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 (+0,31%) (GE Vernova), $RR. (-0,87%) (Rolls-Royce Holdings) and $BE (+0%) (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 (-0,11%) (Quanta Services), $PRY (-0,72%) (Prysmian) and $HPS.A (+2,78%) (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 (+0,19%) (Powell Industries), $VICR (+2,1%) (Vicor) and $MPWR (+0,85%) (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 (+0,32%) (Comfort Systems USA), $EME (+0,36%) (EMCOR Group) and $STRL (-1,28%) (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.
