Many enterprises are off to the races with agentic AI, snapping up technology in areas such as software development, customer support, IT operations and cybersecurity. But the return on this type of AI investment remains murky for some: A very sizable 94% of respondents said their organizations lacked a consistent, enterprise-wide framework for evaluating AI’s ROI.如今各大企业皆在 AI 代理（Agentic AI）的江湖中各显神通，纷纷重金布局软件开发、客户支持、IT 运维及网络安全等要塞。然而，这番投入究竟能换回多少真金白银，对不少人而言仍如雾里看花。调查中，竟有九成四的受访者坦言，自家企业尚无一套行之有效的章法来衡量 AI 的投资回报率（ROI）。

What’s more, only 16% of respondents said they see a positive ROI on more than half their AI projects, while 31% see ROI on less than a quarter.更令人深思的是，仅有一成六的受访者能在过半的 AI 项目中见到正向回报，而有三成一的人，其项目回报甚至不足四分之一。

Those were some of the key takeaways from our just-published Battery Ventures State of Enterprise Tech Spending survey, which includes responses from 100 senior technology leaders representing more than $66 billion in annual technology spend. Overall, the survey still indicated a robust appetite from enterprises to continue spending on AI. Trends in what types of technology these leaders are evaluating, testing and buying—including new AI tools and platforms—provide key insights for tech startups trying to boost sales and scale their companies.以上洞见，皆出自 Battery Ventures 最新发布的《企业技术支出状况调查》。此番调研汇集了百位科技界顶尖高手的真知灼见，其背后掌管的年度技术支出逾 660 亿美元。总而言之，企业对 AI 的胃口依然极好，投入热情未减。这些决策者在评估、测试及采买 AI 工具与平台时的动向，正是科技初创企业洞察先机、谋求扩张的关键指南。

Indeed, enterprise spending on new technologies, notably AI, is brisk today: Seventy-six percent of respondents said they’re already in production with generative AI, with 100% planning to deploy AI programs within two years. Not a single respondent said they were cutting AI spending. An impressive 81% plan to increase overall tech spending over the next 12 months, up from 65% in December 2025.事实上，企业在 AI 等前沿技术上的支出正如火如荼：七成六的受访者已将生成式 AI 投入生产环境，且人人皆有在两年内全面部署的打算。竟无一人打算缩减 AI 开支。更难得的是，八成一的受访者计划在未来十二个月内增加整体技术预算，此比例较 2025 年 12 月的六成五有了显著提升。

In addition, 49% of respondents said they’re actively deploying agentic AI, up from 33% six months ago. Conviction in this type of tech remains high among buyers, with 50% of respondents saying they’re already scaling agentic workflows across business functions.此外，四成九的受访者已在积极部署 AI 代理，较半年前的三成三增长迅速。买家对此类技术的信心坚如磐石，半数受访者表示已将 AI 代理的工作流推广至各项业务职能之中。

But the sobering news is that many of these buyers are still struggling with AI ROI, at least at the enterprise level. Of those who said they still don’t have an enterprise-wide, AI ROI framework, 42% said they were measuring ROI inconsistently across the organization; 43% were still defining their measurement approach; and nine percent reported they still have no way to measure ROI yet.然而，令人清醒的是，许多买家在评估 AI 回报时仍深陷泥潭。在那些缺乏统一 ROI 衡量框架的企业中，四成二的人测量方式杂乱无章，四成三的人尚在摸索门径，更有九成的人坦言至今找不到衡量 ROI 的法门。

Still, 53% of enterprises saw a clear ROI on AI in general terms, with the industries most convinced of AI’s financial and operational impact being IT and healthcare.即便如此，仍有五成三的企业在 AI 整体应用上看到了清晰的回报，其中尤以 IT 与医疗行业对 AI 带来的财务与运营效能最为笃信。

