In the space of six weeks, bitcoin miner investors have been asked to evaluate a $9.1 billion Anthropic lease at Riot, a multibillion-dollar AI hosting agreement at Hut 8, a $3.5 billion bond deal funding Galaxy’s Helios campus, and quarterly reports in which lease revenue outgrew mining revenue for the first time. These announcements share a vocabulary — megawatts, colocation, contracted backlog, anchor tenants — that comes from the data-center REIT world, not the crypto world, and reading them with a miner’s eye leads to systematic mistakes in both directions. This guide is the missing translation layer: the four numbers that matter in every deal, the three traps, and a full worked example using the Riot–Anthropic agreement confirmed on August 11, 2026.
The four numbers in every announcement
1. Megawatts (MW). The industry’s unit of account. A data-center lease is fundamentally a power contract with a building attached: the tenant is renting delivered, grid-connected electrical capacity. When Riot leases Anthropic 191 MW, the relevant scarcity is not floor space but the interconnection — the utility-approved right to draw that much power, which can take five or more years to permit from scratch. This is precisely the asset bitcoin miners accumulated during the decade when nobody else wanted it, and it is why AI companies are signing with miners at all: as Data Center Dynamics put it in its coverage of the Riot deal, the purchase is scarce, already-interconnected capacity. 2. Term. Twenty years (Riot–Anthropic runs to June 2048) versus the 3–5 year hosting contracts of the ASIC era. Term length is what converts a customer into an anchor tenant — and what makes lenders willing to finance construction against the lease. 3. Total contract value. The headline number — $9.1 billion — is the undiscounted sum of decades of future payments, not money received, not money guaranteed, and not comparable to a quarter’s revenue. 4. The ramp schedule. When does capacity energize? Riot’s deal delivers its first 96 MW by end-2027 and the full 191 MW by June 2028; revenue before those dates is approximately zero. A $9.1 billion deal that starts paying in 18 months is a very different security from one paying today.
Backlog is not revenue: the per-megawatt-year test
The single most useful habit: divide everything into dollars per megawatt per year. Riot’s deal: $9.1 billion ÷ 191 MW ≈ $47.6 million per MW over the life of the lease; spread across roughly two decades of full operation, that is on the order of $2.3–2.4 million per megawatt-year. That figure is the sector’s first big public comp, and every future miner-AI deal should be tested against it: a deal pricing meaningfully below suggests weaker power economics or a weaker negotiating position; meaningfully above suggests premium power markets or extra services bundled in. The same arithmetic deflates hype: a “$10 billion” deal over 25 years on 300 MW is $1.3M per MW-year — materially worse pricing than a “$2 billion” deal over 10 years on 80 MW ($2.5M). The headline number rewards duration; the per-MW-year number reveals price.
Now compare against the business being displaced. Riot’s Q2 mining segment produced bitcoin at a direct cost of $49,912 per coin against a roughly $64K market price — a real margin, but one exposed to price, difficulty and a programmed halving every four years. The lease substitutes a flat, contracted dollar stream for that volatile margin. Whether that trade is good depends entirely on your view of bitcoin’s next two decades — which is why the same announcement can be simultaneously bullish for the stock and quietly bearish for the network’s hashrate, and why miner equities and bitcoin itself have decoupled all summer. TeraWulf is the cleanest illustration: Q2 crypto revenue fell 73% year over year while total revenue fell just 6%, because HPC leases — 71% of revenue — filled the hole.
Trap one: nobody mentions who pays for the buildout
A colocation lease is not free money; the landlord builds the building. Converting mining halls to AI-grade data centers means liquid cooling, redundant power distribution, raised-floor or high-density rack infrastructure — costs that run to millions of dollars per megawatt and land on the miner’s balance sheet before the tenant pays a dollar. Riot’s Q2 already showed $18.3 million of data-center fit-out costs inside a $23.2 million data-center segment; that line grows every quarter from here to 2028. The financing cost is the second half of the trap: Galaxy funded its Helios Phase II buildout with $3.507 billion of senior secured notes at 9.875% — the clearest public price yet for pivot capital. When you see a lease announcement, the questions in order: how much capex per MW, funded by what, at what rate, and how much dilution? A signed lease with unfunded capex is an option, not an asset.
