--- 格式版本: 2 标题: "Microsoft, Alphabet, Meta Pivot from Buy to Build in AI" 原文链接: "https://www.datacenterknowledge.com/data-center-construction/hyperscalers-say-ai-race-has-entered-a-new-phase" 发布日期: "2026-07-30" 发布时间校准状态: "found" 发布时间需复核: "否" 发布时间来源: "llm:scrape:original_script_field" 发布时间证据: "datePublished: 2026-07-30T09:59:21.000Z" 发布时间校准原因: "文章结构化元数据中的datePublished字段明确标记为2026-07-30，优先级最高，且未被排除。" 发布时间校准置信度: "1" 发布时间候选数量: 23 发布时间严格候选数量: 9 发布时间原页读取状态: "source template page reused from URL open" 发布时间未找到原因: "" 发布时间校准时间: "2026-08-09T09:18:18+08:00" 发布时间仲裁状态: "confirmed" 发布时间仲裁尝试次数: 1 发布时间仲裁耗时毫秒: 4129 发现时间: "2026-08-09T09:16:28+08:00" 入库时间: "2026-08-09T01:18:23.186Z" 来源平台: "Data Center Knowledge 搜索" 搜索渠道: "source_template" 搜索词: "https://www.datacenterknowledge.com/search?q=Meta" 匹配关键词: - "GPU" - "Vera Rubin" - "deployment" - "performance" 相关厂家: - "Meta" - "Microsoft" - "NVIDIA" - "AMD" - "AWS" - "Google" - "Broadcom" - "OpenAI" 相关专家: [] 内容类型: "网页" 抓取工具: "CDP Render" 清洗工具: "CDP Text + Defuddle/Readability 正文提取" 原始附件: [] AI优质: "否" AI打分: 59 AI分档: "召回候选" AI质检状态: "不通过" AI打分理由: "文章讨论微软、谷歌、Meta的AI基础设施部署与投资，但缺乏超节点/AI Rack具体架构、技术细节或部件信息，仅泛泛涉及数据中心建设，技术细节不足。" AI质检模型: "deepseek-v4-flash" AI质检时间: "2026-08-09T22:21:08+08:00" AI主题相关性: 10 AI来源权威性: 13 AI新颖性: 15 AI技术细节: 2 AI商业部署信号: 10 AI完整性: 9 采集批次: "2026年8月9日8点13分13秒" 采集批次ID: "20260809-081313-600" 去重键: "https://www.datacenterknowledge.com/data-center-construction/hyperscalers-say-ai-race-has-entered-a-new-phase" --- Earnings calls shift the AI race from spending promises to energized campuses, power, and networking. Deployment pace, not GPU supply, is now the hyperscalers’ competitive edge. Microsoft's Fairwater data center in Mount Pleasant, Wisconsin.Image via Microsoft Microsoft, Alphabet, and Meta used their second-quarter earnings to signal a turning point in the AI infrastructure race. Rather than competing on how much they plan to spend, each focused on how quickly it could energize campuses, deploy networking at scale, secure power, and convert new infrastructure into revenue-generating compute. Sid Nag, CEO and chief research officer at Tekonyx, a research and advisory firm, said operators should watch deployment rather than procurement. “Data center operators should watch how quickly hyperscalers convert record AI capex into deployed capacity, because power availability, networking scale, and operational efficiency – not GPU supply – are emerging as the next competitive bottlenecks,” he said. ## Microsoft Turns Deployment into a Competitive Advantage Microsoft devoted much of its earnings call to physical deployment and time-to-serve. CEO Satya Nadella said the company opened 31 data centers during the quarter and 88 during fiscal 2026 across five continents, while adding roughly 1 GW of AI capacity. “We added nearly 1 GW of new capacity this quarter, opened 31 data centers, and reduced dock-to-live times by nearly 50%,” Nadella said, referring to the interval from hardware arrival at the dock to production service. The company said it remains on pace to roughly double total capacity over two years. Nadella added that Microsoft’s [Maia 200 accelerator](https://www.datacenterknowledge.com/infrastructure/microsoft-unveils-maia-200-in-house-inference-chip) supports both OpenAI and internal models while delivering roughly 30% better performance per dollar. Cobalt 200 racks are being deployed in more than 25 data centers. Future deployments are expected to incorporate [Nvidia Vera Rubin](https://www.datacenterknowledge.com/data-center-chips/gtc-2026-nvidia-unveils-vera-rubin-ai-platform-eyes-1t-by-2027) and [AMD Helios](https://www.datacenterknowledge.com/data-center-chips/amd-fires-back-at-nvidia-with-helios-ai-system-epyc-cpus) platforms. Chief Financial Officer Amy Hood said customer demand continues to outstrip available capacity despite record investment. She credited improvements in CPU utilization, GPU efficiency, and deployment processes with helping Azure monetize nearly every increment of new capacity as it comes online, underscoring how quickly AI workloads are absorbing new infrastructure. Microsoft also said roughly two-thirds of capital