The Vertical AI Illusion - What It Actually Takes for Startups to Survive in 2025

I analyzed Menlo Ventures’ “2025 State of Generative AI in the Enterprise” report and came to one clear conclusion: the success formula we knew for SaaS has been completely inverted.

Enterprise AI spending surged from $11.5 billion in 2024 to $37 billion in 2025 - a 3.2x jump that now captures 6% of the global SaaS market. The AI market has decisively shifted from build to buy, and the implications for startups are more unforgiving than most realize. Here are three currents you must face head-on to survive.

Infrastructure Belongs to Big Tech. Applications Belong to Startups.

Infrastructure - foundation model APIs, training clusters, compute - demands capital at a scale only incumbents can provide. That market is already locked up.

But the application layer tells a different story. Startups captured 63% market share, earning nearly $2 for every $1 incumbents made. That’s up from just 36% the year before.

Meanwhile, 76% of enterprises now purchase AI solutions rather than building in-house, a dramatic reversal from the near-even 53/47 split in 2024. Companies have stopped experimenting and started buying.

The one gap startups can confidently exploit is the application layer. The $19 billion application market - larger than infrastructure’s $18 billion - is where the real opportunity sits.

PLG Is 4x Stronger Than Traditional Software - and It’s Rewriting the Sales Playbook

Product-led growth accounts for 27% of all AI application spending. That’s nearly four times the 7% rate in traditional software. Factor in shadow adoption - employees putting AI tools on personal credit cards - and the real number may approach 40%.

The buying power has shifted from executives making top-down decisions to practitioners who try first and decide later. Bottom-up adoption is the default motion in AI.

The conversion numbers reinforce this: AI buyers hit a 47% conversion rate to production, nearly double the 25% rate of traditional SaaS. When users experience value directly, procurement follows.

Coding tools alone represent $4 billion in departmental spending - 55% of the total. The practitioners aren’t waiting for approval.

For Vertical AI, PLG Alone Is a Half-Answer

This is the point that demands the most caution.

Departmental tools like Cursor (coding) and horizontal tools like Gamma (presentations) thrive on pure PLG. A developer downloads Cursor, gets immediate value, and the tool spreads organically through the team.

Vertical AI - products targeting specific industries - doesn’t work this way. The vertical AI market tripled to $3.5 billion, with healthcare alone accounting for $1.5 billion. But these wins weren’t driven by viral bottom-up adoption.

Vertical markets require:

• Exclusive data moats: proprietary datasets that competitors can’t replicate
• Intensive customer management: deep, consultative B2B relationships
• Domain expertise: understanding regulatory, workflow, and integration requirements that generic tools never address

Healthcare’s ambient scribe segment hit $600 million not because doctors casually downloaded an app, but because companies built deep clinical integrations, navigated HIPAA compliance, and embedded themselves in existing EHR workflows.

The lesson is blunt: lightweight, self-serve approaches won’t penetrate vertical markets. You need classic enterprise sales layered on top of product excellence.

The Question Every Startup Must Answer

The old playbook - wrap an API, add a UI, acquire users cheaply - is dead. Only 16% of enterprise deployments and 27% of startup deployments qualify as true agents. Most remain fixed-sequence workflows. The gap between ambition and execution is enormous.

Your service needs to be more than a feature. Does it hold irreplaceable data? Does it embed so deeply into a workflow that ripping it out would be unthinkable?

If the answer is no, the $37 billion market will grow around you, not through you.

Based on Menlo Ventures’ 2025 State of Generative AI in the Enterprise report.