The Thesis in 60 Seconds We believe the shift from monolithic LLM inference to multi-step agentic workflows structurally changes the compute mix inside datacenters. Training-era architectures assumed GPUs would dominate every phase of inference. Agentic workloads have challenged that assumption. When an agent calls a tool, queries a database, waits for human approval, or orchestrates…

Executive takeaway. GTC 2026 reinforced our view that NVIDIA is shifting the competitive field away from performance and toward the economics (TCO) of the AI factory as a whole. We believe the decision is actively shifting to cost per token, tokens per second per watt, deployment speed, utilization, and revenue per megawatt, thus the moat…

Similar to the exercise we did on Anthropic, we wanted to do a similar deep dive on OpenAI. In our view, both companies have meaningful overlap but also important differences in technology, strategy, and market positioning. The exercise for both was a hypothetical: what would an initiation note look like on these companies, given neither…

To quote NVIDIA’s CEO Jensen Huang, from a QA he participated in at the Morgan Stanley TMT conference, “I love constraints. I love constraints. And the reason for that is because in a world of constraint, you have no choice but to choose the best. You can’t squander your choice.” As a series of reports…

While we don’t always favor a sum-of-the-parts analysis and framing for every company, we think it applies perhaps most relevantly to Apple. The deep ecosystem experience is a byproduct of a product excellence that hooks customers, keeps them loyal, and naturally extends their value to that ecosystem over time. The key upside driver for Apple…

In Part 1 we laid out the case for the copper-to-fiber transition via the physics forcing function, the superlinear scaling of connectivity costs, and the TAM repricing that we believe is still underappreciated. Part 2 is our continuation where we connect the thesis to specific companies, specific timelines, and specific risks. We profile nine companies…