Andreessen Horowitz’s (A16Z) “16 Changes to AI in the Enterprise 2025” report reveals a rapidly evolving landscape where enterprises are accelerating their AI adoption with notable shifts in budget, strategy, and technology use. Comparing the 2024 and 2025 insights highlights significant developments, especially around the rise of agentic AI and changing enterprise behaviors.
1. Budgets and Spending Patterns
Enterprise AI budgets are soaring, with leaders expecting an average 75% growth in AI spending over the next year, driven largely by the discovery and implementation of new use cases, particularly customer-facing generative AI applications. This contrasts with earlier focuses on internal use cases. A striking 88% of organizations have increased AI budgets due to agents, with most increasing budgets by at least 10%.
Importantly, this budget growth is sourced differently: innovation budgets have shrunk from around 25% to 7%, while reallocated central IT budgets have jumped from 28% to 39%, and business unit budgets have also increased. This shift indicates generative AI investments are graduating from experimental to permanent budget lines, reflecting a strategic realignment around agentic AI capabilities.
2. Enterprise Behavior Mirrors Consumer Sophistication
Enterprises are adopting a consumer-like approach to AI, using multiple models tailored to specific tasks rather than relying on a single provider. For example, some use Anthropic’s Claude for fine-grained code completion, Google’s Gemini for system design, and OpenAI’s GPT-4.5 for writing. The number of enterprises using five or more models has risen to 37%. This multi-model approach reflects growing sophistication and the recognition that different models excel at different tasks.
3. Market Consolidation and Model Leadership
While many models exist, market share is consolidating around a few leaders. OpenAI remains dominant overall, with 67% of its users deploying non-frontier models in production, whereas Google (41%) and Anthropic (27%) are more concentrated at their highest-end offerings. Google has gained traction especially among large enterprises due to trust, compliance, and favorable performance-to-cost ratios. Large enterprises also favor open source models like Llama for custom builds, while cost reductions in closed-source models are making them more attractive for many customers.
4. Fine-Tuning Declines as Models Improve
Improved model capabilities and longer context windows have reduced the need for costly fine-tuning. Enterprises increasingly rely on off-the-shelf models, simply providing large context inputs instead of fine-tuning parameters. This shift has financial and strategic implications, impacting startups that had positioned fine-tuning as a core service.
5. Reasoning Models and Agentic AI
Enterprises are enthusiastic about reasoning models, which enable solving complex new use cases. While still early in adoption, 23% use OpenAI’s GPT-3.0 in production and 57% report that reasoning models accelerate adoption. Reasoning models and agents are not just improving existing tasks but enabling entirely new workflows.
6. Procurement and Cost Considerations
AI procurement increasingly resembles traditional enterprise software buying, with detailed checklists and emphasis on cost of ownership. Interestingly, reasoning and accuracy have decreased as major buying considerations, while cost has risen, reflecting the move to long-term, widespread usage rather than experimental innovation.
7. Hosting Preferences and Direct Provider Trust
Enterprises are becoming more comfortable hosting AI models directly with providers like OpenAI and Anthropic rather than exclusively through established cloud providers. This shift is driven by the desire for early access to the latest models and performance advantages, as well as employee demand for state-of-the-art tools similar to their consumer experiences.
8. Rising Switching Costs Due to Agentic Workflows
The complexity of agentic AI workflows increases switching costs between models because tasks often require multiple interdependent steps. Enterprises are investing in agent platforms that aim to control a larger ecosystem of use cases, though market forces may eventually drive interoperability. This marks a departure from prior efforts to keep models interchangeable for simple use cases.
9. Model Evaluation Trends
There is a notable rise in reliance on external benchmarks to evaluate models, though internal and project-specific benchmarks have declined slightly. This may be temporary, as current external benchmarks are sometimes insufficiently nuanced, and more sophisticated evaluation methods are anticipated to emerge soon.
10. Build vs. Buy Paradigm Evolves
The enterprise AI market is shifting from predominantly building internal solutions due to lack of options, to increasingly buying third-party AI applications, especially vertical and functional agents. However, this shift is bifurcated: non-regulated industries often adopt off-the-shelf apps with some customization, while heavily regulated sectors like finance and healthcare tend to build custom agentic solutions. The rise of agents that can build other agents supports ongoing internal development in high-value areas.
11. Pricing Models Remain Experimental
Enterprises struggle with outcome-based pricing models due to unclear measurable outcomes and unpredictable costs. Usage-based and hybrid pricing remain more popular, reflecting the experimental phase of AI application monetization[1].
12. Ubiquity of Certain Use Cases
Software development AI adoption has surged, rising from under 40% to over 70% of enterprises using it in production within a year, driven by ROI and improved apps. Other use cases gaining traction include enterprise search, data analysis, and data labeling, while customer service use has slightly declined.
13. Employee Influence on Enterprise AI Adoption
Much enterprise AI growth is driven by “proumer” markets—employees who use consumer AI tools like ChatGPT and push for enterprise adoption. CIOs often cite employee demand as a key driver for purchasing decisions.
14. Incumbents’ Advantages and Emerging AI-Native Competitors
Initial AI infrastructure favored incumbents due to trust, distribution, and capital availability. However, AI-native companies are beginning to outpace incumbents in quality and speed of innovation, particularly in software development tools. Enterprises favor AI-native firms for their rapid innovation pace.
The 2025 landscape is one of rapid acceleration and sophistication. Enterprises now behave more like savvy consumers of AI, juggling multiple models, demanding state-of-the-art capabilities, and integrating agents deeply into workflows. Budgets are growing and becoming permanent, with a clear split between buying off-the-shelf for common needs and building custom solutions for regulated, high-value applications. The rise of agentic AI is reshaping spending, procurement, and switching dynamics, signaling a new phase of AI enterprise adoption where speed, quality, and ecosystem control matter more than ever.
