Artificial Intelligence is no longer a futuristic vision—it’s a crowded, rapidly evolving marketplace where new startups emerge every week. In 2025, identifying the AI companies that will actually endure and thrive requires sharper criteria than ever before. Here’s the framework we, as AI Capital Funds company, use to separate hype from substance.
The Quick Screen: Spotting Signals in Minutes
When looking at a startup, the first filter is a fast yes/no test:
- Acute Problem + Clear Payer: Does the product solve a painful, high-frequency problem that a defined customer segment is already willing to pay for?
- Moat Path: Can they develop defensibility within 12–18 months—through proprietary data, strong workflow integration, or cost advantages?
- Healthy Unit Economics: Is there a believable path to >70% gross margins at scale, accounting for model/API costs, evaluation, and infrastructure?
- Regulatory Readiness: Especially for EU markets, are they prepared for compliance with the EU AI Act (coming into effect August 2025)?
If a startup clears these basics, we dig deeper.
The AI Stack: Where Moats Emerge
In 2025, defensibility is rarely about raw model capability alone. Open-weight models (like Llama 3.1) have narrowed the performance gap with closed models. That means winning startups distinguish themselves through:
- Unique data: proprietary, rights-clear sources and live feedback loops.
- System design: orchestration, tool use, evals, and agent reliability.
- Workflow depth: embedding into the “system of record” (CRM, IDE, EMR).
- Distribution: bottoms-up product-led growth combined with strong partnerships.
- Unit economics: smart use of caching, batching, distillation, and leveraging on-device NPUs for cost, latency, and privacy advantages.
The 30-Minute Diligence Sprint
When we sit down with founders, we focus on:
Product & Traction
- Clear buyer persona and use case.
- Retention metrics (D7/D30 activity, task success rates).
- Evidence of real adoption beyond the founder’s network.
Tech & Data
- Portability across models (no single-vendor lock-in).
- Data rights clarity and feedback loops.
- Robust evals and safety measures.
- Edge strategy: offloading to devices with NPUs where feasible.
Go-To-Market
- Dual motion: organic bottoms-up traction and enterprise readiness.
- Partnerships or distribution deals already in motion.
Compliance & Risk
- A plan for EU AI Act transparency and documentation.
- Security practices around data use and retention.
Market Signals
- Recognition in credible lists like CB Insights AI 100.
- Backing from investors who understand AI economics.
Sector-Specific Markers of Success
- Infrastructure: 10× improvements in cost, latency, or developer experience; proven drop-in compatibility.
- Horizontal Copilots: deeply integrated in workflows, delivering measurable ROI.
- Vertical Solutions: domain expertise, compliance certifications, validated outcomes.
- Edge AI: leveraging NPUs for privacy and real-time performance.
Red Flags in 2025
- Unsustainable Unit Economics: Each task burns more in inference costs than it earns, with no plan to optimize.
- Shallow Retention: flashy demos but poor long-term usage.
- Data Risk: unclear training data rights or disregard for EU regulations.
- Vendor Dependency: locked to a single model provider with fragile prompt chains.
- Hype Over Substance: talking “AGI soon” instead of customer ROI today.
Our Scoring Rubric (100 Points)
- Problem intensity & buyer clarity (15)
- Moat path (20)
- Unit economics (15)
- Product traction (15)
- Tech excellence & portability (10)
- Compliance readiness (10)
- GTM engine (10)
- Team strength (5)
Startups scoring 80+ typically have a real shot at becoming category leaders.
Where We Look for Signals
- AI Index reports for macro trends.
- CB Insights AI 100 for sectoral emergence.
- Model ecosystem updates to track shifting table stakes.
- Hardware roadmaps (NPUs, Copilot+ PCs) for edge adoption.
- Regulatory updates (EU AI Act) to identify compliance-ready winners.
Final Thoughts
By 2025, the AI market is both more crowded and more transparent. Public benchmarks, open-weight models, and clear regulation make it easier to spot who’s building for the long game. The startups that win won’t just be the smartest with models—they’ll be the ones that combine technical strength, strong economics, and regulatory foresight into products customers can’t live without.