This video reflects on our favorite AI case studies of 2026 so far and what they mean for energy, industrials and equity markets. 1-2 year payback periods mean mild supply chain pull, then deep, persistent deflation. Most AI case studies are sub-kW scale, not GW scale. And maybe the few GW-scale deployments are “one and done”. We increasingly think the real upside in AI is in small-scale, distilled applications, which are light on energy and compute, heavy on sensors/actuators and deflationary for energy/materials supply chains?
We have reviewed hundreds of AI case studies in our 1Q 2026 research. In applications from oil and gas, to Chinese coal mines, to discovering new materials.
However some recurring findings are that AI deployments are generally motivated by saving cost and improving efficiency, which in turn, means lowering energy usage and lowering materials usage. This is nicely illustrated by one of our favorite case studies from hyperspectral imaging.
Increasing US electricity demand by each 1% pa requires adding 5GW of constant load. But most of the case studies we have seen likely require kW-scale energy use, rather than MW-GW scale energy use, which we will be flowing through to our data center energy consumption forecasts.
Maybe there are only relatively few examples of tasks that require GW-scale AI training, such as large language models, or autonomous vehicles, which in turn could be tasks that get “solved” and then do not need to be re-solved?
In other examples, we have found that AI helps to recover an order-of-magnitude more energy/materials than it uses (e,g., in waste management) and obviates the need for hundreds of billions of dollars of new grid infrastructure (e.g., via dynamic line ratings).
Hence while it has been fascinating, to deep-dive into all of the supply chains serving the AI eco-system — from GaN, to gas turbines, to capacitors and SSCBs — equity markets cannot live by AI supply chain bread alone, and we think the forward trajectories for these different sub-industries could diverge.
Like everyone else, we are trying to figure out this vast and interesting space. So we hope you enjoy the video, and any questions, pushbacks or challenges are very welcome.