The Venture Capital AI Investment Thesis Is Already Broken

In 2023, the prevailing venture capital thesis for AI investment was: foundation models are the new infrastructure, application layer companies will capture value, and the platform shift creates a window for new entrants to displace incumbents before they adapt.

Two years later, the thesis has partially inverted. The incumbents — Microsoft, Google, Salesforce, Oracle — have adapted faster than expected, primarily by integrating AI capabilities into existing workflows where they have distribution advantages that new entrants cannot match. The application layer companies that were supposed to displace them are instead discovering that their most compelling use cases are being bundled into enterprise software contracts that customers already have.

The window is not closed. But it is narrower than the 2023 thesis assumed, and the companies best positioned to exploit it are not the ones VCs funded in the 2023 wave. The clearest opportunity that remains: highly specialised vertical AI applications in domains where the incumbents have no meaningful presence and where domain expertise creates a durable moat — clinical decision support, legal document analysis, scientific research acceleration.

The companies building generic AI productivity tools on top of foundation model APIs are discovering that “build on GPT-4” was never a business model. It was a prototype strategy.