AI/ML - Decoding the AI Layers

Artificial intelligence now absorbs more than 80% of venture capital dollars deployed globally. But beneath that wave, the market reality is far more uneven: thousands of "wrappers" built around generic models are competing head-on with each other, while a handful of deep tech companies are accumulating structural advantages that no one can replicate in a few months.

Citadelle Capital approaches AI and machine learning as a stack structured in successive layers (from raw compute to models, from orchestration layers to specialized applications). Our team evaluates each opportunity within its own layer: the competitive dynamics, capital intensity, and nature of the barriers to entry are never the same from one layer to the next.

Among these layers, agents and orchestration is one we follow closely, without making it our main focus. It's also the most crowded. Building an agent has become relatively easy, but a good portion of these companies are building features that the model providers themselves could natively integrate within a few months, erasing their differentiation overnight. Those are exactly the companies we look to avoid. The rare companies that will generate a real venture return are the ones that anticipated this risk early, and built their advantage elsewhere: in the workflow data accumulated with their clients over the years, in the depth of integration, and in the switching cost that creates toward a generic model.

Contact us to discuss this thesis in more detail.