The scale is explicit: the International Energy Agency estimates data centers consumed ~415 TWh in 2024 (about 1.5% of global electricity) and projects consumption more than doubling to ~945 TWh by 2030 in its Base Case. In the United States, the U.S. Department of Energy cites the 2024 LBNL report and says domestic data center electricity use is expected to double or triple by 2028, with data centers projected at 6.7%–12% of U.S. electricity by 2028 (up from ~4.4% in 2023).

The IEA explicitly calls out grid connection queues and states that, unless risks are addressed, ~20% of planned data center projects could face delays. That friction shows up in real projects: Joliet, Illinois, just approved a proposed $20B AI data center campus framed as the largest in the state—one more marker that “multi‑gigawatt” is becoming a normal unit in planning decks.

But here’s what most people still miss: efficiency is no longer “J/TH” or “PUE.” Those are table stakes. The winners in this new energy race will stack different types of efficiencies at once.

Site efficiency is how much “useful work” you can extract per delivered MWh once you include cooling, curtailment strategy, and downtime. This is where industrial design and operations discipline matter more than simple spec sheets: stable firmware, repeatable deployment, faster service loops, and fewer forced derates.

Market efficiency is how many revenue lines you can attach to the same electrons. This is where the industry is moving fast: demand response, grid services, “non‑firm” operations, and heat reuse can all turn a single input into multiple revenue lines.

The grid itself is starting to formalize the difference between loads that must be served and loads that can move. PJM Interconnection is now proposing a hard 50 MW line for behind‑the‑meter arrangements, pushing large colocated loads toward a very concrete signal: paying for “firm access” or accepting “non‑firm” service that can be curtailed.

The winning operators will be the ones who treat Bitcoin mining as flexible infrastructure—a power‑native workload that can monetize stranded power, create a “bridge revenue” layer while AI/HPC builds out, and turn waste into product through heat reuse at multiple scales (home, commercial, industrial).

Efficiency is increasingly becoming a full‑stack strategy. The most serious miners increasingly resemble energy and data center companies first—and Bitcoin specialists second.

The early Avalon V1 era proved a decisive truth: no matter how loud the marketing, shipping wins. During the 2013 ASIC race, Avalon shipped “boxy” units early while competitors struggled, and early customers captured outsized first‑mover returns while network difficulty rose quickly.