内容来源：https://aibusiness.com/generative-ai/google-anthropic-expand-partnership

内容总结：

人工智能初创企业Anthropic与谷歌达成价值数百亿美元的深度合作，计划部署百万颗谷歌自研TPU芯片，此举可能打破英伟达在AI硬件市场的长期垄断格局。行业分析师指出，该合作不仅为谷歌云服务锁定重要客户，更首次大规模验证其TPU芯片可作为英伟达GPU的替代方案。

此次合作反映出AI企业正通过与云服务商结盟确保算力供给。Anthropic采用多供应商策略，同时使用谷歌TPU、亚马逊Trainium和英伟达GPU，既保障千兆级算力需求，又避免单一供应商依赖。专家认为，这种多云合作模式将成为AI实验室新范式。

不过双方也面临挑战：Anthropic需适配不同芯片架构，谷歌则需在2026年前实现千兆级TPU产能承诺。同时，作为谷歌竞品的Claude模型获得优先算力支持，可能影响谷歌自研Gemini模型的训练资源，并引发监管机构对科技巨头与AI实验室深度绑定的关注。

中文翻译：

由谷歌云赞助
选择您的首个生成式AI应用场景
要着手应用生成式AI，首先应关注能够优化人类信息交互体验的领域。

这项合作可能削弱英伟达的芯片主导地位，同时帮助Anthropic避免供应商绑定风险。Anthropic与谷歌拓展合作伙伴关系，计划部署多达百万颗这家科技巨头的AI芯片，此举或将冲击英伟达在AI硬件市场的统治地位。

10月23日，Anthropic宣布将扩大采用谷歌张量处理单元（TPU）。该公司表示这份合作价值高达数百亿美元。

这是科技企业新一轮巨额AI投资中的最新举措，此前英伟达始终是主导者。10月15日，英伟达与微软、xAI及贝莱德达成400亿美元协议收购Aligned数据中心。一个月前，英伟达向Anthropic的竞争对手、ChatGPT创造者OpenAI投入千亿美元。

"这延续了AI公司与超大规模云服务商达成合作的模式，"Informa TechTarget旗下机构Omdia分析师托尔斯滕·沃尔克表示，"此类合作能确保可扩展性，让企业专注核心业务，而非过度深入数据中心领域。Anthropic、OpenAI等公司都与这些云服务商建立了合作，因为巨额基础设施投资与风投支持的初创企业模式并不相容。"

Anthropic与谷歌的结盟对长期统治AI硬件领域的英伟达构成挑战。Futurum Group分析师尼克·帕特恩斯指出，此举对谷歌具有双重利好：既锁定了数十亿美元的云服务收入流，也证明了其长期被低估的TPU技术的可靠性。某种程度上，这种强化联盟与微软和OpenAI的紧密关系类似，尽管OpenAI近期通过与英伟达等企业的合作已开始独立发展。

"谷歌获得了与微软-OpenAI联盟抗衡的标志性AI客户，更重要的是使其定制TPU作为英伟达主导性GPU的大规模替代方案获得了重要背书，"帕特恩斯强调，"此外，这也提升了谷歌早期对Anthropic少数股权投资的价值。"他补充说，挑战英伟达市场地位可能迫使这家AI芯片巨头调整定价策略。

沃尔克认为，该协议赋予Anthropic获取更优芯片价格的议价能力，使其能聚焦核心优势开发Claude等强大AI模型。"Anthropic需要借助此类合作专注核心竞争力，而非担忧数据中心基础设施的扩展问题。"

帕特恩斯指出，这一安排还展现了Anthropic正在构建多芯片供应商战略。"通过拒绝绑定单一供应商，他们正基于谷歌TPU、亚马逊Trainium芯片和英伟达GPU构建多元化基础架构。"他补充说，与多家供应商建立合作使Anthropic在确保千兆瓦级算力的同时具备价格谈判优势，这对训练下一代AI模型至关重要。

"这为AI实验室确立了全新的多云战略范本，"帕特恩斯继续阐述，"展示了如何利用谷歌、亚马逊和微软之间的竞争来维持议价能力，规避供应商锁定的关键风险。"不过他也提醒，双方都面临挑战：Anthropic需在不同架构间优化模型，而谷歌则需要兑现承诺。

"谷歌必须在2026年前实现新增超千兆瓦TPU算力的重大基础设施承诺，这个时限相当紧迫，"他分析道，"这可能引发监管机构对少数云巨头与顶级AI实验室之间日益紧密的金融技术关联进行更严格审查。"

