There's a meeting that happens in virtually every mid-market company right now. Someone from engineering makes the case for building a custom AI system: proprietary models, company data, full control. Someone from finance raises an eyebrow at the timeline and cost. And a business unit leader, who has already quietly installed Cursor or Clay for their team, wonders why anyone is still debating this.
The data has caught up with that business unit leader. According to Menlo Ventures' 2025 State of Generative AI in the Enterprise report (one of the most-cited benchmarks in enterprise AI adoption), 76% of enterprise AI use cases are now purchased rather than built internally. Eighteen months ago, it was basically a coin flip. That shift should reframe how your leadership team approaches every AI decision in 2026.
The $37 Billion Signal
Enterprise spending on generative AI hit $37 billion in 2025, up from $11.5 billion in 2024. That's a 3.2x increase in a single year. Menlo's team describes it as the fastest-scaling software category in history, going from $1.7 billion to $37 billion in just three years. It now represents 6% of the entire global SaaS market.
More telling than the total is where the money went. More than half, $19 billion, landed in the application layer (the user-facing software that sits on top of AI models and delivers actual business value). Not infrastructure. Not model APIs. Applications. Stuff that gets used every day by sales reps, engineers, and HR managers.
Gartner independently puts enterprise software spending at over $1.4 trillion in 2026, with AI driving growth rates not seen in the software industry in decades. Enterprises stopped paying for experiments.
Why Build Lost the Argument
For a while, the in-house build narrative was genuinely compelling. Bloomberg trained BloombergGPT for finance. Walmart built its own model called Wallaby for retail. The assumption was that proprietary data plus AI talent plus domain expertise would produce a defensible advantage. By 2024, 47% of enterprise AI solutions were still being built internally, and companies were split almost exactly down the middle.
The vendor ecosystem caught up fast. AI-native startups shipped specialized tools that could be deployed in weeks rather than months, built on top of the same foundation models any internal team would use anyway. Once foundation models became table stakes, the argument for building your own stack mostly collapsed.
Ready-made AI solutions are reaching production more quickly and demonstrating immediate value, according to Menlo's data. Production speed matters enormously right now. The window for competitive advantage through AI is open, but it won't stay open indefinitely.
One data point captures the gap: 47% of AI deals from purchased solutions reach full production deployment, compared to 25% for traditional SaaS. When your procurement converts at nearly twice the rate, you're getting to value faster and burning fewer cycles on implementation.
Startups Are Eating Incumbents' Lunch
At the application layer, AI-native startups captured 63% of the market in 2025, up from 36% just the prior year. Startups now earn nearly $2 in revenue for every $1 earned by incumbents in this space.
On paper, this shouldn't be happening. Established enterprise software vendors have entrenched distribution, existing customer relationships, large sales teams, and balance sheets that startups can't match. Yet in practice, AI-native challengers are outcompeting them across nearly every fast-growing category:
- Coding and developer tools (71% startup share): GitHub Copilot launched with every structural advantage: Microsoft's distribution, existing developer relationships, deep VS Code integration. Cursor (founded by four MIT graduates) built a better product by going deeper on the actual development experience rather than treating AI as a plugin. It scaled to $200M in ARR before hiring a single enterprise sales rep.
- Sales (78% startup share): Tools like Clay and Actively win because they own the workflow around the CRM: research, prospecting, personalization. They become the interface sales reps actually use day to day, with a clear path to replacing the system of record entirely.
- Finance and operations (91% startup share): Accuracy requirements create paralysis inside large incumbents trying to bolt AI onto legacy systems. Startups like Rillet and Campfire built AI-first from the ground up and are winning where speed matters most.
Incumbents are holding their ground in infrastructure: Databricks, Snowflake, and MongoDB together hold 56% of the AI infrastructure market, where reliability and integration depth still outweigh the startup advantage. But in the application layer where your team interacts with AI daily, the newer products are generally winning.