Other findings:另有若干发现如下：

• Humans remain firmly in the AI loop today. Of our respondents, 70% said humans were overseeing or making final decisions on their AI projects; only five percent said they expect fully autonomous systems to be deployed in their enterprises in the next 12 months.当下，AI 的权柄仍牢牢掌握在人手中。七成受访者表示，AI 项目的最终裁决权与监督权皆由人把关；仅有五成的人预期在未来一年内，企业能实现真正的全自动运行。
• Not all spending on AI is net new. Seventy-eight percent of the survey’s respondents said they are funding AI at least partly by reallocating existing tech budgets. Forty-six percent said they blend net-new spending with reallocation.AI 的资金来源并非全是平地起高楼。七成八的受访者表示，其 AI 预算至少部分源于对原有技术预算的重新调配。四成六的人则采取了新旧资金混合的策略。
• With AI technology changing so rapidly and enterprises still experimenting with use cases, enterprise contract lengths have shortened significantly. Only 29% of our respondents said they were signing tech contracts of 25 months or longer. Historically, 64% of organizations signed contracts of 25 months or .鉴于 AI 技术一日千里，企业尚处于摸索阶段，技术合同的期限亦随之大幅缩短。仅有二成九的受访者愿意签署 25 个月以上的长约。而在往昔，长约的比例高达六成四。
• Still, enterprises seem to believe AI represents a core, long-term, structural shift in their operations, particularly in software development. In the survey, 84% said they expect AI to materially reshape their software-delivery cycle within two years. This is one factor driving increasing technology budgets overall.尽管如此，企业仍视 AI 为一场深刻且长久的结构性变革，尤其是在软件开发领域。八成四的受访者预计，AI 将在两年内彻底重塑其软件交付周期。这正是驱动整体技术预算水涨船高的核心动力。
• And AI has risen to the top of our respondents’ list of their top five enterprise priorities. Right now, those priorities are generative AI; agentic AI; cloud infrastructure; data warehousing; and cybersecurity.AI 已然跃居受访者五大优先事项之首。目前，这五大要务依次为：生成式 AI、AI 代理、云基础设施、数据仓库，以及网络安全。
• Finally, we saw some emerging trends in which LLMs enterprises are increasingly turning to. A strong 75% of respondents ranked Claude, from Anthropic, among their top three model choices, ahead of Microsoft (72%) and OpenAI (58%). Still, Microsoft’s Azure OpenAI/Copilot product remains the single most-used model.最后，我们观察到企业在 LLM（大语言模型）选择上的新趋势。七成五的受访者将 Anthropic 旗下的 Claude 列为首选三甲，竟压过了微软（七成二）与 OpenAI（五成八）。不过，微软的 Azure OpenAI/Copilot 依然是目前市占率最高的模型。
• Code generation and testing are the top engineering re-evaluations by a wide margin: 82% of enterprises are re-evaluating code generation and refactoring, and 75% are re-evaluating testing and QA automation. These are the two highest numbers in the entire re-evaluation dataset. Code security and vulnerability detection is at 70%. Together these three paint a clear picture of where AI is restructuring the engineering stack and where opportunities for AI startups exist.在工程重构方面，代码生成与测试遥遥领先：八成二的企业正重新审视代码生成与重构，七成五的企业则在重构测试与 QA 自动化。这两项指标在所有数据中高居榜首，代码安全与漏洞检测亦占七成。三者合力，勾勒出 AI 重构工程栈的版图，亦指明了初创企业的机遇所在。
• Security review is the biggest breaking point in the AI SDLC, according to the survey. When asked to name the biggest SDLC breaking points as AI agents become more common in the delivery pipeline, respondents named security review and vulnerability management as their top concerns, followed by compliance and auditability and developer skill degradation.调查显示，安全审查已成为 AI 软件开发生命周期（SDLC）中最大的关隘。当问及 AI 代理普及后最大的交付痛点时，受访者首推安全审查与漏洞管理，随后是合规审计与开发者技能退化等隐忧。
• AI governance is moving to the C-suite, but ownership is still fragmented. Twenty-five percent of enterprises now have a “chief AI officer”, up from 23% in Q3 2025. But 54% still have AI sitting under existing tech leadership with no dedicated owner, and seven percent have no formal AI leadership role at all.AI 治理虽已进入决策层视野，但权责依然分散。二成五的企业现已设立“首席 AI 官”，较 2025 年第三季度的二成三略有增长。然而，五成四的企业仍将 AI 挂靠在现有技术领导下，缺乏专人统筹，更有七成企业尚未设立正式的 AI 领导岗位。

Loading...

Taking too long?

Reload document

| Open in new tab

The information contained in this market commentary is based solely on the opinions of Scott Goering, Evan Witte and Nick Elsner, and nothing should be construed as investment advice. This material is provided for informational purposes, and it is not, and may not be relied on in any manner as legal, tax or investment advice or as an offer to sell or a solicitation of an offer to buy an interest in any fund or investment vehicle managed by Battery Ventures or any other Battery entity. The views expressed here are solely those of the authors.本市场评论所含信息仅代表 Scott Goering、Evan Witte 及 Nick Elsner 之个人观点，绝非投资建议。本文旨在提供资讯，不得被视为法律、税务或投资建议，亦不构成任何销售邀约或购买 Battery Ventures 所管理基金的要约。文中所述观点仅代表作者本人。

The information above may contain projections or other forward-looking statements regarding future events or expectations. Predictions, opinions and other information discussed in this publication are subject to change continually and without notice of any kind and may no longer be true after the date indicated. Battery Ventures assumes no duty to and does not undertake to update forward-looking statements.文中可能包含关于未来事件或预期的前瞻性陈述。此类预测与观点会随时间不断更迭，恕不另行通知，且在发布日期后可能不再准确。Battery Ventures 不承担更新前瞻性陈述之义务。