Bitcoin gained 21.5% from the Aug. 17 close through Aug. 21, yet six of seven large US-listed miners finished the same trading stretch much lower. MARA Holdings rose 16.1% and came closest to $BTC, while Cipher Digital fell 14.8%, TeraWulf lost 11.2%, Hut 8 dropped 8.1%, and IREN declined 6.8%.
The stocks still sold as Bitcoin proxies had separated during one of Bitcoin’s strongest weeks of the year.
QQQ fell 2.3% over those sessions as long-term yields stayed volatile, placing the miners inside a weaker technology-equity market. Their corporate structure helps explain this, as several former mining specialists now derive revenue, financing, or forward valuation from long-duration>Debt and Bitcoin sales are financing those buildouts, adding construction schedules, equipment procurement, and customer concentration to valuations that once depended only on hash price.
Company filings place our little group of public miners pivoting to AI at several different stages of that conversion, with TeraWulf generating $31.9 million of its $44.8 million second-quarter revenue from high-performance-computing leases and roughly $12.8 million from digital assets.
Hut 8 says its Beacon Point leases cover 949 megawatts of contracted IT capacity and carry $26.6 billion of base-term contract value, subject to future delivery and tenant performance.
IREN reported $70.5 million of AI cloud revenue and $66.7 million of Bitcoin mining revenue in its June quarter, moving AI above mining in its current revenue mix. Its Aug. 27 release also put operating annual recurring revenue at $1 billion as of Aug. 26 and contracted ARR tied to 2026 capacity at $4 billion, with the latter targeted to become operational by Dec. 31 subject to commissioning, testing, and customer acceptance.
The quarter included a $450.4 million impairment largely tied to decommissioned mining hardware as sites convert to AI.
Cipher still recorded second-quarter revenue from Bitcoin mining, but it has contracted 700 megawatts of high-performance-computing capacity across three sites and began delivering the first capacity at Black Pearl in August.
Riot Platforms sits closer to the middle, reporting $113.7 million of mining revenue, $23.2 million from data centers and $37.3 million from engineering in a $174.2 million quarter. Its 241 megawatts of contracted AI capacity carry roughly $9.8 billion of company-estimated long-term revenue, giving investors a contract book to value alongside 11,380 Bitcoin held at June 30.
CleanSpark would have served as a mining-only control earlier in the year, though that classification expired before the August breakout. The company signed a 20-year, $6.6 billion> MARA now provides the closest large mining-led comparator, even as it explores adjacent energy and computing businesses.
The split shows up in the beta
CryptoSlate analyzed Alpaca/IEX historical equity closes for HUT, WULF, IREN, CIFR, RIOT, MARA, CLSK, and QQQ, together with Alpaca’s $BTC/USD closes, from Aug. 22, 2024 through Aug. 24, 2026.
Daily stock returns used exchange trading days, and each Bitcoin return covered the interval between consecutive stock-market dates, so Monday observations included the weekend. Ten-year Treasury yields came from the Federal Reserve’s DGS10 series.
Close-to-close returns from Aug. 17 through Aug. 21 capture the initial separation by pairing Bitcoin’s rally with QQQ as a broad technology-equity reference for the same sessions.
The longer calculation used rolling 90-trading-day correlations and univariate Bitcoin betas, with one comparison ending Aug. 22, 2025, and the current one ending Aug. 24, 2026.
A Bitcoin beta of 1.10 means that a 1% daily Bitcoin move was associated with an average 1.10% move in miners during that window, while correlation measures how consistently their directions and magnitudes traveled together.
Bitcoin beta declined from the comparable 2025 window for six companies, with IREN holding near 0.93, while Bitcoin correlation fell for six and edged up for WULF from an already low 0.17 to 0.22.
QQQ correlation exceeded Bitcoin correlation for all seven companies in the current window, meaning their daily returns tracked the Nasdaq proxy more consistently than the coin they mine.
MARA retained the group’s highest Bitcoin correlation and beta, matching its heavier dependence on mining economics.
HUT, WULF, and CIFR occupy the bottom half of current Bitcoin correlation as their data-center contracts carry more weight in investor estimates, while MARA occupies the top. IREN breaks the simple ordering because its Bitcoin beta stayed steady even as its QQQ correlation reached 0.60, allowing an operating mining base and a large AI pipeline to appear in the same return series.
CryptoSlate also ran a 2026 year-to-date regression that included daily Bitcoin returns, QQQ returns, and the daily move in the 10-year Treasury yield. The three-factor model explained roughly 28% to 45% of daily variation across the seven miners, while the estimated effect of a 10-basis-point yield increase ranged from a 0.52% decline for WULF to a 0.79% gain for CIFR.
Four rate coefficients were negative, and three were positive, so the sample doesn’t support a common duration trade across the group.
Mixed coefficients block an easy bond analogy because higher yields reduce the present value of cash flows expected years from now and raise project financing costs, while daily equity returns also absorb tenant announcements, construction updates, Bitcoin holdings, power prices, and capital raises.
Contracts explain why the businesses carry different exposures, while the data provide no common rate trade across the group.
Contracts give the miners a second risk map
TeraWulf is the best example of a company whose income statement has already crossed into a new category. High-performance-computing leases supplied about 71% of second-quarter revenue, and its filings describe repurposing or curtailing mining equipment as computing capacity expands.
A Bitcoin rally can improve the residual mining operation, while the equity valuation now also depends on tenant payments, construction delivery and the financing attached to the campus.
Cipher shows how the stock-market identity can move before the revenue mix catches up. Its June quarter still reflected mining, though investors can model contracted computing capacity and the company-estimated $793 million of average annual net operating income associated with its base lease terms.
The gap between current revenue and promised capacity makes delivery dates, capital cost, and counterparty quality central inputs for the share price.
Riot and IREN carry both sets of exposures more visibly, although their current revenue mixes have separated. Mining supplies most of Riot’s current revenue, while IREN’s June quarter put AI cloud revenue just above Bitcoin mining revenue.
Their mining operations retain asset-price sensitivity, and signed AI agreements add another stream of projected cash flows. Each quarterly filing can shift the weight on Bitcoin production versus data-center delivery, making beta an output of the business mix rather than a permanent company trait.
Contract announcements carry plenty of uncertainty because base-term value represents payments expected across many years, revenue and net operating income estimates depend on timely delivery, and project-level debt protects a parent balance sheet only within the terms of its structure.
A multiyear headline value can help classify the company’s direction, though it can’t substitute for a discounted cash-flow model or completed capacity.
The Aug. 17 breakout captured a real separation, and the longer sample confirms a broader sector reclassification. MARA rose alongside the coin, every company retained a positive current Bitcoin beta, and mining cash flow still funds or supports several AI buildouts.
Bitcoin has become one factor among several, with its weight lowest where contracted computing capacity has become the center of the equity case.
“Bitcoin miners” now describe these companies’ origin more reliably than their destination. Investors buying the group through a mining basket can receive exposure to Bitcoin production, hyperscaler credit, construction schedules, power-delivery risk, project finance and technology-equity multiples in different proportions.
The contracts help explain why the stocks separated, while the remaining Bitcoin betas show their old identity still travels with them.
