What enterprise VCs actually want to fund in 2025

Silicon Valley's deepest pockets are officially moving on from the foundation model arms race.

A new TechCrunch survey of 20 enterprise venture capitalists reveals exactly where smart money wants to go in 2025. The consensus is clear: investors are rotating capital away from raw model training and toward task-specific workflows, enterprise resilience, and the plumbing needed to keep autonomous agents from breaking.

Features vs. companies: the 2021 SaaS ghost returns

Intel Capital senior managing director Mark Rostick pointed out that base models have largely been established. The next wave belongs to models tailored for specialized functions, especially when paired with agentic execution. Insight Partners managing director Mike Hayes echoed this, noting that he is hunting for startups solving messy enterprise friction that legacy software never managed to fix. That includes end-to-end vertical workflows and security tools that remediate threats instead of just firing off alerts.

Yet every investor surveyed seems haunted by the same ghost: the 2021 SaaS bubble. Back then, single-feature tools raised enormous valuations, only to be crushed in 2023 when CIOs slashed budgets and consolidated vendors. Boldstart Ventures general partner Ed Sim emphasized that the core challenge today is figuring out whether an AI startup is building a standalone business, a product, or just a feature destined to be swallowed by incumbents.

From agent pricing to system resilience

Reliability has suddenly become a marquee investment thesis. Team8 managing partner Liran Grinberg pointed directly to the massive CrowdStrike outage as proof of how fragile digital infrastructure remains, arguing that businesses desperately need anti-fragile architecture built into their systems by design. Battery Ventures investor Jason Mendel also highlighted observability and reliability as prime targets.

At the foundational level, AI infrastructure still has plenty of runway, but the focus has shifted. Janelle Teng, vice president at Bessemer Venture Partners, noted that momentum is shifting toward infrastructure that supports reasoning models, edge deployments, computer-use agents, and even startups solving how to price agentic labor.

Why it matters

This survey confirms what enterprise insiders have felt for months: the hype cycle has given way to cold, hard ROI calculations.

For founders, the takeaway is brutal. Thin wrappers around frontier APIs are uninvestable. If your startup is a single workflow tool that can be copied by Salesforce or Microsoft in a quarterly update, you will not survive enterprise consolidation. To win now, founders must either bury themselves so deep inside industry-specific workflows that incumbents cannot easily reach them, or build the critical guardrails and billing engines that enterprises need before letting AI agents touch production code.

For enterprise buyers, this pivot is welcome news. Vendors will be forced to compete on reliability, compliance, and tangible cost takeout rather than vague promises of artificial general intelligence. That said, we think caution is warranted around agentic automation. Building software that can independently remediate security flaws or operate business systems sounds appealing, but in mission-critical environments, proven dependability will trump novel autonomy every single time.