博通的定制芯片和网络设备显示，AI支出正从通用图形处理器扩展到更广泛的领域。
戴尔、Snowflake和Palo Alto揭示了下一个瓶颈：系统、可用数据和安全。
当预期极端时，强劲的业绩仍可能令人失望，这使得现金创造和执行更加重要。
人工智能繁荣开始看起来不再像单一交易，而更像一条供应链。
在9月1日至2日之间，戴尔科技、Palo Alto Networks、Snowflake和博通均报告了强劲的AI相关需求。然而，投资者反应各异。戴尔上涨，Snowflake盘后大涨，而Palo Alto下跌，博通小幅下滑。
AI支出仍然强劲。更难的问题是哪些领域的经济效益最佳，以及投资者已为此支付了多少价格。
博通设计定制AI处理器以及连接数据中心内数千颗芯片的网络技术。AI半导体收入同比增长超过两倍，而管理层现在预计在2027财年AI半导体收入约为115亿美元。
大型AI开发商越来越希望芯片能围绕自身工作负载进行设计。对于特定任务，定制芯片在成本或能效上可能优于通用图形处理器。博通还销售连接这些芯片的网络设备，使其同时涉足两大瓶颈。
然而财报发布后股价下滑。近期指引大致符合市场预期，而市场预期本就极高。强劲的需求可能是真实的，但可能仍不足以支撑股价。
戴尔组装服务器和存储系统，将芯片转化为可运行的基础设施。AI服务器订单创下历史新高，公司再次上调展望，确认客户正从计划转向部署。
现金流值得关注。戴尔在季度营收为47亿美元的情况下，产生了约1亿美元的标准自由现金流。其调整后数字更高，因为它加回了与客户融资和租赁设备相关的现金。硬件增长仍可能需要大量现金。
Snowflake帮助公司存储、组织和使用数据。产品收入加速增长，指引上调，管理层表示AI产品推动了近期增长加速的约一半。当AI能够安全地与公司自有数据协作时，其价值会更大。
Palo Alto Networks提供安全层。更多的互联数据、模型和自动化代理带来了更多需要保护的东西。业绩强劲，但股价下跌，因为投资者质疑增长有多少来自收购，而非基础业务。
博通控制着稀缺的定制计算和网络技术。戴尔受益于物理部署，但占用了更多资本。Snowflake将通过数据使用变现。Palo Alto受益于日益增长的安全复杂性。
对投资者而言，更好的问题应从“谁有AI敞口？”转变为“这家公司控制着什么瓶颈，以及它如何高效地将需求转化为现金？”
主要风险在于基础设施支出最终可能超过实际AI使用。客户集中度也很重要，尤其是当少数超大规模厂商驱动需求时。市场预期仍然苛刻。
AI热潮的第一阶段奖励了稀缺的计算能力。这四份财报表明，第二阶段正变得更加有趣。
博通显示定制芯片和网络正成为AI基础设施的重要第二支柱。戴尔证实实体部署依然强劲。Snowflake显示AI正在进入企业数据领域，而Palo Alto则显示更多自动化带来了更多控制需求。
经验之谈并非猜测哪家公司的标志能在AI竞赛中胜出，而是要问：谁控制着瓶颈？谁能获得经常性收入？谁将需求转化为现金？随着AI向技术栈各处扩散，这些问题可能比整体增长率更为重要。
Broadcom’s custom chips and networking show AI spending is broadening beyond general-purpose graphics processing units.
Dell, Snowflake and Palo Alto reveal the next bottlenecks: systems, usable data and security.
Strong results can still disappoint when expectations are extreme, making cash generation and execution more important.
The artificial intelligence boom is starting to look less like one trade and more like a supply chain.
Between 1 and 2 September 2026, Dell Technologies, Palo Alto Networks, Snowflake and Broadcom all reported strong AI-linked demand. Yet investors reacted differently. Dell rallied, Snowflake jumped after hours, while Palo Alto fell and Broadcom slipped.
AI spending still looks strong. The harder question is where the best economics sit, and how much investors have already paid for them.
Broadcom designs custom AI processors and the networking technology connecting thousands of chips inside data centres. AI semiconductor revenue more than tripled year on year, while management now expects about 115 billion USD of AI semiconductor revenue in fiscal 2027.
Large AI developers increasingly want chips designed around their own workloads. Custom chips can offer better cost or power efficiency than general-purpose graphics processing units for specific tasks. Broadcom also sells the networking equipment connecting them, giving it exposure to two bottlenecks at once.
Yet the shares slipped after results. Near-term guidance landed around market expectations, which were already enormous. Strong demand can be real and still not be strong enough for the share price.
Dell assembles servers and storage systems that turn chips into working infrastructure. AI server orders hit record levels and the company raised its outlook again, confirming that customers are moving from plans to deployments.
Cash flow deserves attention. Dell generated about 1 billion USD of standard free cash flow on 47 billion USD of quarterly revenue. Its adjusted figure is higher because it adds back cash tied up in customer financing and leased equipment. Hardware growth can still require plenty of cash.
Snowflake helps companies store, organise and use their data. Product revenue accelerated, guidance rose and management said AI products drove roughly half of the recent growth acceleration. AI becomes more useful when it can work safely with a company’s own data.
Palo Alto Networks provides the security layer. More connected data, models and automated agents create more things to protect. Results were strong, but the shares fell as investors questioned how much growth came from acquisitions versus the underlying business.
Broadcom controls scarce custom computing and networking technology. Dell benefits from physical deployment but ties up more capital. Snowflake monetises data usage. Palo Alto benefits from growing security complexity.
For investors, the better question is shifting from “who has AI exposure?” to “what bottleneck does this company control, and how efficiently does it turn demand into cash?”
The main risk is that infrastructure spending eventually runs ahead of real AI usage. Customer concentration also matters, especially when a few hyperscalers drive demand. Expectations remain demanding.
The first phase of the AI boom rewarded scarce computing power. These four earnings reports suggest the second phase is becoming more interesting.
Broadcom shows custom silicon and networking becoming a serious second pillar of AI infrastructure. Dell confirms physical deployment remains strong. Snowflake shows AI moving into enterprise data, while Palo Alto shows that more automation creates more need for control.
The lesson is not to guess which logo wins the AI race. It is to ask who controls a bottleneck, who gets paid repeatedly and who converts demand into cash. As AI spreads through the stack, those questions may matter more than the headline growth rate.
正在载入文档……