Anthropic、Google、Broadcom 合作，获取数吉瓦AI算力

Anthropic 与 Google 和 Broadcom 合作，获取数吉瓦下一代AI算力

加利福尼亚州旧金山 – 2026 年 4 月 7 日 – AI 安全与研究领域的领导者 Anthropic 今日宣布，大幅扩展其与科技巨头 Google 和 Broadcom 的合作伙伴关系。这项新协议承诺 Anthropic 将获得数吉瓦的下一代张量处理单元 (TPU) 算力容量，预计将于 2027 年开始投入使用。这项里程碑式的投资凸显了 Anthropic 积极扩展其计算基础设施的战略，旨在为前沿 Claude 模型提供动力，并满足其企业客户群不断飙升的全球需求。

此次合作标志着 Anthropic 的一个关键时刻，确保其拥有维持快速创新步伐所需的资源。Anthropic 首席财务官 Krishna Rao 表示：“与 Google 和 Broadcom 的这项开创性合作是我们扩展基础设施的纪律性方法的延续。我们正在做出迄今为止最重大的算力承诺，以跟上我们前所未有的增长步伐，使 Claude 能够定义 AI 发展的前沿。” 这一战略举措巩固了 Anthropic 在 AI 前沿的地位，为日益复杂和强大的模型提供了骨干支持。

前所未有的增长推动大规模基础设施投资

大幅扩展算力容量的决定直接反映了 Anthropic 爆炸性的商业成功和市场采纳。2026 年全年，Claude 模型的需求以惊人的速度增长。Anthropic 的年度经常性收入现已超过令人印象深刻的 300 亿美元，比 2025 年底的约 90 亿美元增长了三倍多。

在此显著增长的基础上，每年支出超过 100 万美元的商业客户数量在不到两个月内翻了一番，现已超过 1,000 家。这种企业参与度的快速扩张凸显了 Claude 在推动各行业商业价值和数字化转型中的关键作用。为了维持这一发展轨迹并继续有效地服务其不断增长的客户群，对核心基础设施进行如此重大的投资，不仅有利，而且绝对必要。

Anthropic 的增长里程碑

指标	2025 年底预估	2026 年初（当前）	增长系数（约）
年度经常性收入	约 90 亿美元	超过 300 亿美元	约 3.3 倍
年度支出超过 100 万美元的客户数	500+	超过 1,000	超过 2 倍

此表展示了 Anthropic 必须在 AI 算力方面进行重大投资的积极增长指标，使其在市场中保持持续领先地位。

加强美国AI基础设施和全球影响力

新获取的算力中有很大一部分将战略性地部署在美国境内。这与 Anthropic 致力于加强美国 AI 基础设施的更广泛承诺相符，并建立在其 2025 年 11 月承诺向国内计算能力投资 500 亿美元的基础上。通过建立强大的、以美国为中心的 AI 算力足迹，Anthropic 不仅提高了其运营安全性和效率，还为国家技术主权以及先进计算领域的经济增长做出了贡献。

此次合作也深化了 Anthropic 与 Google Cloud 现有的合作关系，这是去年十月宣布增加 TPU 算力容量的进一步发展。这种长期关系突显了双方在推动 AI 开发和部署方面的共同愿景。尽管很大一部分算力将部署在美国，但 Anthropic 致力于全球布局的承诺坚定不移，确保其模型全球客户均可访问。

战略性云和硬件多元化以增强韧性

Anthropic 在 AI 硬件和云部署方面采取了显著的多元化方法，旨在最大化性能、效率和韧性。该公司在包括 AWS Trainium、Google TPUs 和 NVIDIA GPUs 在内的一系列尖端 AI 硬件上训练和运行其 Claude 模型。这种多供应商、多架构的策略使 Anthropic 能够精心匹配特定 AI 工作负载到最适合的芯片，从而优化速度和成本效益。

“这种平台的多样性为依赖 Claude 进行关键工作的客户带来了更好的性能和更强的韧性，”该公司表示。这种分布式且适应性强的基础设施对于为所有人扩展 AI 至关重要，特别是对于需要坚定不移的可靠性和吞吐量的大型企业。此外，尽管 Amazon 仍然是 Anthropic 的主要云提供商和训练合作伙伴，并继续在 Project Rainier 项目上密切合作，但 Claude 仍是唯一一个在全球三大云平台：Amazon Web Services (Bedrock)、Google Cloud (Vertex AI) 和 Microsoft Azure (Foundry) 上均可供客户使用的前沿 AI 模型。这种广泛的云可用性，以及他们的亚马逊合作，为采用先进 AI 的企业提供了无与伦比的灵活性和选择。

