My last bottleneck posts were about HBM/Memory, Power & Cooling and Advanced Packaging. All three topics show that the critical points in the AI stack are shifting further and further into the physical infrastructure.
Today it's all about energy/grid.
Modern AI systems not only require computing power, but also enormous amounts of of stable energy. Several bottlenecks are now emerging in parallel. What I find particularly remarkable is that many investors think of energy almost exclusively in terms of electricity production. The actual bottleneck is spread across several levels of the infrastructure.
Level 1: Generation
This is where the energy itself is generated. For me $GEV (-0,95 %) (GE Vernova), $RR. (-0,52 %) (Rolls-Royce Holdings) and $BE (-0,34 %) (Bloom Energy) are exciting examples. The bottlenecks here are mainly in base load, flexible generation and local energy supply around new AI and data center load profiles. And Rolls-Royce could also become a player in Small Modular Reactors (SMR).
Level 2: Transmission
This is about being able to transport large amounts of energy at all. This sector is currently almost the biggest physical grid bottleneck. I find it particularly relevant here $PWR (+0,05 %) (Quanta Services), $PRY (+0,23 %) (Prysmian) and $HPS.A (+0 %) (Hammond Power Solutions). Typical bottlenecks are HVDC, high voltage, transformers and the massive expansion of the grid infrastructure.
Level 3: Distribution
Ultimately, the energy must be available locally, for industry, cities, data centers or critical infrastructure. There I look above all at $POWL (-0,61 %) (Powell Industries), $VICR (+1,7 %) (Vicor) and $MPWR (-0,65 %) (Monolithic Power Systems). The problems here are increasingly arising with power delivery, medium voltage, AI rack power and stable local grids.
Level 4: Implementation
This area is often underestimated. Because even if technology, capital and planning are available, projects still have to be physically implemented in the end.
That's why I think $FIX (+0,7 %) (Comfort Systems USA), $EME (-0,15 %) (EMCOR Group) and $STRL (+0,18 %) (Sterling Infrastructure) interesting. The bottlenecks here often lie in EPC capacities (plant construction), skilled workers, construction speed and the actual realization of large infrastructure projects.
Energy/Grid is currently so exciting for me because the bottleneck is not in a single place, but runs through the entire energy chain. but runs through the entire energy chain. That's why I consider this topic to be one of the most important bottlenecks around AI infrastructure at the moment.
