Grayscale swaps its bitcoin miner index for compute that earns five to eight times per megawatt-hour

AI cloud services fetch an estimated median of $940 per megawatt-hour, against $113 to $179 for bitcoin mining ^[15]. Same electricity, five to eight times the revenue ^[1]. Core Scientific's second quarter of 2026 shows that landing on a profit and loss statement: $136.7 million of colocation revenue against $27.5 million from mining ^[14]. Hosting was about 83 percent of the two lines combined ^[2].

The equity market has already paid for the switch. Miners carrying AI or HPC contracts trade at about 12.9 times forward sales, against 3.7 times for those without, Cryptopolitan reported ^[16]. The contracted names carry three and a half times the multiple ^[3], on revenue the same report calls tens of billions of dollars announced and unbilled ^[7]. The index reconstitutes quarterly ^[5]. A name can qualify for the pivot half of the portfolio on a publicly announced intention to move into AI workloads ^[4].

Grayscale's case for durability rests on lead times. The release cites CBRE data putting North American data centre vacancy at a record low, with roughly six months of spare capacity in reserve ^[9]. New facilities can take as long as five years to build ^[10]. Sixty months of construction against six months of slack is a ratio of ten to one ^[4]. Grayscale's head of index, Steve Vanourny, wrote in the September 22 announcement: "GCPU is a direct extension of that thesis, giving investors access to the companies closing the gap between AI's demand for compute and the physical infrastructure needed to meet it" ^[8].

Scale cuts the other way. Goldman Sachs projects AI-related capital spending above $1 trillion a year ^[11]. The public mining sector's entire announced contract book is 7 to 10 percent of one year of that ^[5]. The miners' asset is the grid position. An already-permitted, grid-connected site is worth far more than the rigs sitting inside it, and a 2,600-gigawatt US interconnection backlog stands while OpenAI pushes roughly $280 billion of compute commitments, according to Cryptopolitan ^[17].

So the fund is priced on a queue. I would expect that queue to outlast the next several rebalances, because nothing built in a quarter answers a six-month vacancy buffer. The disappointment comes through the $940. That figure is a median, and if capacity arrives faster than the five-year figure implies, the convert sleeve compresses from 12.9 times toward 3.7 with every contract still intact. TeraWulf is up near 73 percent and Riot Platforms roughly 94 percent over the year to date ^[13]. A GCPU buyer is entering after that run. The test is CBRE's next vacancy print and the median hosting price: either one loosening, and half the portfolio re-rates down while every customer stays put.