3 top VCs gathered at SaaStr Annual to do a deep dive on AI in B2B today, from what customers really want in the enterprise to what VCs are funding:
- Karen Page, General Partner at B Capital
- Rudina Seseri, Managing Partner at Glasswing Ventures, which focuses on early-stage AI and Frontier Tech investments in Enterprise and cybersecurity markets.
- Joselyn Goldfein, Managing Director at Zeta Venture Partners, which invests in AI and data infrastructure-focused startups from inception through seed stage
And see everyone at 2025 SaaStr Annual, May 13-15 in SF Bay!!
What VCs Are Funding in AI Today
The AI funding landscape has evolved rapidly in 2023-2025. Here’s where the smart money is flowing:
- Vertical-specific AI applications that solve industry-specific problems in healthcare, fintech, and life sciences are attracting significant investment
- Enterprise AI governance tools to help large organizations manage model deployment, security, and compliance as AI becomes mission-critical
- AI developer productivity tools continue to see strong funding as they demonstrate immediate ROI through measurable engineering efficiency gains
- Data infrastructure for AI including vector databases, annotation tools, and data quality platforms that support the foundation of AI systems
- AI agents with specialized capabilities rather than general-purpose assistants, particularly those that integrate with existing workflows and enterprise systems
Notably, VCs are becoming more selective about “AI-native” startups, requiring demonstrable technical advantages beyond simple API integrations with large language models.
4 Unexpected Learnings:
- AI adoption follows reverse logic: Unlike traditional tech, start with employee-facing tools rather than customer-facing applications for higher success rates
- “AI Application as a Service” (AIS) is emerging as a new category distinct from traditional SaaS, with different implementation and go-to-market requirements
- Incumbents have unexpected advantages in AI adoption due to data assets and distribution networks, making direct competition challenging for startups
- Narrow AI solutions often win against broader platforms, contrary to the usual technology adoption pattern where comprehensive platforms typically dominate
AI Adoption: Interest vs. Implementation
The enterprise interest in AI right now is off the charts. Every public company earnings call mentions AI, and every boardroom discussion includes it. Yet there’s a massive gap between interest and implementation.
The harsh reality: Most enterprises are adopting AI due to FOMO (Fear Of Missing Out) rather than for specific business outcomes. A chief data officer at a top-five global bank recently shared they have 150 generative AI projects in the lab but zero in production.
If 2023 was the “year of the demo,” 2024 should be the “year of production” – but most companies are still struggling to deploy AI successfully. This creates both challenges and opportunities for founders.
Project Selection: Where Enterprises Go Wrong
Many companies stumble by deploying AI in high-risk, customer-facing applications first (like chatbots). This is exactly backward.
The winning approach: Start with employee-facing tools that deliver measurable productivity gains. Software engineering teams have been early adopters of AI coding assistants precisely because they provide an immediate, measurable lift.
The current state of AI adoption resembles the early days of cloud: great infrastructure exists, but there’s a lack of applications to use it effectively. Everyone is trying to DIY without knowing how.
Distinguishing True AI from AI Washing
Be wary of companies creating “thin wrappers” around OpenAI offerings and claiming to be AI companies. Investors have become increasingly skeptical of startups that simply add “AI” to existing products without deep integration or understanding.
The actual tech stack matters. To build a successful AI company in 2024 and beyond, you need:
- More substantial tech depth
- Proprietary technology advantages
- A clear path to production use cases
Navigating the “False Indication of Interest”
AI founders face a unique challenge: the “false indication of interest” phenomenon. Potential customers express enthusiasm for AI without actual buying intent.
How to overcome this: Focus on solving specific problems rather than selling “AI.” Whether it’s improving productivity, enhancing data quality, or optimizing processes, the value proposition must be concrete.
Building a Moat in the AI Era
The commoditization risk in both the infrastructure layer and thin application layer is substantial. To build a barrier to entry:
- Own the middle layer of your tech stack
- Address widely recognized problems that others haven’t solved
- Target areas where the market need is identified but solutions aren’t mature
Success Stories: FeatureByte, Weavi, and Laiva
FeatureByte
This company leverages AI to automate feature engineering, articulating clear ROI: “We can automate a process that would take months and $2M per dataset.”
