A week ago, AMD's blowout earnings lifted neocloud and AI infrastructure stocks across the board. The read-through for Bitcoin miners pivoting to AI data centres was clear: if AI compute demand is as strong as hyperscalers are signalling, the miners racing to repurpose their power assets may be timing the cycle correctly.
The earnings from MARA and CleanSpark reported Monday offer the first detailed look at how that transition is actually progressing — and how much it is still costing.
MARA reported Q1 2026 revenue of $174.6 million, down 18.3% year-over-year, and a net loss of $1.3 billion. Approximately $1 billion of the Q1 loss came from an unrealized mark-to-market adjustment on digital assets tied to Bitcoin's price decline during the quarter — not a cash loss, but a real accounting hit under the FASB fair value rules the company adopted this year. Adjusted EBITDA was negative $1 billion, also dominated by the Bitcoin mark-to-market. EPS came in at -$3.31 against analyst expectations of around -$1.41, a substantial miss on the headline number.3
Operationally, the picture is more constructive. MARA delivered a record energised hash rate of 72.2 exahash per second in Q1, up 33% from 54.3 EH/s in Q1 2025, and its share of available mining rewards increased to 5.5% from 4.8% the prior quarter. The company also made significant balance sheet moves: it sold approximately $1.5 billion of Bitcoin during the quarter, using proceeds to repurchase more than $1 billion of convertible notes at a discount and reduce its credit line by $200 million. The debt retirement reduced potential dilution by as many as 46 million shares, or roughly 9% on a fully diluted basis. MARA has not used its at-the-market equity programme since the end of Q3 2025, a deliberate choice to fund operations through Bitcoin monetisation rather than equity dilution.
The strategic headline is the Long Ridge acquisition. MARA announced on 30 April a definitive agreement to acquire Long Ridge Energy & Power from FTAI Infrastructure — a deal that would increase its owned and operating capacity by roughly 65%, from approximately 1.3 gigawatts to 2.2 gigawatts at closing, and 2.4 gigawatts including expansion capacity. The company's current plan for Long Ridge includes an initial 200-megawatt AI build-out, with construction beginning in H1 2027 and initial capacity expected online in mid-2028. CEO Fred Thiel said MARA is already in advanced discussions with prospective tenants across multiple sites. The existing Bitcoin mining operations at Hannibal will continue until the data centre campus requires the power — a sensible hedge given BTC's recovery since quarter end.
CleanSpark's Q2 result told a similar story from a different position. Revenue from Bitcoin mining was $136.4 million, down 24.9% from $181.7 million a year ago, with a net loss of $378.3 million driven by a $224.1 million fair value loss on Bitcoin, increased depreciation and amortisation of $115.9 million, and higher operating expenses. Shares fell 6% in after-hours trading following the release, having missed analyst consensus on both revenue ($136.4 million versus $152.3 million expected) and EPS (-$1.52 versus -$0.41 expected).
The operational metrics were stronger. Power costs declined to 5.2 cents per kilowatt hour from 6 cents a year earlier, a meaningful efficiency gain that supports margins in a depressed revenue environment. Megawatts under contract roughly doubled year-over-year, including 585 MW of ERCOT-approved capacity, and post-quarter figures showed 50.0 EH/s of operational hashrate and 640 BTC mined in April alone. The balance sheet remained solid: working capital of $1.0 billion as of March 31, $260.3 million in cash, and Bitcoin holdings valued at $925.2 million.
On the AI front, CleanSpark CEO Matt Schultz outlined progress across four areas: land and power development with ERCOT approval of 300 MW in Brazoria; leasing progress in Georgia; constructive financing conditions; and ongoing construction at the new Sandersville parcel. CFO Gary Vecchiarelli was pointed about the direction of travel: "CleanSpark is no longer a single-track business. Bitcoin mining generates the cash flow, AI infrastructure monetises the assets over the long term, and our Digital Asset Management function optimises capital and liquidity across cycles."
The message from both companies is structurally identical: the mining business is cash flow while the AI build-out is the long game, and the Q1 numbers reflect a quarter when Bitcoin was down sharply rather than any deterioration in the underlying strategy. That framing is fair as far as it goes. The more honest version of the same narrative is that CleanSpark and MARA are building AI capacity without yet having the customers to fill it — the AI premium in both stocks is currently being priced on potential rather than contracted revenue. Long Ridge gives MARA a credible pathway to prove the thesis. CleanSpark's 1.8 gigawatts under contract and its ERCOT approvals give it real optionality. What neither company has yet shown is a signed hyperscale or enterprise AI lease.
The market's reaction to both sets of results was more nuanced than the headline misses might suggest. CLSK closed Monday at $14.30, up $0.10 on the day despite the EPS miss — a sign that investors came in expecting the worst and found the operational picture credible enough to hold. MARA, which reported after market hours, was trading at $13.79 by end of day Monday, up from a close of $12.94 the prior session — a 6.6% gain that came partly on the back of Bitcoin's own recovery above $80,000 during the same period. Both stocks remain well off their 52-week highs — MARA's range over the past year spans $6.66 to $23.45, while CLSK has traded between $7.82 and $23.61 — reflecting just how much ground the miners have to recover even as BTC stabilises.
The AMD earnings showed the demand side of the AI infrastructure equation is real and accelerating. What MARA and CleanSpark need to show, in Q3 and beyond, is that the supply they are building is landing with paying tenants. For now, the market appears willing to price that possibility in — but the patience has a timer on it.
