调查显示OpenAI智能体平台企业使用率为Anthropic两倍

亿邦AI 2026-10-14 10:06

亿邦AI 2026/10/14 10:06

邦小白快读

全文速览

总1：核心数据速览

1. OpenAI在智能体平台使用率43%，是Anthropic（22%）的两倍；首选份额OpenAI25%，谷歌18%，Anthropic16%。

2. 生产环境中，OpenAI模型接入41%，Gemini39%，Anthropic15%；云服务谷歌云51%、Azure36%、AWS23%。

总2：实操干货

1. 多供应商并行是主流，如家得宝用谷歌生态存数据，用Azure调OpenAI校验，Claude写代码，OpenAI审核。

2. 选型时谨防统计口径差异，Ramp数据显示Anthropic付费企业比例（43.8%）高于OpenAI（39.8%），与大型企业调查结论不同。

3. 模型替换风险高：62%企业换过模型，59%性能下降，75%需几天工程量；切换需重新训练上下文，形成粘性，初期选型要留足迁移余地。

总1：品牌竞争格局

1. 智能体平台使用率OpenAI43% vs Anthropic22%，首选OpenAI25% vs 谷歌18% vs Anthropic16%；OpenAI品牌暂居上风。

2. 未来12个月采购意向Anthropic33%超过OpenAI30%和谷歌25%，且计划采购Anthropic的企业中60%为新客户，品牌势能可能逆转。

总2：品牌渠道与用户行为

1. 云渠道是重要通路：谷歌云覆盖51%，Azure36%，AWS23%；68%企业用谷歌云或Gemini，62%用Azure或OpenAI，模型品牌与云渠道强绑定。

2. 家得宝案例展示了品牌并行：重度谷歌生态，同时用Anthropic、OpenAI补位，说明企业客户喜欢混合采购而非单一品牌。

3. 用户粘性来自切换成本：每次切换模型需重新上下文训练，62%企业换过模型，59%性能下降。品牌商需在易迁移和深度绑定之间平衡。

总1：增长市场与需求变化

1. 智能体平台当前OpenAI43%、Anthropic22%、谷歌21%；未来计划采购Anthropic33%、OpenAI30%、谷歌25%，新采购需求集中在Anthropic。

2. 生产环境云侧谷歌云51%、Azure36%、AWS23%；模型OpenAI41%、Gemini39%、Anthropic15%。多供应商并行成主流，家得宝案例说明了混合架构的可行性。

总2：机会与风险

1. 模型替换服务是机会：62%企业换过模型，59%遭遇性能下降，75%需要数天工程量，围绕模型迁移、校验、集成可形成业务。

2. 风险提示：不同统计口径结果差异大，Ramp样本显示Anthropic付费比例更高，需警惕样本偏差；智能体调查样本较小，决策不要只看单一数据。

3. 可学习点：关注Anthropic IPO进度的企业，提前布局与Anthropic的合作代理、集成方案；同时可提供开源框架（如LangChain）方案。

总1：生产环境技术选型

1. 云服务谷歌云51%领先，Azure36%、AWS23%；大模型OpenAI41%、Gemini39%、Anthropic15%。

2. 检索工具使用率：OpenAI检索引擎58%、Vertex AI搜索50%、独立工具（Pinecone/Elasticsearch）12%-14%，合计40%至少用一款独立工具，可用于工厂知识库和数据检索。

总2：数字化启示

1. 家得宝案例：重度用谷歌BigQuery和ADK，再通过Azure调OpenAI校验、用Claude写代码、OpenAI审核，说明可用多供应商组合满足生产需求。

2. 风险控制：模型替换并非低成本，62%企业换过模型，59%性能下降，75%需数天工程；工厂转型应先搭建稳定平台，避免频繁更换模型造成停产风险。

3. 商业机会：模型切换需要重新上下文训练，形成供应商粘性；若工厂自研AI应用，可与多云服务商合作，降低被锁定风险。

总1：行业趋势

1. 企业AI平台三强格局：OpenAI智能体平台使用率43%、Anthropic22%、谷歌21%；主平台首选OpenAI25%、谷歌18%、Anthropic16%。

2. 未来需求向Anthropic偏移：33%计划更换或新增平台的企业考虑Anthropic，其中60%尚未使用过。

总2：客户痛点与解决方案

1. 痛点：模型替换频繁且风险高，62%企业换过模型，59%性能下降，75%工程量大；咨询师客户81%换过模型，服务商应重视迁移评估与返工支持。

2. 解决方案：多供应商并行。例如家得宝混合使用谷歌、Azure OpenAI、Anthropic；检索层可用OpenAI检索工具（58%）与Vertex AI搜索（50%），独立工具仍有40%渗透率，可提供统一检索抽象层。