Weavi
Founded in 2020, they anticipated the growing importance of unstructured data and embeddings. By predicting the rise of multimodal AI, they positioned themselves with the necessary infrastructure (including vector databases) to serve the next generation of applications.
Laiva
Becoming the platform of choice for life science companies and research institutions by creating a two-sided marketplace with significant SaaS components. They use AI for price discoverability and optimization, with a setup that drives annual retention.
Deployment Strategies: Top-Down vs. Bottom-Up
Infrastructure companies typically deploy top-down, while application-layer tools are more likely to follow bottom-up adoption patterns. The new term “AI application as a service” (AIS) describes companies selling AI-powered applications to mid-market and enterprise customers.
Product-led growth (PLG) motion applies well to AI-powered products. Examples like ChatGPT and Perplexity demonstrate how AI applications can be adopted through PLG strategies.
SaaS vs. AI: A Misleading Analogy
Unlike SaaS, AI isn’t necessarily disruptive to the existing tech stack. Incumbents can plug in AI-powered features without a complete rewrite.
The incumbent advantage: Distribution networks and vast data resources give established players a significant edge. Going head-to-head with a system of record simply because you’re “the AI-powered version” is often a losing strategy.
Building a Durable AI Business
To create a lasting AI business:
- Develop applications with stickiness and resilience to system-of-record functionality
- Build a clear moat in a market facing commoditization
- Solve previously unaddressable problems or develop advantages incumbents cannot easily replicate
Starting an AI company requires significantly higher setup effort than traditional SaaS, including building and training algorithms and collecting data.
Determining Demand and Budgets
When entering an existing category, understand go-to-market strategies and adoption signals. For new categories, thorough market research is essential.
Unlike the early days of SaaS, starting an AI company is expensive. Be prepared for higher capital requirements.
Current AI Capabilities and Limitations
AI excels at:
- Summarization
- Content generation
- Voice and image recognition
AI struggles with:
- Complex planning
- Multi-step workflows
Understanding “good enough” accuracy for your specific use case is critical. Tolerance for errors varies dramatically by industry and application.
Creating a Data Moat
A “data moat” refers to a unique and defensible data set providing competitive advantage. Options include:
- Curation and annotation of otherwise commoditized data
- Access to unique data sources (like wet lab data in life sciences)
- Creating new workflows that generate proprietary data
Having a narrow focus often leads to higher quality data access and lower likelihood of disruption by generalist infrastructure companies.
The Governance Opportunity
Many organizations are testing AI infrastructure that lacks governance controls. Large enterprises have an immediate need for governance solutions to handle AI at scale.
This represents an under-recognized opportunity for B2B AI startups focusing on compliance, risk management, and administrative controls.
Where It’s Harder to Get VC Funding in AI Today
Despite the overall enthusiasm for AI, several areas face increasing investor skepticism:
- “AI-washing” startups that simply integrate OpenAI APIs with minimal proprietary technology or differentiation
- General-purpose AI assistants competing directly with well-funded incumbents like ChatGPT and Claude, particularly those without specialized domain expertise
- Pure infrastructure plays without clear paths to adoption, as the market is saturated with well-funded options and cloud providers are rapidly expanding their offerings
- AI companies with high compute requirements but uncertain unit economics, as investors grow more concerned about the long-term cost structure of AI businesses
- Undifferentiated LLM fine-tuning businesses that lack proprietary data or novel techniques to create sustained competitive advantages
- AI startups targeting SMBs exclusively without a clear path to enterprise or mid-market customers, as monetization remains challenging in this segment
VCs increasingly expect AI startups to demonstrate not just technical capabilities but also clear paths to sustainable business models with attractive margins. The bar has risen significantly from the “growth at all costs” mindset of 2021-2022.
Final Thoughts
The AI space is well-funded but still maturing. The most successful companies will combine unique data advantages with strong algorithms and effective go-to-market strategies.
For founders, the key is balancing long-term vision with short-term proof points that can secure ongoing financing and market traction.
