AI预算来源拆解：55%新增，定价需超2倍竞品以获客
AI Spend Hit $177 Billion This Year. Only 22% of It Came Out of Software Budgets. The 4th Annual State of Enterprise AI Report from Menlo Ventures,
直接给出B2B AI产品的定价策略判据（2倍于竞品）和资金来源分析，帮助创业者避开狭小的软件预算池，极具实操价值。
Menlo Ventures just released its fourth annual State of Enterprise AI report. Enterprises will spend $176.8 billion deploying AI in 2026, up 5.4x from $33 billion in 2025. AI deployment spend now equals about 30% of public cloud revenue, up from about 6% a year ago.
Menlo Ventures 刚刚发布了其第四份年度企业 AI 状态报告。2026 年，企业在部署 AI 上的支出将达到 1768 亿美元，较 2025 年的 330 亿美元增长 5.4 倍。目前，AI 部署支出约占公有云收入的 30%，而一年前这一比例约为 6%。
For B2B founders, the more useful number is where that money comes from.
对于 B2B 创始人而言，更有用的数字是这些资金来自哪里。
Menlo surveyed roughly 400 U.S. enterprise decision-makers who together control more than $4 billion in AI spend, and asked which budget their incremental AI dollars came from. 55% was net-new budget. 22% came from software budgets. The rest came from labor and other lines.
Menlo 调查了大约 400 名美国企业决策者，他们共同掌控着超过 40 亿美元的 AI 支出，并询问其增量 AI 资金来自哪个预算。55% 为新增预算。22% 来自软件预算。其余部分来自劳动力及其他科目。
Of the 22% from software budgets, 13 points went to incumbents converting their own seat licenses into AI products. AI-native vendors got 9 points, or $12.6 billion.
在来自软件预算的 22% 中，13 个百分点流向了 incumbent（现有厂商），它们将自身的席位许可证转化为 AI 产品。AI 原生厂商获得了 9 个百分点，即 126 亿美元。
Five takeaways for anyone building, selling, or investing in B2B + AI:
任何构建、销售或投资 B2B + AI 的人都有五点启示：
- 55% of AI spend is new budget. Products that have to win a software budget are competing for the smallest pool.
- Incumbents kept 13 of the 22 points that moved out of software budgets.
- AI cut SaaS growth by about 2 points, from 13% to 11% on a market of more than $600 billion.
- The average enterprise spent about 2.5x its AI plan. The median spent about 1.2x. A small group of power users drives the totals.
- Incumbents held or gained share in nearly half the categories Menlo tracks. ServiceNow has 91% of AI spend in IT, and Workday has 68% in HR.
- 55% 的 AI 支出来自新增预算。必须争夺软件预算的产品竞争的是最小的资金池。
- 现有厂商保留了从软件预算中流出的 22 个百分点中的 13 个百分点。
- AI 使 SaaS 增长率下降了约 2 个百分点，在超过 6000 亿美元的市场中，从 13% 降至 11%。
- 平均每家企业的实际支出约为其 AI 计划的 2.5 倍。中位数约为 1.2 倍。一小群重度用户拉高了总体数据。
- 在 Menlo 跟踪的近半类别中，现有厂商保持或增加了市场份额。ServiceNow 占据了 IT 领域 91% 的 AI 支出，Workday 占据了 HR 领域 68% 的 AI 支出。
#1. $12.6 Billion Moved From Software Budgets to AI-Native Vendors. The Labor and Net-New Pools Are Several Times Bigger.
#1. 126 亿美元从软件预算流向 AI 原生厂商。劳动力和新增预算池要大数倍。
Here’s how incremental AI spend breaks down by source:
以下是增量 AI 支出的来源细分：
- 55% net-new budget
- 22% from software budgets
- 13 points to incumbents converting existing SaaS licenses to AI
- 9 points ($12.6 billion) to AI-native vendors
- About 23% from labor and other budgets
- 55% 为新增预算
- 22% 来自软件预算
- 13 个百分点流向现有厂商，将其现有的 SaaS 许可证转化为 AI
- 9 个百分点（126 亿美元）流向 AI 原生厂商
- 约 23% 来自劳动力及其他预算
The matchup most founders picture is an AI-native startup and a legacy seat vendor fighting over the same software line. That’s the smallest of the three pools, and incumbents are winning most of it.