Trap two: counterparty quality is the whole ballgame
A 20-year revenue stream is worth exactly as much as the tenant’s ability and obligation to keep paying. Things to check in every deal: Is the tenant named? Riot’s deal spent 24 hours as a Bloomberg-sourced mystery before Anthropic was confirmed; unnamed-tenant deals deserve a discount until they are not. Is the tenant creditworthy? An AI lab burning venture capital is a different credit from AMD or a hyperscaler — and most AI-era leases have not lived through a downturn yet. What are the termination and assignment clauses? Rarely disclosed in press releases, always material; look for the details in the next 10-Q. Concentration: Riot now has two tenants on one campus — better than one, but Rockdale itself (grid, weather, Texas power pricing) remains a single point of failure. None of this is a prediction that any tenant defaults; it is the discount-rate conversation the press releases skip.
Trap three: the network sees it differently
Every megawatt leased to an AI tenant is a megawatt not hashing. For the bitcoin network, the miner-AI migration is a slow structural shift in who secures the chain and where: hashrate concentrates toward operators with the cheapest stranded power and the least attractive AI real estate, while the best-connected campuses graduate out of mining entirely. Add the treasury dimension — MARA has pledged 18,750 BTC (over half its stack) against $750 million of loans, and Strategy has sold coins four weeks running — and the once-reliable miner-HODL supply sink is now, at the margin, a supply source. Neither dynamic says anything about bitcoin’s price this quarter; both say the industrial structure underneath the network in 2030 will look nothing like 2020’s.
The checklist, then the worked example
| Question | Where to find it | Riot–Anthropic answer |
|---|---|---|
| How many MW, and is the interconnect approved? | Press release; utility filings | 191 MW at Rockdale — existing, energized campus |
| Term and extensions? | Press release; 8-K | 20 years to June 2048; two 5-yr options (→ ~$16.1B) |
| $ per MW-year? | Your own division | ≈$2.3–2.4M — the sector’s new public comp |
| Ramp schedule? | Press release | 96 MW by end-2027; full 191 MW by June 2028 |
| Who funds the buildout, at what rate? | 10-Q; financing announcements | Riot: $1.2B liquidity; fit-out already $18.3M/qtr. Sector rate: Galaxy’s 9.875% |
| Named, creditworthy tenant? | Confirmation, not reports | Anthropic — confirmed Aug 11; second tenant AMD on same campus |
| What happens to the mining fleet? | Earnings calls | Mining continues (cost/BTC $49,912) but capex priority has visibly shifted |
Run the checklist and the Riot–Anthropic deal grades as the sector’s strongest to date: named credit-worthy tenant, energized site, second tenant already on campus, defined ramp, and per-MW pricing that every subsequent deal will be measured against. It still carries the two open risks every such deal carries — execution on an 18-month buildout, and a tenant class (AI labs) whose long-term economics remain unproven. Which is the honest summary of the whole miner-AI trade in August 2026: the leases are real, the math is checkable, and the discount rate is a matter of opinion. Now you have the tools to form yours.
What is a colocation lease?
The tenant rents delivered power capacity and data-center space from the landlord (the miner), typically supplying its own computing hardware. The miner earns contracted dollar revenue; the tenant gets grid-connected capacity years faster than building from scratch.
Is a $9.1 billion deal really worth $9.1 billion?
Not in present-value terms. It is the undiscounted sum of ~20 years of future lease payments, contingent on buildout, execution and tenant solvency. Divide by MW and years (≈$2.3–2.4M per MW-year here) to compare deals honestly.
Why do AI companies lease from bitcoin miners?
Grid interconnection is the bottleneck. Miners hold approved, energized capacity accumulated over a decade — buying access to it is faster than permitting new sites.
What should I check first in the next deal announcement?
Whether the tenant is named and confirmed, then the per-MW-year price, then who funds the capex and at what rate.
Does this help or hurt bitcoin?
Mixed: healthier miner balance sheets reduce forced selling, but megawatts leaving mining and treasuries being pledged or sold change the network’s industrial base. This is context, not investment advice.
Disclaimer: This article is for informational purposes only and does not constitute investment advice. Cryptocurrencies and crypto-linked equities are volatile and you can lose money. Do your own research and consult a licensed financial advisor before making investment decisions.