spending now goes toward shorter-lived assets such as CPUs and GPUs, giving the company flexibility to adjust hardware purchases as demand evolves. ## Alphabet Treats AI Infrastructure as a Long-Term Growth Engine Alphabet raised its 2026 capital expenditure outlook to between $195 billion and $205 billion after Google Cloud revenue rose 82% year over year. Executives also said infrastructure spending will rise again in 2027. About 60% of spending is going to servers, with the remainder focused on data centers and networking. Alphabet said it continues to supplement internal deployments with third-party infrastructure while expanding its own fleet, reflecting both surging customer demand and the realities of bringing new capacity online. The company noted its [Tensor Processing Units](https://www.datacenterknowledge.com/data-center-chips/google-launches-ironwood-tpu-for-next-gen-ai-inference) are generating commercial revenue as Google’s custom AI silicon reaches more customers. “Our AI investments are redefining what’s possible across every part of our business,” CEO Sundar Pichai said. “Google Cloud revenues accelerated to 82% growth, driven by demand for AI infrastructure and AI solutions.” Nag said the higher spending matters less than what it signaled about the company’s long-term strategy. “Alphabet’s biggest message wasn’t higher capex. Instead, it was confidence that AI infrastructure has shifted from a discretionary investment to the foundational operating layer for long-term revenue growth and competitive differentiation,” he said. Nag added that Alphabet is “executing one of the industry’s most disciplined AI infrastructure strategies,” scaling compute, networking, and custom silicon in lockstep “to create a durable competitive advantage rather than simply chasing model leadership.” ## Meta Plans for Persistent Scarcity in AI Capacity Meta framed its infrastructure strategy around scarcity rather than spending. Chief Financial Officer Susan Li said the company believes industry capacity will remain constrained. “The industry has underbuilt historically for the wave of AI adoption, making existing capacity, including our own, extremely valuable,” Li said, adding that industry capacity is expected “to remain tight for the foreseeable future.” Meta spent $31.1 billion on capital expenditures during the quarter, driven by investments in servers, data centers, and network infrastructure, and maintained full-year capex guidance of $130 billion to $145 billion. Li said Meta is building enough infrastructure to maximize capacity through 2027 while preserving flexibility beyond that horizon by prioritizing long-lived assets, including data centers, networking, land, and power. Asked about external demand, CEO Mark Zuckerberg said Meta has received “a lot of offers for compute at a significant premium over what we paid for it,” but indicated the company sees greater long-term value in using that capacity to power AI products, APIs, and business agents than in simply selling compute. Meta is financing its buildout with long-duration debt, infrastructure partnerships, and continued [investment in custom silicon](https://www.datacenterknowledge.com/infrastructure/meta-expands-broadcom-partnership-to-co-develop-custom-ai-silicon) to improve long-term flexibility and supply-chain leverage. ## What Comes Next Together, the three companies described different responses to the same challenge. Microsoft is compressing deployment timelines to bring capacity online faster. Alphabet is scaling compute, networking, and custom silicon as an integrated system while supplementing its own buildout with third-party infrastructure. Meta is locking in land, power, and financing before industry capacity tightens further. Attention now turns to Amazon. It is scheduled to report second-quarter earnings on Thursday, July 30. Beyond cloud revenue, investors will be looking for evidence that AWS is converting record infrastructure spending into energized AI capacity and whether CEO Andy Jassy identifies power availability, construction timelines, and grid access as the next constraints on AI expansion.