3. 数据参考：云侧谷歌云51%、Azure36%、AWS23%，模型OpenAI41%、Gemini39%、Anthropic15%，但Ramp样本显示Anthropic付费比例43.8%高于OpenAI39.8%，做方案需明确客户群体。

总1：平台市场结构与客户需求

1. 平台使用率OpenAI43%、Anthropic22%、谷歌21%；首选OpenAI25%、谷歌18%、Anthropic16%。云侧谷歌云51%、Azure36%、AWS23%，模型接入OpenAI41%、Gemini39%、Anthropic15%。

2. 客户普遍多供应商并行，家得宝案例说明平台商不能指望客户单一绑定；62%企业换过模型，平台商应提供平滑切换、多模型管理功能以响应需求。

总2：运营与风控

1. 云渠道统治力：68%企业用谷歌云或Gemini，62%用微软Azure或OpenAI，平台商可与云厂商深度整合扩大生态。

2. 风险：统计口径差异大，Ramp数据显示Anthropic付费比例更高，与大型企业调查相反；未来Anthropic采购意向强（33%考虑，60%新客户），需防范客户向Anthropic平台迁移。

3. 建议：关注Anthropic IPO进程，其资本化可能加速市场洗牌；平台商可提前布局与之对接或提供跨平台迁移工具。

总1：产业新动向

1. 智能体平台OpenAI使用率43%对Anthropic22%，但未来采购意向Anthropic33%反超OpenAI30%，且60%计划采购者为新客户；模型更换率升至每月8%历史高位，过去一年62%企业换过模型。

2. 新问题：模型替换带来59%性能回退、75%高工程量；咨询师报告客户换模型比例81%，企业内部仅51%，受访者偏差值得关注。统计口径方面，OpenAI年化营收500亿与700亿差异源于云伙伴销售，Ramp与大型企业调查结果相反，对产业统计方法提出挑战。

总2：商业模式与政策启示

1. 云与模型深度捆绑，68%企业用谷歌云或Gemini，62%用微软Azure或OpenAI；家得宝混合使用谷歌、OpenAI、Anthropic是典型，Anthropic递交IPO并举办投资者日显示资本加速。

2. 样本量仅109/102/113，建议建立标准化统计体系；模型切换成本高企影响竞争，政策可关注互操作性和数据可迁移性。

返回默认

声明：快读内容全程由AI生成，请注意甄别信息。如您发现问题，请发送邮件至 run@ebrun.com 。

我是品牌商卖家工厂服务商平台商研究者帮我再读一遍。

Quick Summary

Overview 1: Key data at a glance

1. OpenAI leads agent-platform adoption at 43%, double Anthropic's 22%; for primary platform preference, OpenAI is at 25%, Google 18%, and Anthropic 16%.

2. In production environments, OpenAI models are integrated by 41% of enterprises, Gemini 39%, and Anthropic 15%; on the cloud side, Google Cloud is used by 51%, Azure 36%, and AWS 23%.

Overview 2: Practical takeaways

1. Multi-vendor parallelism is mainstream. Home Depot, for example, stores data in the Google ecosystem, calls OpenAI through Azure for validation, uses Claude to write code, and uses OpenAI for review.

2. Be careful about differences in measurement methodology when evaluating vendors. Ramp's data shows Anthropic's share of paying enterprise customers (43.8%) is higher than OpenAI's (39.8%), the opposite of the larger-enterprise survey.

3. Model switching is risky: 62% of enterprises have switched models, 59% saw performance decline, and 75% needed days of engineering work. Switching requires retraining on context, creating vendor stickiness, so early choices should leave enough room for migration.

Overview 1: Brand competitive landscape

1. On agent-platform adoption, OpenAI leads with 43% vs. Anthropic's 22%; for primary platform preference, OpenAI is 25%, Google 18%, and Anthropic 16%. The OpenAI brand is ahead for now.

2. Over the next 12 months, purchase intent favors Anthropic, at 33%, over OpenAI at 30% and Google at 25%. Moreover, 60% of enterprises planning to buy Anthropic are new customers, so brand momentum could reverse.