大多数创始人设想的对决场景是 AI 原生初创公司与传统席位供应商争夺同一笔软件预算。这是三个资金池中规模最小的，且现有厂商赢得了其中的大部分。
I’ve used a related rule in Fund II this year. I won’t invest in an agentic B2B company unless it unlocks budget beyond software, and the evidence I look for is deal sizes at least 2x what the pre-AI competitors get. A product priced like the seat product it replaces is competing for the 22%. A product priced at 2x or more that customers still buy is being paid for out of the other 78%.
我在今年 Fund II 中应用了一条相关规则。除非一家 agentic B2B 公司能够解锁超出软件范畴的预算，否则我不会投资它；我所关注的证据是其交易规模至少是 AI 前竞争对手的两倍。定价与其所取代的席位产品相当的产品，是在争夺那 22% 的份额。而定价为其两倍或以上、客户仍愿意购买的产品，则是从那 78% 的其他预算中支付的。
We’ve seen the same thing at SaaStr. Our agent spend came out of headcount we didn’t backfill. We run with 3 humans and 20+ agents now, and revenue is higher than it was with a much larger team. Any vendor that wanted that budget had to take over work people were doing.
我们在 SaaStr 也看到了同样的情况。我们的 Agent（智能体）支出来自我们没有补员的裁员名额。现在我们由 3 名员工和 20 多个 Agent 组成团队，收入比之前拥有庞大团队时还要高。任何想要获得这部分预算的供应商都必须接手人们原本在做的工作。
#2. AI Took About 2 Points Off SaaS Growth: 13% Down to 11%
#2. AI 使 SaaS 增长减少了约 2 个百分点：从 13% 降至 11%
Menlo estimates AI cut application SaaS growth from about 13% to about 11% on a market of more than $600 billion. That’s roughly $12 billion of growth that didn’t happen. For an individual seat-priced company, it can show up as a flat year.
Menlo 估计，AI 将应用类 SaaS 的增长率从约 13% 拉低至约 11%，市场规模超过 6000 亿美元。这意味着大约 120 亿美元的增长未能实现。对于按席位定价的公司来说，这可能表现为全年业绩持平。
It matches the public company results we’ve covered all year. Total software spend is growing faster than the median SaaS company, and the categories priced per human seat (CRM, sales, marketing, collaboration) are the ones that slowed.
这与我们全年报道的上市公司结果相符。总软件支出增速快于中位数 SaaS 公司，而按人类席位定价的类别（CRM、销售、营销、协作）则是增速放缓的部分。
Menlo found little evidence that enterprises are replacing their systems of record (CRM, HRIS, ERP) with AI. Those systems are staying. What’s shifting is how much new budget goes to them and how much goes to tools built around them.
Menlo 发现很少有关于企业用 AI 替换其记录系统（CRM、HRIS、ERP）的证据。这些系统依然存在。发生变化的是分配给它们的新预算比例，以及围绕它们构建的工具所获得的预算比例。
#3. ServiceNow Has 91% of AI Spend in IT. Workday Has 68% in HR.
#3. ServiceNow 占据了 IT 领域 91% 的 AI 支出。Workday 在 HR 领域占据 68%。
Menlo tracked AI app spend by department. Incumbents gained ground in nearly half the categories:
Menlo 按部门追踪了 AI 应用支出。 incumbent（现有主导厂商）在近一半的类别中扩大了份额：
Where the incumbent owns the workflow, the data, and the permissions, the incumbent captures the AI spend. ServiceNow runs the ticket queue. Workday holds the employee record. For both, AI is an upsell to a base they already control.
