for infrastructure, and for self-study chips, for ongoing development and deployment, which will require significant expenditures.
Moreover, as data centres and servers become operational, depreciation of equipment, peacekeeping and energy costs are also capital expenditures that Google needs to bear in the long term.
Earlier, Google had revised its 2026 capital expenditure projections upwards to 1950 to $205 billion, and management had even indicated that the figure would increase significantly in 2027.
Investors need to judge whether these inputs will ultimately yield sufficient returns. Even if income continues to grow, profit performance and cash flows will continue to be under pressure if the new income does not cover the new costs.
This pressure is not unique to Google, as is the third quarter of Meta ' s first negative free cash flow since 2012, highlighting the cost of artificial intelligence booms.
Many large technology companies are investing huge amounts of money in AI infrastructure, fearing that the lack of investment will lose their competitive edge and give way to potential markets.
But if all companies continue to expand their construction, the extent to which the industry can eventually absorb and the profits from these calculations becomes an inescapable issue for every financial account.
While Google ' s cloud service and chip operations are growing rapidly, this does not guarantee the desired return on all inputs.
The most important reason for this year ' s challenges to the AI narrative is the slow pace of commercialization, the increasing reliance of technology companies on external financing and the visible burden of capital expenditure on their operations.
For Google, the AI competition entered a more realistic phase. Modelling capacity and a financial reserve are important, and investors will ultimately have to see whether these inputs translate into healthy income performance.
Auto-fetched and translated by AI Gem Box. All rights belong to the original author.