Don't assume your existing enterprise software vendor's AI add-on is the best option just because it's convenient. In most functional categories, there's an AI-native startup with a product that's demonstrably better. Start your evaluation with "who built the best product for this job" and work backwards to integration and procurement, not the other way around.
Where the ROI Is Happening Right Now
Coding is the clearest proof point. $4 billion in 2025 (55% of all departmental AI spend) went to coding and developer tools, making it the largest category across the entire application layer. Half of developers now use AI coding tools daily, rising to 65% in top-quartile organizations. Teams report velocity gains of 15% or more. Cursor reached $2 billion in annualized revenue in roughly 18 months, up from essentially nothing. If your engineering team isn't using these tools, that's a gap that's compounding.
The depth of this category warrants attention: Menlo tracks the shift from $550M to $4B in a single year as a reflection of a genuine capability change, not just hype adoption. Models can now navigate complex codebases, handle multi-file edits, and run autonomous debugging loops that would have required a developer's full attention a year ago. The tools got good enough to justify the spend.
Customer success and marketing come next. Each captures about 9% of departmental AI spend. The use cases are clear (support ticket resolution, content generation, personalization) and the feedback loop on results is fast. AI-native tools are winning here because they were designed specifically for these workflows, not retrofitted onto them.
HR at 5% is likely underweighted. Candidate screening, onboarding automation, and policy Q&A are all high-value, low-risk use cases that AI handles well, freeing HR teams for higher-judgment work.
Vertical AI: from $1.1B to $3.5B in one year. Spending on industry-specific AI tripled in 2025, with healthcare leading at $1.5 billion (ambient scribing alone represents $600 million). If your business operates in a regulated or specialized vertical, there are almost certainly domain-specific AI tools that have already solved the compliance and workflow integration problems your internal team would spend 18 months wrestling with.
The Procurement Shift That Most Executives Are Missing
Product-led growth now drives 27% of all AI software spend, four times the rate of traditional software. Factor in shadow AI (employees using tools their company hasn't formally procured), and the real figure is closer to 40%.
Cursor is the canonical example. It reached $200 million in ARR without a single enterprise sales rep, growing almost entirely through individual developers who paid $20 a month out of pocket, found the product so useful they couldn't stop using it, and eventually pulled it into their organizations. By late 2025, enterprise customers accounted for 60% of Cursor's revenue.
Shadow AI is a security problem, yes, but it's also your clearest signal about which products are worth formalizing. Run the audit before you run the RFP. The best tools are often already in the building.
A Buy-First Framework for 2026
Start with engineering. AI coding tools are the first deployment to make. The ROI evidence is overwhelming, the tools are mature, and every engineering team that isn't using them is already behind.
Buy for standard workflows, build only for real differentiation. Customer support automation, sales prospecting, HR self-service, and meeting summarization are all well-solved by existing products. Build only when the use case represents your actual competitive advantage, and even then, consider buying the platform and building the customization layer on top.
Evaluate startups alongside incumbents. The 78% startup market share in sales AI and 91% in finance/operations aren't flukes. Treat AI procurement like a new product evaluation, not a software renewal. Run pilots. Look at production rates. Ask vendors for references from companies your size.
Compress your procurement timeline. The Menlo data shows AI deals convert to production at 47%, nearly twice the 25% rate of traditional SaaS. Buyers move faster because the value is clear. An AI procurement process that drags on for six months is leaving productivity gains on the table.
Audit shadow AI first. What tools are your teams already using? That answer is your clearest signal about what delivers real value versus what looked good in a demo.
The Window Is Real
The companies winning the AI application layer right now (in coding, sales, HR, and customer success) are establishing habits, workflows, and integration depth that will be hard to displace. Incumbents with entrenched distribution advantages couldn't keep up. Smaller vendors without differentiated AI capabilities won't either.
Mid-market companies running production deployments in 2026 will ship faster in 2027 than competitors still in evaluation mode. The gap compounds.
The market voted with $37 billion.