Data Centers Lack 'Transmission,' Not 'Generation': AI's Real Power Bottleneck
- Gartner expects AI to lift power consumption by 26%, but the real bottleneck is transmission, not generation
- Grids and transmission capacity can't keep up, constraining data-center expansion
- For investors it is a core clue to whether AI infrastructure can scale up
Citing Gartner, ITmedia notes AI will lift power consumption by about 26% — but flags a key, often-missed fact: a data center's real bottleneck is not whether power can be generated, but whether it can be delivered. That distinction is crucial to judging whether AI infrastructure can scale. Why is transmission more binding than generation? Even with enough generation, power must travel from plant to data center via transmission lines and substations, and grids take long to build, are complex to permit and hard to site — so they lag demand's explosion. The result is a paradox: land, servers, even generation are in place, but because the 'last stretch' of grid can't connect, the center can't run at full load. That is the hardest reality behind today's 'data center beside homes' story too. For investors, the value of this clue is that realizing AI profits depends not only on chips and orders but on the grid gate. Regions with ample transmission become new data-center hubs; where transmission is stuck, even a hot theme struggles to land. Watch grid and transmission investment, and the shift of siting toward power-rich regions.