MARA Q2 2026: The 91% Sell-Down and the AI Pivot That Hasn't Paid Yet

The number that froze me wasn't in the headline. It was buried in MARA Holdings' Q2 2026 filing: 2,213 BTC sold against 2,422 BTC mined. Ninety-one percent of everything they dug out of the ground, shipped straight to the market. That's not treasury management. That's survival mode. Then there's the March window. In a single gulp, MARA moved 15,133 BTC — roughly $1.1 billion. That's more than six quarters of mining output, flushed in weeks. The world's largest publicly-traded Bitcoin miner just walked its own vault to the exit door. I've been reading these confessions since 2021, when I was live-streaming CryptoPunks floor-price chaos from Buenos Aires. The LUNA collapse taught me that balance sheets are emotional documents. And this one? This one reads like a blink. Let me rewind the tape to understand how we got here. MARA used to be the poster child of the HODL-to-survive school of miner treasuries. Mine it, hold it, borrow against it — let the appreciating asset do the heavy lifting. During the 2024 ETF sprint, that playbook minted heroes. It's been annihilated. BTC's average price in Q2 2026 fell 28% year-over-year; the period-end price sat roughly 45% lower. Meanwhile MARA pushed hash rate up 22% to 70.3 EH/s — a massive engineering sprint — and production crawled forward just 3% to 2,422 BTC. Network difficulty ate the growth. From the peak to the pit, this is what miner math looks like when the music stops. The pain metric that kept me up: electricity cost per BTC hit $38,690. That's 53% of the average BTC price during the quarter. Stack labor, debt service, depreciation on top, and the full-cycle cash cost starts flirting with the coin itself. Based on my experience modeling mining operations, that's the danger zone where marginal machines start debating whether to unplug. Then the cuts: 15% of staff shown the door. EBITDA swinging from +$1.2 billion to -$360 million. The sector's biggest player just transitioned from printing money to burning it. And in Texas, where ERCOT power prices swing like a pendulum, that's a double-edged sword. One summer heat wave and the grid signals miners to shut off. One winter storm and the hedge books explode. I've watched mining CFOs age a decade in a single rolling blackout. Spot On Chain analysts have already named the dynamic: liquidating the Bitcoin treasury to fund operations creates a supply overhang on BTC itself. Chasing the alpha through the noise, the real signal isn't the hash rate. It's the treasury. Let me walk you through the balance sheet like I do for every miner I audit. 35,577 BTC held — about $2.1 billion — but down 29% year-over-year. That's not a bump; that's a drawdown of the strategic reserve itself. Of those coins, 9,270 BTC — 26% — are lent out or pledged as collateral. The company earned $4.3 million in quarterly interest by lending 4,742 BTC. Annualized, that's roughly 4.9%. Four-point-nine percent on your Bitcoin. That's not a "strategic reserve" yield. That's a money-market yield. MARA has internally reclassified Bitcoin from a fortress asset to a yield-bearing liquid instrument. The philosophy shift is quiet, but it's the loudest thing in this report. Now the operational math. Hash rate up 22%, output up 3%. On the surface, that makes no sense. But network difficulty rose faster than MARA's new machines could compensate. The cost per PH/s actually improved 4% — the engineers did their jobs — yet the electricity bill per BTC ballooned. The only explanation: the new hashrate is high-cost capacity. Those machines were plugged in for a bull market that never showed up. Compare that to CleanSpark, which earned its reputation buying the newest machines at the deepest discounts, or Riot, which stuck to developing its own land and power. MARA's acquisition path bought speed, but it also bought the most expensive electrons in the business. And here's a read I can't shake: part of that 70.3 EH/s is probably already reserved for AI/HPC workloads, not SHA-256 mining. The company explicitly says it's "continuing to channel more capital toward AI and high-performance computing." That's a datacenter story wearing a miner's hat. The land story is genuinely interesting. MARA controls up to 4.8 GW of potential power capacity — Matagorda County, the Long Ridge acquisition — and the entire sector is doing the same dance. Core Scientific signed its AI contracts first, locking in CoreWeave's rent. MARA is still chasing that proof point. Riot, CleanSpark, all of them are pivoting from coin-printing to compute-renting. Breaking silos, one block at a time. But here's the tension: AI revenue is not on the income statement yet. The $174.9 million top line is still 97% mining-dependent. $4.3 million in lending interest is the only non-mining income, and that's 2.4% of revenue. The market is being asked to value MARA as an AI-infrastructure company while it operates as a forced seller of Bitcoin. If BTC keeps sliding, those counterparties on the 26% collateralized stack have rights. Margin calls. Forced sales. The same dynamic that sank over-leveraged funds in 2022 could reach into MARA's books this time. The conventional read: MARA is transforming from a commodity miner into diversified compute infrastructure. Buy the transition. I get it. But let me offer the uncomfortable side. Hype, heartbeats, and hard data. The hard data says these BTC sales are procyclical. Price drops → MARA sells more to cover debt and operating costs → supply pressure pushes price down further. That's a feedback loop, and it doesn't stop at MARA. The sector-beta effect is real: when the flagship miner reports a loss and starts firing people, fund managers re-rate every mining stock on their watchlist. MARA's bad news becomes Riot's discount, Core Scientific's headache. The whole group trades as one giant high-beta bet on BTC these days. Here's the blind spot nobody on Crypto Twitter is talking about: when the largest public miner lends out its own Bitcoin for yield, it quietly erodes the "digital gold" thesis. Every 4.9% interest payment is a small admission that Bitcoin is just another carry asset. That's the kind of slow, structural sentiment shift that doesn't show up in the earnings call but shapes institutional allocation for years. And that March sale — 15,133 BTC in one window — wasn't an accident. It was a decision to de-risk before the door closed. The stock trades with daily swings of 5-15%; this report bundles a loss, a selldown, layoffs, and a pivot narrative all at once. The market will decide which story to price. Watch Q3 like a hawk. Three signals: Does MARA sell more than it mines again? Do named AI clients sign leases on that 4.8 GW? Does BTC hold above that $38,690 electricity breakeven? If marginal machines shut down, hashrate drops, difficulty adjusts, and the survivors win. If AI leases land, MARA becomes something genuinely new. If neither happens — and the selling continues — the biggest miner in the world becomes the market's most visible source of supply. The race isn't to the fastest miner anymore. It's to the last one still holding.