CleanSpark has an AI tenant. MARA has a pipeline. Both have shrinking mining revenue.

Two of the largest listed bitcoin miners reported their June quarters on Thursday, and the numbers went the same direction: down. MARA Holdings booked second-quarter 2026 revenue of $174.9 million, off 27% from $238.5 million a year earlier. CleanSpark, whose fiscal third quarter covers the same April-to-June window, reported $138.0 million, down 30.5% from $198.6 million, according to The Block's coverage of both releases.

The losses were larger than the revenue decline suggests. MARA's net loss came to $611.3 million, or $1.60 per diluted share, against net income of $808.2 million a year earlier — a swing driven in part by a $343 million fair-value loss on its digital assets. Adjusted EBITDA went from a $1.2 billion profit to a $360.9 million loss. CleanSpark lost $239.8 million, or $0.89 per basic share, versus $257.4 million of net income a year prior, with a $116.3 million fair-value loss on its bitcoin and adjusted EBITDA of negative $113.0 million. Combined: roughly $851 million of red ink in three months.

The mechanical cause is visible in MARA's own operating stats. Energized hashrate rose 22% year over year to 70.3 EH/s and cost per petahash per day improved 4% to $27.7 — more machines, running more efficiently — while revenue still fell 27%. MARA mined 2,422 BTC at an average price of about $71,325 and sold 2,213 of them at an average of $73,078, per The Block. That the quarter closed with large fair-value writedowns implies bitcoin ended the period well below those averages; CleanSpark's 13,924 BTC were carried at $814.9 million on June 30, or roughly $58,500 a coin. The Block's news wire on Friday carried a separate headline putting bitcoin above $65,000 after a US jobs miss.

The Bernstein split

A day later, on Friday, analysts at Bernstein led by Gautam Chhugani published paired notes that The Block reviewed: Outperform on CleanSpark with a $24 price target, Market-Perform on MARA with a $17 target. Bernstein's stated reason was execution, not strategy — CleanSpark has a signed anchor tenant and a build underway; MARA has not yet landed a single commercial AI contract.

CleanSpark's contract is a 20-year, $6.6 billion triple-net lease covering 175 IT MW at its Sandersville site in Georgia, with what the company describes only as a "high investment-grade global technology company" as tenant. The tenant is not named. The lease carries an exclusivity agreement over CleanSpark's entire 885 MW Texas portfolio, and CleanSpark has hired the tenant's preferred engineering and construction firm for the Sandersville build — a step Bernstein said de-risks the deployment. The first data hall is expected in the fourth quarter of 2027.

The financing is not closed. Bernstein put project-level capex at roughly $1.9 billion, derived from CleanSpark's own guidance of $10 million to $12 million per IT MW, and said the equity portion is funded while the company seeks secured project financing at a 90% loan-to-cost ratio. Bernstein cited about $900 million of liquidity as of June 30; CleanSpark's release listed $202.6 million in cash plus $814.9 million of bitcoin, against total assets of $2.7 billion, long-term debt of $1.8 billion and working capital of $761 million.

There is a regulatory overhang. A Texas audit requirement for data center projects is expected to delay ERCOT's Batch Zero process, with a Public Utility Commission of Texas hearing set for Aug. 20. CleanSpark's management told Bernstein it does not expect the delay to move the energization timeline for its 285 MW Sealy site (mid-2027) or the 300 MW first phase at Brazoria, both of which it expects in ERCOT's base load for Batch Zero.

MARA's case is a pipeline. Bernstein said recent transactions could lift its planned power portfolio to 4.2 GW, including the 2 GW Matagorda County site in Texas bought for $600 million — with the purchase price tied to development milestones — and the pending 1 GW Long Ridge acquisition in Ohio, which needs Federal Energy Regulatory Commission approval. MARA says Long Ridge will deliver immediate positive EBITDA and expand AI capacity at its Hannibal campus once cleared. Its partnership with Starwood is meant to pair MARA's power with Starwood's engineering and construction arm and more than 7 GW of delivered infrastructure; the two are working through lease discussions across multiple sites. Management reiterated confidence in signing at least two AI leases by year-end.

Bernstein values both on a sum-of-the-parts basis: roughly 8x EV to estimated 2027 adjusted EBITDA for the mining business, bitcoin holdings marked at current prices, and $3 million per megawatt for incremental power available for AI. That last input is why the power-capacity numbers matter more than they look.

