An exhaustive report from VC titan Andreessen Horowitz does an impressive job of capturing current details into the state of GenAI use in enterprises, covering everything from investment levels to model selection to vendor preferences.
While the report, “16 Changes to AI in the Enterprise: 2025 Edition,” goes broad and deep into a wide range of topics, I’m going to present five highlights that particularly piqued my interest for their alignment with ongoing coverage at Cloud Wars. Those five are:
- Enterprise investment levels in AI
- Enterprise usage of AI models
- How AI technology acquisition is evolving
- Why customers increasingly buy — instead of build — AI agents
- Why enterprises are embracing AI-native startups
More details on these 5 points are below, but the entire report and accompanying analysis are extremely insightful and worth reviewing in their entirety. The data comes from research, including responses from 100 CIOs across 15 industries.
AI Spending on the Rise
Spending on Large Language Models (LLMs) has grown quickly and shows no sign of slowing down. Enterprise leaders expect an average of 75.7% growth over the next year, from $7 million in May 2025 to $12.3 million on average in 2026. The figure was $2.5 million in 2024.
Spending growth is driven partially by enterprises discovering more relevant use cases and increasing employee adoption. Another big factor is the shift in mindset toward applying AI to customer-facing use cases rather than just internal use cases. That’s an encouraging sign, and a useful counterpoint to other data that has found a major focus on use cases more narrowly focused on driving internal efficiency.
In a related sign of maturity in AI usage, 39% of firms have reallocated GenAI to central IT budgets, vs. 28% one year ago, while just 7% are now funding GenAI projects from an innovation budget, vs. 24% one year ago, indicating GenAI is moving more and more out of the experimental stages. Additional AI spending is expected to transition to “core” budgets, the analysis indicated.
Multiplicity of Models
Andreessen Horowitz’s analysis noted it’s become the norm to have multiple models deployed in production use cases. Customers are seeking to avoid lock-in but also seeking the best model for a given use case: 37% of 2025 respondents are now using 5 or more AI models, vs. 29% of respondents last year.
This finding demonstrates an increasing level of comfort and confidence in selecting and deploying models that are best suited to the use cases at hand, another indicator of growing maturity.
AI Tech Acquisition Gains Enterprise Qualities
While they continue to invest in a range of models, enterprises increasingly apply more traditional software buying practices to AI, including stringent price controls and evaluation measures for functionality and quality. The cost of ownership as a product consideration criterion increased significantly vs. the number of respondents selecting it one year ago.
Buying Not Building
The data show that early in the AI product cycle (think 2024), enterprises often opted to work directly with AI models and build their own applications. However, the latest data reflect a shift toward buying third-party applications, at least in part reflecting the expanding ecosystem of AI apps. The report indicated companies are finding that internally developed tools are difficult to maintain and frequently don’t give them a business advantage, which furthers their interest in buying instead of building apps.
In the case of customer support — an oft-cited AI use case and one ripe for the power of AI copilots and agents — over 90% of survey respondents noted that they are testing third-party apps. Software development is another function with high percentages of respondents using third-party AI for production and testing.
AI-Native Innovators
When it comes to AI functionality and product development speed, AI native firms in many cases have an edge over larger, established vendors that are either building AI software from scratch or adding AI functions to existing products.
The chart below shows that innovation, native AI software (which hasn’t been retrofitted), and flexibility/responsiveness of a vendor are the top 3 factors in enterprises’ preference for selecting software from AI-native firms.
The gap between established vendors and upstarts is highlighted powerfully in software development and, more specifically, AI coding: Users who have adopted Cursor, an AI-native code editor, show notably lower satisfaction with previous-gen tools, including GitHub Copilot. While GitHub Copilot users who haven’t used Cursor register a NetPromoter Score (NPS) of 50, those who’ve used both give GitHub Copilot a 24, which compares with the Cursor NPS of 67.
AI Agent & Copilot Summit is an AI-first event to define opportunities, impact, and outcomes with Microsoft Copilot and agents. Building on its 2025 success, the 2026 event takes place March 17-19 in San Diego. Get more details.