Claude 的未来：企业级规模及更广阔前景

Google 和 Broadcom 注入的数吉瓦下一代 TPU 算力容量标志着 Anthropic 及其 Claude 模型的新时代。这种资源的大规模扩展不仅是为了满足当前需求；它旨在支持下一代 AI 能力，推动前沿模型所能达到的极限。对于企业而言，这意味着能够获得更强大、更可靠、更复杂的 AI 助手，能够以前所未有的准确性和安全性处理复杂任务。

随着 Anthropic 的持续快速增长，这项战略性的算力承诺确保其研发团队拥有探索 AI 新前沿所需的强大动力，从更高级的推理能力到增强的多模态理解。与 Google 和 Broadcom 等行业巨头的合作，使 Anthropic 能够巩固其领导地位，推动创新，在未来几年内造福其庞大的客户群和更广泛的 AI 生态系统。

常见问题

What is the primary objective of Anthropic's expanded partnership with Google and Broadcom regarding AI compute?

The core objective of Anthropic's newly announced partnership with Google and Broadcom is to secure a massive expansion of its AI compute infrastructure, specifically multiple gigawatts of next-generation Tensor Processing Unit (TPU) capacity. This significant commitment, set to come online starting in 2027, is designed to power the continued development and deployment of Anthropic's frontier Claude models. It directly addresses the exponential growth in demand from its global customer base, ensuring that Anthropic has the necessary computational resources to maintain its leadership in AI innovation and deliver high-performance, reliable services to enterprises adopting its advanced AI solutions.

How has Anthropic's customer base and revenue grown, influencing this infrastructure investment?

Anthropic has experienced an unprecedented surge in customer demand and financial growth, which directly necessitates this substantial infrastructure investment. As of early 2026, the company's run-rate revenue has soared past $30 billion, a significant leap from approximately $9 billion at the close of 2025. Furthermore, the number of business customers each spending over $1 million annually with Anthropic has doubled in less than two months, now exceeding 1,000. This rapid, sustained growth across its enterprise client base underscores the critical need for a massive expansion in compute power to both support existing users and enable future advancements of its Claude models.

What role does this partnership play in Anthropic's commitment to strengthening American AI infrastructure?

This partnership represents a major pillar in Anthropic's broader strategic commitment to strengthening American AI infrastructure. The vast majority of the newly secured compute capacity from Google and Broadcom will be sited within the United States. This move significantly expands upon Anthropic's previously announced $50 billion investment initiative from November 2025, which aims to bolster domestic computing capabilities. By housing such critical AI infrastructure on U.S. soil, Anthropic not only addresses its own operational needs but also contributes to national technological resilience and job creation within the rapidly evolving AI sector, fostering a robust domestic AI ecosystem.

How does Anthropic's multi-platform strategy enhance the performance and resilience of Claude models?

Anthropic employs a deliberate multi-platform strategy to enhance the performance and resilience of its Claude models. This involves training and running Claude across a diverse range of AI hardware, including AWS Trainium, Google TPUs, and NVIDIA GPUs. This approach allows Anthropic to strategically match specific workloads to the most suitable chips, optimizing for efficiency and specialized processing. By not relying on a single hardware vendor or cloud provider, Anthropic gains significant flexibility, reduces dependency risks, and ensures that customers receive optimal performance and increased operational resilience for their critical AI-driven tasks, a key factor for enterprise adoption and trust in frontier AI.

Which major cloud platforms offer Anthropic's Claude models to their customers?

Anthropic's Claude models are uniquely available across all three of the world's largest cloud platforms, providing unparalleled accessibility and choice for enterprise customers. Specifically, Claude is offered through Amazon Web Services (AWS) via its Bedrock service, on Google Cloud through Vertex AI, and within Microsoft Azure via its Foundry program. This broad availability ensures that organizations, regardless of their existing cloud infrastructure or strategic partnerships, can seamlessly integrate and leverage Anthropic's frontier AI capabilities, making Claude a highly versatile and accessible option in the competitive AI market.

What is the expected timeline for the new compute capacity to become operational?

The new compute capacity, secured through the expanded partnership with Google and Broadcom, is anticipated to become operational relatively soon. Anthropic expects this significant expansion of next-generation TPU capacity to come online starting in 2027. This phased rollout ensures a steady increase in computational power, strategically aligned with the projected growth in demand for Anthropic's Claude models and the continuous advancement of its AI research and development efforts, providing a clear roadmap for scaling its infrastructure to meet future challenges and opportunities in the AI landscape.