The great pivot

Bitcoin miners are re-underwriting their megawatts from SHA-256 hashing to AI/HPC colocation. The arithmetic is simple: a MW leased to a hyperscaler is worth 3–5× more per year than the same MW mining BTC at current hashprice.

Source: analyst calculation. HPC lease benchmark ~$1.44M/MW/yr triangulated from CORZ CoreWeave contract disclosure (Q1-2026 and Q2-2024 earnings calls, S&P Global); self-mining economics at ~$100k BTC, ~$50/PH/day hashprice, ~30 J/TH, $0.04/kWh.

The four names sit at four different points on the curve. CORZ is the template — 243 MW invoicing CoreWeave against a 590 MW commitment, $3.3B project bond closed. RIOT crossed the signed-lease line in Q1-2026 (AMD, doubled intra-quarter from 25 to 50 MW, +150 MW of options). MARA has bought the power (Long Ridge 505 MW CCGT, Starwood JV) but has not signed a tenant. CLSK, the historic HODL holdout, spent fiscal Q2-2026 rebadging 1,175 MW as AI-suitable and marketing Sandersville to a lead prospective tenant.

Source: analyst compilation from RIOT, CORZ, MARA and CLSK Q1-2026 (calendar) / Q2-2026 (CLSK fiscal) earnings calls and presentations, S&P Global.

Disaggregating into signed / optioned / remaining HPC-suitable MW clarifies the shape — only CORZ has a signed book bigger than its pipeline that is actually billing today; CLSK's 835 MW "signed" figure is largely a re-badging of MW currently earning BTC.

Source: analyst compilation from company Q1-2026 (calendar) / Q2-2026 (CLSK fiscal) earnings calls and presentations, S&P Global.

Ranking on transition progress

The pivot has to be paid for, and consensus has the group's quarterly CapEx roughly doubling from Q1-2026 exit levels into Q3–Q4. CORZ has already been spending like a hyperscaler for four quarters; CLSK's curve ramps hardest into year-end — a spend the current stack cannot fully cover without either a signed HPC lease unlocking project debt (the CORZ template) or fresh paper.

CORZ

• Company-guided 2026 CapEx: ~$2.0B for FY26, explicitly "assumes no new customer contract"; includes ~$700M for the Hunt County (TX) + Polaris (Muskogee, OK) site acquisitions plus long-lead equipment procurement to precede ~1 GW of new billable capacity
• Build-cost benchmark: ~$11M/MW total build cost (20% cash outlay against 80% project-finance LTC) Source: CORZ, Q1 2026 Earnings Call Presentation; CORZ, Q1 2026 Earnings Call

RIOT

• Company-guided 2026 CapEx: No full-year 2026 CapEx figure disclosed
• Build-cost benchmark: AMD 50 MW lease total CapEx $172.7M, or ~$3.5M/MW blended ($3.6M/MW on initial 25 MW; $3.3M/MW on 25 MW expansion, "a significant reduction … driven by a leaner build-out scope"). Core-and-shell CapEx "unchanged from prior guidance" but not dollar-quantified Source: RIOT, Q1 2026 Earnings Call; RIOT, Q1 2026 Earnings Call Presentation

MARA

• Company-guided 2026 CapEx: No full-year 2026 CapEx figure disclosed
• Build-cost benchmark: Long Ridge (505 MW CCGT) closing brings owned-operated capacity to ~2.2 GW; MARA is explicit that the Starwood JV structure is designed "to drop [assets] into the partnership with limited amount of capital needs" — the site-level CapEx is meant to sit outside MARA's own P&L Source: MARA, Q1 2026 Earnings Call

CLSK

• Company-guided 2026 CapEx: No full-year 2026 CapEx figure disclosed
• Build-cost benchmark: AI build cost $9–11M/MW ("the range that we're seeing … the overwhelming majority [of CapEx] is related to AIDC"); Bitcoin-mining CapEx being wound down two years out from the next halving Source: CLSK, Q1 2026 Earnings Call

The next twelve months either validate the peer group's pipeline into contracted MW — or expose which names are perpetually two quarters away.