This graph alone shows how global AI power consumption is getting out of hand very quickly - and it's not just about hyperscalers or OpenAI (www.techradar.com)

A new chart from Lennox Data Centre Solutions highlights a rapid shift in data‑center design: where racks once drew only a few kilowatts, AI‑focused racks are projected to hit about 1 MW each by 2030, while average racks climb to 30–50 kW. That means a single AI rack could consume 20–30× the energy of a general‑purpose rack, moving power delivery and cooling from supporting roles to strategic priorities for operators and designers. Technically, the industry is rethinking power and thermal architectures: high‑voltage DC (≈±400 V) is being adopted to cut losses and cable bulk, central CDUs will route liquid to rack manifolds and cold plates, and advanced approaches such as chip‑level microfluidics (Microsoft trials show up to 3× better heat removal and a 65% reduction in GPU temperature rise) plus AI hotspot mapping are being tested. The trends pressure supply chains and factory throughput—creating openings for smaller, agile players—but mean data‑center futures will be defined as much by heat rejection and power distribution as by compute.