KOSUNI CRYPTO
Korea's Crypto Pulse, in English
$BTC $ETH $XRP $SOL $DOGE
Mining
Neutral

AI Power Management Mimics Bitcoin Mining Approach

Published September 27, 2026 5:32 AM · 9 views $BTC
AI Power Management Mimics Bitcoin Mining Approach

AI data centers are expanding their operational models by adjusting computing loads in response to power grid conditions, mirroring the approach used by Bitcoin (BTC) mining facilities that turn equipment on and off based on electricity costs and peak demand. Google announced in a March 19, 2026 post that it has included 1GW of data center demand-response capacity in long-term power contracts across multiple U.S. regions. The company explained it can limit or reschedule some machine learning workloads to reduce power consumption. Luxor Energy and Bentaus reported in an August 24 press release that they reduced AI inference GPU power usage by up to 75% within 500 milliseconds of ERCOT's 4CP signal, maintaining checkpoints so tasks could resume without interruption.

The key is not just energy savings but whether data centers can operate as flexible loads rather than fixed-demand systems. By distinguishing between time-sensitive tasks like AI chatbot responses and batch processing or overnight workloads, power usage can be adjusted across different times. This expands the discussion around AI infrastructure beyond GPU availability to include power contracts, demand response, cooling, and grid connection. BTC mining was an early adopter of this structure; Riot Platforms ($RIOT) noted in its Q2 2026 results that participating in ERCOT's demand-response program allows it to offset electricity invoices through tailgate credits when operations are temporarily suspended.

While miners can shut down without affecting customer response times, AI workloads involve service level agreements, latency tolerance, training schedules, cooling, and network equipment. Luxor Energy and Bentaus' experiment was limited to single GPUs; scaling across a campus requires determining which tasks can be delayed and controlling power restoration for thousands of devices. ERCOT reported 91,134MW of electricity demand on July 22, 2026, while Texas officials stated in an August 3 press release that ERCOT is reviewing over 47.4GW of connection requests, with about 90% related to data centers.

The International Energy Agency (IEA) projects global data center power consumption could more than double by 2030, reaching approximately 945TWh. In the U.S., data centers are expected to account for nearly half of electricity demand growth through 2030. Lawrence Berkeley National Laboratory estimates they could consume 11.8% of U.S. power by 2030, ranging from 9.5% to 15.3%. CPower, Bentaus, and Supermicro reported in a February 2026 press release that they reduced total GPU infrastructure power usage by up to 75% within 20 milliseconds using CAISO signals, though overall data center power reduction varies due to cooling, storage, network, and power conversion equipment.

Korean Source

This article is an English localization of a Korean-language crypto news report. Original headline: AI 전력 제어, 비트코인 채굴장 방식 닮아간다