Bitcoin Miners Switch On 750MW of AI Data Center Capacity in Q2 Filings

The 20-Watt Claim Unravels

A viral comparison making the rounds on LinkedIn and X holds that the human brain thinks on roughly 20 watts while the fastest supercomputer on record draws millions more — proof, the posts imply, of AI's extravagant energy appetite. The gap itself is real: LineShine in Shenzhen, which tops the global supercomputer ranking for the first time since 2017, performs 2.198 exaflops on entirely domestic Chinese CPUs while drawing 42.2 million watts. But the numbers circulating alongside it do not survive scrutiny. The claim that a brain needs just 12 watts traces to a 2023 paper in Frontiers in Artificial Intelligence that offers no citation at all — the sort of unverified statistic that spreads as classic FUD. That same paper produced the 2.7-billion-watt estimate for digitally recreating a human brain, extrapolated from a 10-million-neuron simulation scaled to mouse size and multiplied by a thousand. It also states the simulation ran about 30,000 times slower than biology, a detail social posts drop; secondary sources then credit the figure to the Blue Brain Project, which never published it. Even baselines wobble — the oft-quoted 86 billion neurons rests on four male brains and is under dispute in the journal Brain. Harder numbers do exist: Epoch AI pegged a typical ChatGPT query at 0.3 watt-hours in early 2025, and a peer-reviewed study in Joule reached 0.31, though reasoning models producing longer answers can cost several times as much.

tops the global supercomputer rankinghttps://x.com/T3chFalcon/status/2069744850031804448?ref_src=twsrc%5Etfw

Where silicon has caught up is architecture. Cortical firing averages below 1 hertz, with energy chasing change rather than clock cycles — the same sparse principle behind Kimi K2 activating 32.6 billion of its 1.04 trillion parameters per token, about 3.1%, down from Mixtral's roughly 28% in 2023, with DeepSeek-V3 now at 5.5%. DeepSeek trained a 671-billion-parameter model in eight-bit precision, and NVIDIA has since pretrained a 12-billion-parameter model in four-bit. The one difference silicon has not copied is the brain's fusion of memory and computation in one physical place: Stanford's Mark Horowitz showed that fetching an operand from memory can burn hundreds of times the energy of the arithmetic itself. Hardware built explicitly to imitate neurons has fared worse — Intel's Hala Point packs 1.15 billion artificial neurons across 1,152 chips yet remains a research prototype at Sandia National Laboratories, BrainChip reported $700,000 in customer receipts against $5.3 million of operating outflow last March quarter, and Rain AI explored a sale in 2025 after failing to raise $150 million.

Bitcoin Miners Hold the Power Cards

Efficiency has become the whole ballgame because electricity, not silicon, now gates AI expansion. The International Energy Agency put global data center consumption at 485 terawatt-hours in 2025, with AI-focused facilities growing 50% that year alone and expected to triple by 2030. Median time from an interconnection request to commercial operation now exceeds five years, according to Lawrence Berkeley National Laboratory, and Microsoft chief executive Satya Nadella said in November that his company holds processors it cannot plug in — the shortage is powered buildings, not chips. Bitcoin miners spent a decade solving precisely that problem. They hold energized sites, signed power agreements and interconnection rights that newcomers wait years to secure, and VanEck estimates retrofitting a working site at $3 million to $4 million per megawatt against $10 million to $12 million for greenfield construction. Public miners have signed AI contracts worth more than $70 billion in aggregate, yet second-quarter 2026 filings show roughly 750 megawatts actually energized across the sector: Core Scientific accounts for about 437 of them, Galaxy's Helios campus delivered 133, TeraWulf 102, IREN 50 and Riot 25. Hut 8 has contracted 949 megawatts and switched on none of it.

The pivot has been expensive. Combined quarterly losses at MARA and CleanSpark hit $851 million in August, and preliminary Cambridge survey data presented in July showed only about 10% of miners had allocated power to AI. The obstacles are physical: mining tolerates interruption, while AI tenants demand firm power, dense cooling and fiber that remote sites rarely have. Even so, the direction is unmistakable — Core Scientific now earns 83% of its revenue from colocation and just 13% from mining itself, and miners holding signed leases trade at far higher multiples of their energized power. Readers tracking the market in real time can follow live spot and futures prices on MEXC.

Power, Not Chips, Decides the Build

COINOTAG's read is that the two stories are one. Evolution optimized under a hard skull ceiling — a 200-watt brain would have killed its owner — while AI has faced only a capital ceiling, and capital stretches far more easily than electricity. With the IEA projecting a tripling of AI facility demand by 2030, the same scrutiny that dogged Ethereum's proof-of-work years now lands on data centers, and the next AI bet is increasingly a bet on power rather than chips. Bitcoin miners, holding energized sites the newcomers lack, are the unexpected incumbents — and AI-adjacent crypto narratives, from Fetch.ai to AI crypto wallets, trace back to the same compute-and-power race.