当 incumbent 掌握工作流、数据和权限时，incumbent 就能捕获 AI 支出。ServiceNow 管理工单队列。Workday 持有员工记录。对两者而言，AI 都是对其已控制基础业务的向上销售（upsell）。
Customer success went to AI-natives. Sierra and Decagon lead the fastest-growing department at 4.3x, because AI can handle a support conversation from start to finish and the answer can be checked.
客户成功领域则流向 AI 原生厂商。Sierra 和 Decagon 以 4.3 倍的增速领跑该部门，因为 AI 可以处理从开始到结束的支持对话，且答案可被验证。
Menlo says the fastest-growing functions share three conditions:
Menlo 表示，增长最快的功能领域具备三个共同条件：
- Agents can take action in connected systems
- The context they need is available
- Outcomes can be verified
- Agent 能够在互联系统中采取行动
- 所需上下文可用
- 结果可验证
Support meets all three. HR and IT do too, but the incumbent owns the systems and the context. If you’re building in one of those categories, your plan for getting that data belongs in the first pitch.
支持服务满足所有这三个条件。HR 和 IT 也满足，但 incumbent 拥有系统和上下文。如果你正在构建其中一个类别的产品，获取数据的计划应放在你的首次推介（pitch）中。
#4. Sales Is a $1.4 Billion AI Category. Coding Is $34 Billion.
#4. 销售是一个 14 亿美元的 AI 类别。编程是 340 亿美元。
Coding is $34 billion, up 7.7x. Workplace assistants and chatbots are $25 billion. Together they make up 76% of AI application spend. Every other horizontal category combined comes to $8.8 billion. Vertical AI is $10.1 billion, led by healthcare at $3.7 billion and legal at $1.8 billion.
编程领域为 340 亿美元，增长了 7.7 倍。职场助手和聊天机器人领域为 250 亿美元。两者合计占 AI 应用支出的 76%。其他所有水平类别加在一起仅为 88 亿美元。垂直 AI 领域为 101 亿美元，其中医疗保健领域以 37 亿美元领先，法律领域为 18 亿美元。
Sales AI at $1.4 billion is about 4% of what enterprises spend on AI for coding. We’ve written for months that agents have taken over SDR work, much of first-line support, and a lot of CSM work, while there’s still no good AI account executive. The data fits. Sales AI is growing 3.6x a year from a small base, and the leaders are prospecting and data tools like Clay.
销售额达 14 亿美元的 Sales AI，仅占企业在编程 AI 上支出的约 4%。我们已撰文数月指出，AI Agent 已接管 SDR（销售开发代表）的工作、大量一线支持工作以及许多 CSM（客户成功经理）的工作，但至今仍未出现优秀的 AI 客户经理（Account Executive）。数据印证了这一点：Sales AI 从较小的基数开始，年增长率达 3.6 倍，领先者包括 Clay 等潜在客户挖掘和数据工具。
The coding number also shows how much the labs are taking. Claude Code and Codex account for $24 billion, or 70% of coding spend, up from about $1 billion a year ago. Independent coding apps (Cursor, Cognition, Factory, Lovable, and others) total $10.1 billion, up 2.9x. A product that’s a thin layer on a frontier model now competes directly with the model’s maker.
编程领域的支出数据也显示了各大实验室（Labs）所占的份额。Claude Code 和 Codex 占据了 240 亿美元，即编程支出的 70%，而一年前这一数字约为 10 亿美元。独立的编程应用（Cursor、Cognition、Factory、Lovable 及其他）总支出为 101 亿美元，增长了 2.9 倍。如今，建立在前沿模型之上的轻量级产品，正直接与模型制造商展开竞争。
#5. The Average Enterprise Spent 4.7x Last Year’s AI Budget. The Median Spent 2.2x.
