Panel, Conference Presentation
Venture Capital Investing in AI | RAISE Summit 2024 | Paris
- Paris emerges as the new European AI capital: Nicola Granatino cited the rapid rise of the French ecosystem following the "Llama" paper, noting a shift from London to Paris as the hub for AI talent and innovation within a single year.
- Strategic shift to product-first AI: Venture Capitalists at Balderton and AIC are prioritizing founders who focus on solving specific customer ROI problems over those attempting to build foundational models from scratch.
- Foundation model commoditization: James (Balderton) argues that investing in base layers (transformers on internet-scale data) is a high-risk "death by commoditization" play due to massive capital requirements and limited winners.
- Differentiation via proprietary models: Successful startups like Writer (Palmyra) and PhotoRoom are being backed for their ability to evolve from product-first wrappers into building their own sustainable, proprietary diffusion models to maintain competitive moats.
- Data moats in biotech are superior to consumer internet: Charlie (ARK) highlighted that bio-health data offers a stronger defensive moat due to its "messy," highly siloed, and causal nature, contrasting it with the correlative, internet-hoovered data used for general LLMs.
- Clinical trial scale drives value: Ark's portfolio includes Freenome, which raised $1.3 billion to conduct large-scale clinical studies (1.3 billion dollar raise for colorectal cancer detection), demonstrating the high cost and necessity of rigorous data validation in health AI.
- Trust as a primary currency: General Catalyst views relationships with hospital systems and providers as a critical moat, noting that earning stakeholder trust is a prerequisite for accessing and solving complex disease problems.
- Shift in data value perception: The industry is pivoting from valuing raw Electronic Health Records (EHR) to prioritizing high-value human feedback generated when clinicians interact with and prompt AI models.
- EU governance advantage: While the U.S. lags in data governance questions, Europe is noted as being "further ahead" in establishing regulatory frameworks for AI and data rights in healthcare.
- Europe's defense tech deficit: Only five European defense startups have raised over $35 million in the same period that 100 U.S. startups have achieved this, a gap attributed to a cultural reticence in Europe to acknowledge the current reality of war and the resulting opportunity for meaningful impact.
- Corporate adoption gap: Antoine (AIC) identifies the lack of willingness among large European and French corporations to be early adopters as a fundamental barrier preventing the ecosystem from scaling, unlike the U.S. enterprise market which historically drives software growth.
- Global talent strategy: The consensus suggests Europe should market itself not just as a source of talent but as a destination, requiring startups to adopt a "bi-continental" mindset from day one, modeling success after Israeli founders who target the U.S. market immediately.
- Inference costs remain a hurdle: The high expense of model inference, combined with the difficulty of accessing proprietary data silos (e.g., Salesforce), presents a significant financial barrier to building full-stack AI businesses with high margins in the near term.
- Market saturation warning: Investors warn that while hundreds of billions in capital are currently flowing into GPU acquisition and model training, a "wall" is expected within a year where only startups with validated use cases and differentiated applications will secure further funding.
- Investment decisions disclosed: James admitted to passing on Mistral, while acknowledging he also passed on Moderna at a $100 million valuation, noting that early-stage prediction of success in foundational AI remains highly unpredictable.