The VCs who bet big on AI are now asking the quiet question: what if these companies can't hold their lead?
The Signal
Matt McIlwain at Madrona dropped something honest at March Capital's summit that most investors won't say out loud. AI application companies are growing faster than anything he's seen in twenty years. That's the headline everyone wants. But his follow-up question cuts deeper: can they stay durable?
This matters because the entire AI app thesis rests on defensibility through fine-tuning and model orchestration. The pitch deck version says: we combine multiple models, tune them for our specific use case, build proprietary workflows, and create something competitors can't replicate. It sounds right. Companies like Harvey for legal work or Glean for enterprise search are racing to prove it.
But McIlwain's doubt points to the structural problem. If your moat is fine-tuning today's models, what happens when next quarter's foundation models are better out of the box? When OpenAI or Anthropic ship capabilities that make your custom stack obsolete? The AI app layer might be growing fast precisely because it's easy to build right now. Speed of growth and durability of business are not the same thing.
The smart money is still flooding in, but they're asking harder questions about retention, switching costs, and what happens when the foundation layer keeps moving. Growth is easy to fund. Businesses that last are harder to find.
The Implication
If you're building an AI app, your edge can't just be technical. Model fine-tuning is table stakes, not a moat. Focus on workflow capture, data network effects, or becoming so embedded in operations that ripping you out costs more than the subscription. Investors are learning to tell the difference between fast growth and staying power. You should too.
Source: The Information