Key facts

• MARA Q2 2026 revenue $174.9M, −27% YoY; net loss $611.3M ($1.60/diluted share); adjusted EBITDA −$360.9M — The Block, from MARA's results
• CleanSpark fiscal Q3 revenue $138.0M, −30.5% YoY; net loss $239.8M ($0.89/basic share); adjusted EBITDA −$113.0M — The Block, from CleanSpark's results
• MARA mined 2,422 BTC (avg ~$71,325), sold 2,213 (avg $73,078); hashrate 70.3 EH/s, +22% YoY; cost/PH/day $27.7 — The Block
• MARA treasury 35,577 BTC (~$2.1B), down 29%, fourth-largest public holder; CleanSpark 13,924 BTC, eleventh — Bitcoin Treasuries via The Block
• CleanSpark Sandersville lease: $6.6B, 20 years, triple-net, 175 IT MW, unnamed investment-grade tenant; first data hall Q4 2027 — CleanSpark statement and Bernstein via The Block
• Bernstein: CLSK Outperform, $24 target; MARA Market-Perform, $17 target; CLSK closed $12.75 and MARA $10.65 on Aug. 6 — Bernstein notes via The Block
• PUCT hearing on the Texas data center audit requirement: Aug. 20 — Bernstein via The Block

The real-world read

The two Block stories disagree on MARA's power number. Thursday's earnings piece said the Matagorda site "could push its power portfolio toward 4.8 GW"; Friday's Bernstein piece said recent transactions could lift the planned portfolio to 4.2 GW. That is a 600 MW gap, and on Bernstein's own $3 million-per-MW input it is worth about $1.8 billion of implied value. Neither figure is contracted capacity — both describe planned power, and 1 GW of it is contingent on FERC.

A stated comparison in the Bernstein writeup doesn't compute. The Block described CLSK's $12.75 close as "an 88% discount to Bernstein's target" and MARA's $10.65 as "60% below its $17 target." Those percentages are the upside from price to target, not the discount from target to price: $12.75 is about 47% below $24, and $10.65 is about 37% below $17. The direction of the call is unchanged; the framing overstates the gap.

"$6.6 billion" is the headline number CleanSpark wants read, and it is a 20-year gross figure. Averaged, it is about $330 million a year, against roughly $1.9 billion of upfront project capex on Bernstein's estimate — and the first data hall isn't due until Q4 2027. The debt to build it isn't raised yet; a 90% loan-to-cost package is being sought. President Gary Vecchiarelli's line about converting "infrastructure optionality into durable cash flows and long-term shareholder value" is a company officer's characterization of a business that lost $239.8 million in the quarter he was describing.

The exclusivity clause cuts both ways. One tenant now has exclusivity over all 885 MW of CleanSpark's Texas portfolio. That is presented as validation; it also means the optionality Vecchiarelli invokes has been sold to a single counterparty that CleanSpark won't name. Investors are being asked to underwrite a credit they can't see.

MARA sold 91% of what it mined. For a company still marketed as the fourth-largest corporate bitcoin holder, a 29% treasury drawdown alongside 2,213 coins sold out of 2,422 mined is the more informative fact. Miners sell when mining economics compress; the fair-value loss and the sales are the same story.

Two claims sit unresolved. CleanSpark's management says a regulatory delay it acknowledges exists won't change its energization dates — the Aug. 20 PUCT hearing, not the company, will settle that. And MARA's "at least two AI leases by year-end" was reiterated, meaning it was promised before and hasn't happened yet. Bernstein's Market-Perform is, in effect, a decision not to pay for that promise until it lands.

Note the reading environment. The Block pages carrying this reporting also carry paid placements — an LMAX Digital promotion and a Polymarket unit — which are advertising, not reporting. The Block also discloses that Foresight Ventures is its majority investor and that Bitget is an anchor LP for Foresight.

Opinion, and whose

• Bernstein (Gautam Chhugani and team): CleanSpark Outperform, $24; MARA Market-Perform, $17. Execution — a signed tenant versus none — is the differentiator. A first commercial AI contract "could be a re-rating catalyst" for MARA.
• Bernstein, on method: ~8x EV/2027E adjusted EBITDA for mining, bitcoin at market, $3 million per MW for incremental AI-available power. These are assumptions, not observations, and small changes to the per-MW figure move the targets materially. Bernstein is a sell-side broker; ratings are its product.
• CleanSpark's Gary Vecchiarelli: the company holds "scarce, grid-connected power assets and multiple pathways to commercialization."
• MARA's Fred Thiel: "Bitcoin mining provided the foundation," with digital infrastructure and Exaion positioning MARA "across multiple layers of the AI infrastructure value chain."
• MARA management: at least two AI leases signed by year-end. Unverified forecast from an interested party.

Sources

• The Block, Naga Avan-Nomayo (Aug. 7, 2026) — the Bernstein ratings, price targets, Sandersville lease mechanics, capex and financing assumptions, ERCOT/PUCT timeline, MARA's 4.2 GW pipeline figure and Bernstein's valuation method, based on notes The Block reviewed.
• The Block, Daniel Kuhn (Aug. 6, 2026) — MARA and CleanSpark quarterly results, balance sheet and production detail, executive quotes, the 4.8 GW figure, and Bitcoin Treasuries holdings rankings.
• Bernstein research notes (Aug. 7, 2026) — primary source for the ratings and targets; not read directly here, quoted via The Block. Sell-side research: an interested party in the equities it rates.
• Company statements from CleanSpark and MARA — quoted via The Block; the Vecchiarelli and Thiel quotes are corporate communications and are treated as such.
• Bitcoin Treasuries — holdings and rankings, via The Block.
• Not used as reporting: the LMAX Digital and Polymarket placements on The Block's pages, which are advertising.

Nothing here is investment advice.