Overview 2: Brand channels and user behavior

1. Cloud channels are a key route to customers: Google Cloud reaches 51% of enterprises, Azure 36%, and AWS 23%; 68% use Google Cloud or Gemini, and 62% use Azure or OpenAI. Model brand and cloud channel are strongly tied.

2. The Home Depot case shows brand parallelism: heavy use of the Google ecosystem, supplemented by Anthropic and OpenAI, indicates that enterprise customers prefer mixed procurement over a single brand.

3. User stickiness comes from switching costs: every model switch requires retraining context; 62% of enterprises have switched models and 59% saw performance decline. Brand vendors need to balance ease of migration against deep customer lock-in.

Overview 1: Growth markets and demand shifts

1. Current agent-platform adoption is OpenAI 43%, Anthropic 22%, and Google 21%; future purchase plans point to Anthropic 33%, OpenAI 30%, and Google 25%. New procurement demand is concentrated on Anthropic.

2. In production, cloud usage is Google Cloud 51%, Azure 36%, and AWS 23%; model usage is OpenAI 41%, Gemini 39%, and Anthropic 15%. Multi-vendor parallelism is becoming the norm, and the Home Depot case proves the viability of hybrid architectures.

Overview 2: Opportunities and risks

1. Model-replacement services are an opportunity: 62% of enterprises have switched models, 59% saw performance degradation, and 75% needed days of engineering. This creates a business around model migration, validation, and integration.

2. Risk note: different measurement methods produce very different results. Ramp's sample shows a higher paid-adoption rate for Anthropic, so watch for sample bias; the agent survey has a small sample, so decisions should not rely on one data point.

3. Actionable insight: enterprises watching Anthropic's IPO progress should proactively build partner and integration offerings around Anthropic early; also consider offering open-source frameworks such as LangChain.

Overview 1: Production-environment technology selection

1. On the cloud side, Google Cloud leads at 51%, followed by Azure 36% and AWS 23%; for large models, OpenAI is used by 41%, Gemini 39%, and Anthropic 15%.

2. Retrieval tool adoption: OpenAI's retrieval engine 58%, Vertex AI Search 50%, and standalone tools such as Pinecone or Elasticsearch 12%–14% individually; combined, 40% use at least one standalone tool. These are all relevant for factory knowledge bases and data retrieval.

Overview 2: Digital-transformation lessons

1. Home Depot's approach — heavy use of Google BigQuery and ADK, then calling OpenAI through Azure for validation, using Claude for code, and OpenAI for review — shows that a multi-vendor combination can meet production needs.

2. Risk control: model replacement is not low-cost. 62% of enterprises have switched models, 59% saw performance decline, and 75% needed days of engineering. Factories should first build a stable platform to avoid downtime from frequent model changes.

3. Business opportunity: switching models requires retraining context, creating supplier stickiness. If a factory develops its own AI applications, it can work with multiple cloud providers to reduce the risk of lock-in.

Overview 1: Industry trends

1. The enterprise AI platform landscape is basically three players: OpenAI has 43% agent-platform adoption, Anthropic 22%, and Google 21%; for primary platform preference, OpenAI is 25%, Google 18%, and Anthropic 16%.

2. Future demand is shifting toward Anthropic: 33% of enterprises planning to switch or add a platform are considering Anthropic, and 60% of those have never used it.

Overview 2: Client pain points and solutions

1. Pain point: model replacement is frequent and risky — 62% of enterprises have switched models, 59% saw performance decline, and 75% faced heavy engineering. For consultant-served clients, the switching rate is 81%, so service providers should emphasize migration assessment and rework support.

2. Solution: multi-vendor parallelism. Home Depot, for example, mixes Google, Azure OpenAI, and Anthropic. For the retrieval layer, OpenAI's retrieval tool (58%) and Vertex AI Search (50%) are common, while standalone tools still have 40% penetration; a unified retrieval abstraction layer can be offered.

3. Data reference: cloud usage is Google Cloud 51%, Azure 36%, AWS 23%; model usage is OpenAI 41%, Gemini 39%, Anthropic 15%. However, Ramp's sample shows Anthropic's paid-customer ratio at 43.8% vs. OpenAI's 39.8%, so solution design must be tailored to the client segment.

Overview 1: Platform market structure and customer demand

1. Platform adoption is OpenAI 43%, Anthropic 22%, Google 21%; primary preference is OpenAI 25%, Google 18%, Anthropic 16%. Cloud-side usage is Google Cloud 51%, Azure 36%, AWS 23%; model integration is OpenAI 41%, Gemini 39%, Anthropic 15%.

