Nine public bitcoin miners spent $5.11 billion on capital assets in the first half of 2026. Their AI and high-performance computing arms brought back $341.2 million. That is the 15-to-1 figure the trade press ran all week, and read alone it sounds like a verdict: bitcoin miners threw money at AI and got a rounding error. The number is real. The reading is early.
The count comes from BlocksBridge Consulting, whose Miner Weekly newsletter tallied capital spending as cash purchases of hardware, property and equipment, net of asset sales. So the $5.11 billion is the full buildout. The $341.2 million is only what the finished racks billed. Those two clocks are not synced, and that gap is the whole story.
Capex lands upfront, revenue arrives quarters later
A datacenter costs its whole price the quarter the concrete gets poured. It earns nothing until a tenant plugs in, and that can sit quarters away. Put half a year of construction against the trickle of revenue from the fraction already energized and you get a scary multiple by design. It has to fall as capacity fills.
Look at the quarter instead of the half and the bend shows. AI and HPC revenue across the nine miners hit $205.8 million in the second quarter, up 52 percent from the first. Six months of it totaled $341.2 million; a single recent quarter was most of that. That is revenue accelerating while the capex meant to carry it was still being poured.
Core Scientific shows the mechanism in one company
Core Scientific spent $797.5 million on capital assets in the second quarter alone. Its colocation revenue over the same stretch was $136.7 million, up from $77.5 million three months before. Billed capacity reached 437 megawatts by the middle of July. The revenue nearly doubled in a quarter because the megawatts came online, not because demand appeared out of nowhere. Read that company at a snapshot and it looks like it is bleeding. Read it across two quarters and it looks like a plant coming up to speed.
That $30.7 billion lumps clouds in with miners
The wider figure making the rounds is $30.7 billion of capex across 15 miners and data-center firms, a 42.6 percent jump over 2025. That bucket is where the comparison breaks. It quietly folds pure-play AI clouds in with bitcoin miners retrofitting old halls into GPU space, and the two are nothing alike on revenue.
| Company | 2026 capex | Recent quarterly AI revenue |
|---|---|---|
| Nine bitcoin miners (AI/HPC only) | $5.11B (H1) | $205.8M (Q2) |
| Core Scientific | $797.5M (Q2) | $136.7M (Q2) |
| CoreWeave | $14.12B (H1) | $2.58B (Q2) |
| Nebius | $8.13B (H1) | $582.3M (Q2) |
CoreWeave and Nebius alone account for more than $22 billion of that $30.7 billion, and CoreWeave books $2.58 billion in a single quarter against its spend. Averaging a cloud already at scale with a miner that energized its first tenant in June produces a ratio that fits neither. The 15-to-1 belongs to the miners, not the group the headline attached it to.
None of this makes the bet safe. The math that pushed miners toward AI in the first place still assumes tenants that many of the nine have only signed on paper, with leases that do not cash-flow until 2027. The debt raised to build is due whether the GPUs fill or not, and Core Scientific borrowed $3.3 billion to get its capacity finished. TeraWulf already books more from HPC than from mining, and HIVE grew its HPC line 94 percent to $19.5 million, so the pivot is real revenue, not a slide-deck promise. The ratio will keep dropping as racks energize. The question for next quarter is whether it drops fast enough to service what was borrowed, the same pressure now bearing on bitcoin miners even as bitcoin difficulty slipped year on year.