#5. 企业平均 AI 预算是去年的 4.7 倍。中位数是 2.2 倍。
Enterprise AI adoption went from 55% to 65%, and spend per enterprise nearly quintupled. The average and the median tell different stories:
企业 AI 采用率从 55% 上升至 65%，每家企业的支出几乎翻了近五番。平均值和中位数揭示了不同的故事：
- Average: spend reached 4.7x the prior year, against a planned 1.9x
- Median: spend reached 2.2x, against a planned 1.8x
- 平均值：支出达到上一年的 4.7 倍，而计划增长为 1.9 倍
- 中位数：支出达到上一年的 2.2 倍，而计划增长为 1.8 倍
Menlo attributes the gap to agents that run in parallel and keep working after the user logs off. Spend tracks the amount of work done, so a company running 20 agents overnight spends at a different scale than one whose employees open a chatbot a few times a day.
Menlo 将这一差距归因于并行运行且在用户下线后仍继续工作的 Agent。支出追踪的是完成的工作量，因此一家整夜运行 20 个 Agent 的公司，其支出规模与一家员工每天仅打开几次聊天机器人的公司截然不同。
Two implications for founders:
这对创始人有两点启示：
- Most of your pipeline looks like the median. The typical enterprise buyer is spending about 2x last year and is still early.
- Per-seat pricing caps what you can earn from power users. These are the customers whose spend ran far past plan, and a seat price limits their bill to their employee count. Consumption, outcome, and resolution pricing scale with their usage.
- 你的大部分销售管道看起来像中位数水平。典型的企业买家支出约为去年的 2 倍，仍处于早期阶段。
- 按席位定价限制了你能从重度用户那里获得的收入。这些客户的支出远超计划，而按席位计价会将其账单限制在员工数量内。基于用量、结果或解决方案的定价则能随其使用量扩展。
#6. The Labs Took 69% of AI Deployment Dollars
#6. 实验室占据了 AI 部署资金的 69%
Anthropic, OpenAI, and Google captured $122.7 billion, or 69% of deployment spend, up from 50%. Their spend grew 7.5x, versus 3.3x for everyone else. AI-native companies overall hold 77% of application spend (up from 63%) and 90% of infrastructure spend. Excluding the labs, startups still hold 61% of app spend.
Anthropic、OpenAI 和 Google 捕获了 1227 亿美元，即部署支出的 69%，高于去年的 50%。它们的支出增长了 7.5 倍，而其他所有公司的增长仅为 3.3 倍。整体而言，AI 原生公司占据了 77% 的应用支出（去年为 63%）和 90% 的基础设施支出。排除实验室后，初创公司仍占据 61% 的应用支出。
This fits the departmental data. AI-natives lead in categories that didn’t exist before or that AI can handle end to end: coding, assistants, support, legal. Incumbents lead where AI is added to a workflow they already own: IT tickets, HR records, data platforms. Which kind of category you’re in should shape how you sell.
这与部门数据相符。AI 原生公司在以前不存在或由 AI 端到端处理的类别中处于领先地位：编程、助手、支持和法律。 incumbent（现有巨头）则在 AI 被添加到其已拥有工作流的领域占据主导：IT 工单、HR 记录和数据平台。你所处的类别类型应塑造你的销售策略。
A disclosure for reading the report: Menlo is an investor in many of the companies it covers, including Anthropic, OpenRouter, Lovable, Factory, Fireworks, Modal, and OpenEvidence. The budget-source, overspend, and departmental figures come from Menlo’s survey of 400 enterprise decision-makers. The agent-versus-human token figures come from OpenRouter, whose users skew toward developers and startups.
报告披露：Menlo 是其覆盖的众多公司的投资者，包括 Anthropic、OpenRouter、Lovable、Factory、Fireworks、Modal 和 OpenEvidence。预算来源、超支及部门数据来自 Menlo 对 400 名企业决策者的调查。智能体与人工令牌数据来自 OpenRouter，其用户群体偏向开发者和初创公司。
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力