2. Customers generally run multiple vendors in parallel. The Home Depot case shows that platform providers cannot expect single-vendor lock-in; with 62% of enterprises having switched models, platforms should offer smooth switching tools and multi-model management.

Overview 2: Operations and risk management

1. Cloud channels dominate: 68% of enterprises use Google Cloud or Gemini and 62% use Microsoft Azure or OpenAI. Platform providers should invest in deep integration with cloud vendors to expand their ecosystems.

2. Risk: statistical methods vary. Ramp's data shows higher paid adoption for Anthropic than the larger-enterprise survey suggests; future purchase intent for Anthropic is strong (33% considering it, 60% new customers), so platforms need to guard against migration to Anthropic.

3. Recommendation: monitor Anthropic's IPO process; its capitalization could accelerate market reshuffling. Platform providers should prepare integration paths or cross-platform migration tools in advance.

Overview 1: New industry dynamics

1. OpenAI's agent-platform adoption is 43% vs. Anthropic's 22%, but future purchase intent is higher for Anthropic at 33% vs. OpenAI's 30%, with 60% of planned buyers being new customers. The model-switching rate has risen to a historical high of 8% per month, and 62% of enterprises switched models in the past year.

2. New problems: model replacement leads to 59% performance regression and 75% high engineering workload. Consultants report that 81% of their clients have switched models, while internal enterprise data says 51% — respondent bias is worth attention. On measurement, OpenAI's annualized revenue figures of $50 billion and $70 billion differ because of cloud-partner sales, while Ramp and large-enterprise survey results are also opposite, challenging industry statistical methods.

Overview 2: Business model and policy implications

1. Cloud and models are deeply bundled: 68% of enterprises use Google Cloud or Gemini, 62% use Microsoft Azure or OpenAI. Home Depot's combined use of Google, OpenAI, and Anthropic is a typical example; Anthropic's IPO filing and investor day signal accelerating capital activity.

2. Sample sizes are only 109/102/113, so a standardized statistical system is advisable. High model-switching costs affect competition, and policy should focus on interoperability and data portability.

Disclaimer: The "Quick Summary" content is entirely generated by AI. Please exercise discretion when interpreting the information. For issues or corrections, please email run@ebrun.com .

I am a Brand Seller Factory Service Provider Marketplace Seller Researcher Read it again.

2026年9月发布的多份AI产业专项调查数据，覆盖企业级AI应用的平台选型、基础设施、模型替换等多个维度。智能体编排平台是为AI智能体分配任务、调用工具、约束操作权限的核心软件，在使用这类平台的受访企业中，43%接入OpenAI的Agents SDK与Responses API，22%使用Anthropic的Claude Agent SDK及Managed Agents，21%选择谷歌的Enterprise Agent Platform。

在主平台选择维度，25%的受访企业将OpenAI相关产品列为首选，18%选择谷歌产品，16%选择Anthropic产品，剩余企业采用Salesforce Agentforce、微软Copilot Studio、亚马逊Bedrock，以及LangChain等开源框架。

同月份开展的AI基础设施与算力调查覆盖生产环境技术选型。云服务领域，51%的受访企业使用谷歌云，36%采用微软Azure，23%选择亚马逊云服务。大模型层面，41%的企业接入OpenAI模型，39%使用谷歌Gemini模型，15%采购Anthropic模型。统计生产环境核心供应商时，32%的企业选择谷歌云或Gemini，22%选择OpenAI，仅6%选择Anthropic。该统计存在口径差异，部分通过微软Azure调用OpenAI模型、通过亚马逊或谷歌云调用Anthropic Claude模型的企业，可能直接填报云服务商而非模型厂商。若叠加云服务与自有模型业务，68%的受访企业使用谷歌云或Gemini，62%使用微软Azure或OpenAI。

家居零售商家得宝是多供应商并行的典型案例。家得宝执行副总裁兼CIO Angie Brown在近期录制的行业播客采访中提到，公司为重度谷歌生态用户，绝大多数数据存放于谷歌BigQuery数据仓库，内部部署大量谷歌智能体开发套件ADK。团队同时通过微软Azure调用OpenAI模型校验谷歌模型的输出结果，用Claude编写代码、用OpenAI模型完成代码审核，并非单一绑定某一家供应商。