沃尔克同时指出，Anthropic的Claude与谷歌Gemini生成式AI模型平台存在竞争关系。"此协议让谷歌能观察Claude在大规模生产环境中的表现，"他解释道，"根据使用情况，谷歌将能推断Claude的收益模式、性能特征和定价策略。"此外他补充说，Anthropic通过支付费用获得谷歌TPU优先使用权，这可能影响谷歌Gemini模型的训练和推理能力。

您可能还喜欢

英文来源：

Sponsored by Google Cloud
Choosing Your First Generative AI Use Cases
To get started with generative AI, first focus on areas that can improve human experiences with information.
The deal could erode Nvidia's chip dominance. It also helps Anthropic avoid vendor lock-in.
Anthropic and Google's expanded partnership, in which the independent generative AI vendor plans to deploy up to a million of the tech giant's AI chips, looks to threaten Nvidia dominance of the AI hardware market.
Anthropic on Oct. 23 revealed plans to expand its use of Google's tensor processing units (TPUs).
Anthropic said the partnership is worth tens of billions of dollars.
The deal is the latest in a round of tech vendors spending huge sums on AI, with Nvidia the main player up to now. On Oct. 15, Nvidia, Microsoft, xAI and BlackRock agreed on a $40 billion deal to buy Aligned Data Centers. A month earlier, Nvidia invested $100 billion in Anthropic rival and ChatGPT creator OpenAI.
"This just continues the pattern where AI companies strike deals with hyperscalers to ensure scalability and the ability to focus on their core business, instead of getting too deep into the data center business," said Torsten Volk, an analyst at Omdia, a division of Informa TechTarget. "Anthropic, OpenAI and friends all have deals in place with these hyperscalers, as massive infrastructure investments do not mix well with VC-funded startups."
The Anthropic-Google deal challenges Nvidia, which has dominated the AI hardware arena for years.
Anthropic's expanded partnership is beneficial for Google because it not only secures a multi-billion-dollar cloud revenue stream but also proves the credibility of its long-underestimated TPU technology, said Nick Patience, an analyst at the Futurum Group. In some ways, the strengthened alliance parallels the close relationship that has existed between Microsoft and OpenAI, although OpenAI has branched out on its own in recent months, with the Nvidia deal and others.
"It gains a banner AI tenant to rival Microsoft's OpenAI partnership and crucially, achieves a massive public validation of its custom TPUs as a credible, large-scale alternative to Nvidia's dominant GPUs," Patience said. "Plus, it gives a boost to Google's early minority investment in Anthropic."
Challenging Nvidia's market dominance could force the AI chip giant to lower its pricing, Patience added.
The deal gives Anthropic more leverage for sourcing chips at better prices and enables it to focus on its strengths, building powerful AI models such as Claude, Volk said.
“Anthropic needs these deals to be able to focus on its core competencies, instead of worrying about scaling data center infrastructure,” Volk said.
The arrangement also shows how Anthropic is opting not to work with just one chip vendor and instead has built a multi-chip strategy, Patience said.
"By refusing to lock in with a single vendor, it is building a diversified foundation using Google's TPUs, Amazon's Trainium chips and Nvidia's GPUs," he said.
He added that by forming partnerships with all these vendors, Anthropic can negotiate prices while securing gigawatt-level compute power, which is critical for training the next generation of AI models.
"It establishes the new multi-cloud playbook for AI labs, demonstrating how to use competition between Google, Amazon, and Microsoft to maintain negotiating use and avoid the critical risk of vendor lock-in," Patience continued.
However, both vendors face some challenges. Anthropic will now have to optimize its models across different architectures, while Google will need to execute, Patience added.
"It must now deliver on the monumental infrastructure promise of bringing over a gigawatt of new TPU capacity online by 2026, which is obviously pretty soon," he said. "It potentially also brings heightened regulatory scrutiny over the deepening financial and technical ties between a handful of hyperscalers and the few top-tier AI labs."
Another challenge is that Anthropic Claude competes with Google's Gemini generative AI line of models and platforms, Volk noted.
"This deal allows Google to observe how Claude behaves in massive-scale production scenarios," he said, adding that based on usage, "Google will be able to deduce Claude's revenue patterns, performance characteristics and pricing strategies."
Meanwhile, Anthropic is paying for priority access to Google's TPUs, which might interfere with Google's Gemini training and inferencing capacity, he added.
You May Also Like