不同统计口径下，两家头部大模型厂商的市场表现存在差异。企业支出管理平台Ramp9月9日披露的数据显示，其平台内43.8%的企业为Anthropic的订阅服务或token付费，39.8%为OpenAI付费。Ramp的样本仅覆盖使用其企业卡及软件的客户，这类企业普遍比平均水平更具科技属性，且统计时无论人员规模均按单家企业计数，中小微企业占比更高因此权重更大，与前述面向100人以上组织及相关服务咨询师的调查样本、统计维度均存在区别。企业完全可能同时采购Anthropic的订阅和API用于编码工具，同时在OpenAI平台上运行智能体。

营收相关数据同样受统计口径影响。当前OpenAI年化营收约为500亿美元，低于9月广泛传播的700亿美元数值，后者将其通过云合作伙伴产生的销售额纳入统计，与Anthropic的营收统计口径一致。10月8日该更低营收数据披露后，美股科技板块出现下跌。Anthropic于今年6月秘密提交IPO申请，市场公开信息显示，其将于本周三在旧金山总部举办IPO前投资者日，邀请部分机构投资者参与。

未来12个月的企业采购意向出现变化。在计划新增、更换智能体编排平台的受访企业中，33%将Anthropic平台纳入考量范围，30%考虑OpenAI，25%考虑谷歌，其中计划采购Anthropic平台的企业里有60%目前尚未使用其产品。

企业更换AI模型的频次正在升高。Ramp10月6日披露的数据显示，截至9月的三个月里，平台内企业平均每月有8%更换AI模型供应商，比例为该数据统计以来最高。前述智能体编排调查同步统计，62%的受访企业过去一年曾在保留外层应用的前提下，替换生产应用内的AI模型，涵盖切换至其他厂商模型、升级同厂商新版本模型两类场景。这些企业中59%遇到性能下降问题，需要返工调整应用或换回原模型，仅19%的企业称替换几乎不需要工程开发工作，75%的企业反馈替换需要至少数天的工程工作量，甚至最终中途放弃。家得宝方面在采访中提到，每次模型切换都需要重新完成上下文训练，这形成了一定的用户粘性。不同受访群体反馈的替换比例存在差异，81%的咨询师称其主要客户过去一年有过模型替换行为，企业内部受访者中该比例为51%，推高了整体统计数值。

第三份智能体上下文层调查显示，在生产环境运行检索功能的企业中，58%使用OpenAI的检索工具，50%使用谷歌Vertex AI搜索，独立检索工具Pinecone和Elasticsearch的使用率分别为12%到14%，总计40%的企业至少使用一款独立检索工具。

本次引用的三份9月调查分别回收109份、102份、113份有效问卷，受访者均为100人以上规模组织的员工或主要服务这类组织的咨询师。月度调查样本来自行业读者群体及专业研究panel，结果仅反映受访群体情况，其中OpenAI在智能体编排平台使用率上的领先达到95%统计置信度，计划采购Anthropic平台的受访企业样本量为20家。10月的同主题调查将沿用相同问卷，追踪相关数据的变化。

本文首发于亿邦动力官方网站

文章来源：亿邦动力

FAQ回顾

企业级AI智能体平台有哪些主流选择？

据2026年9月企业AI专项调查，智能体编排平台使用率中，OpenAI的Agents SDK与Responses API为43%，Anthropic的Claude Agent SDK为22%，谷歌Enterprise Agent Platform为21%。另有Salesforce Agentforce、微软Copilot Studio、亚马逊Bedrock及LangChain等开源框架可供选择。

OpenAI和Anthropic的智能体平台在企业市场表现有什么差距？

从使用率看，OpenAI为43%，Anthropic为22%，OpenAI约为Anthropic的两倍。主平台首选方面，25%企业选OpenAI，16%选Anthropic。但在未来12个月计划更换平台的考量中，33%考虑Anthropic，略高于OpenAI的30%，且其中60%目前尚未使用Anthropic产品。

企业更换AI模型时会遇到哪些问题？

62%的受访企业过去一年曾替换生产应用内AI模型，其中59%遇到性能下降，需返工或换回原模型；75%反馈替换需要至少数天工程工作量，部分最终中途放弃。家得宝CIO提到，每次模型切换都需重新完成上下文训练，形成用户粘性。

AI厂商市场份额统计口径差异是如何产生的？

差异源于部分企业通过云服务商调用模型，如经微软Azure使用OpenAI、经亚马逊或谷歌云使用Claude，可能被统计为云厂商而非模型厂商。Ramp数据显示43.8%企业付费给Anthropic、39.8%给OpenAI，但样本以中小科技企业为主，与面向100人以上组织的调查口径不同。