Hyperscale Data Sold Its Last 211 BTC at $76K, and Left Roughly $8M on the Table
GPUS finished dumping its bitcoin reserve on Nov. 8, closing out a $74.4 million exit that averaged about $66,800 a coin. The cash goes to debt and 10 megawatts of AI capacity in Michigan. Here's the math on what the trade actually cost, and what comes next on Nov. 23.
What happens when a public company spends a summer telling you it's building a $100 million bitcoin treasury, then sells every coin before Thanksgiving? You get Hyperscale Data. The NYSE American-listed firm, ticker GPUS, unloaded its final 211.56 BTC for about $16 million, according to the company's own release. Zero coins left that weren't mined in-house.
The last fill came in around $76,000 per coin, all wrapped by Nov. 8. And that's the version of the story where things went well.
The math on the way out
Zoom out and the full exit is uglier than the final print. Since late July, Hyperscale has sold roughly 1,113 BTC for about $74.4 million. Run the division and you land near $66,800 a coin.
Bitcoin was hanging around $74,200 when the numbers were being run. Mark that same 1,113 BTC to the current tape and you get about $82.6 million. So the company gave up somewhere in the neighborhood of $8 million by selling into weakness instead of sitting still.
Is $8 million a disaster for a company this size? No. But it's real money, and it's the exact gap every treasury company gets judged on when the cycle turns. You don't get to say "bitcoin is pristine collateral" in August and then quietly liquidate in November without the tape noticing.
Here's the sequence. Late July, Hyperscale reported holding more than 1,106 BTC and floated a $100 million digital asset treasury, with a stated goal of reaching parity between its bitcoin stack and its market cap. A week later it disclosed a first sale of about 150.5 BTC, leaving roughly 958.5 BTC as of Aug. 2. Its Ault Capital Group arm even bought about 15 BTC on the open market around then, which reads like a hedge against looking like a seller.
Four months later, the whole position is gone.
From a $100 million target to flat
Let's be fair to management. Hyperscale was never a clean bitcoin play. It's a pile of businesses wearing a trench coat. Sentinum runs the data center side. Ault Capital Group is a hybrid private equity and operating outfit with lending, defense, industrial and hotel exposure. That's not a pure miner. It's a holding company that happened to own coins.
But the reversal still matters, because it's the same movie playing out across the sector. Keel dumped its entire treasury in the third quarter for $192 million. Riot and Hut 8 sold thousands of BTC to fund AI buildouts and buybacks. MARA said Monday it had sold 13,597 BTC for roughly $1.1 billion after Sept. 30, with more selling planned to cover AI infrastructure, convertible note puts and share repurchases.
That's not a coincidence. That's a migration. The capital that once chased hash rate and cold storage now chases megawatts and interconnection queues. And ngl, the AI data center pitch is easier to sell to a board than a coin with 60% drawdowns.
Think about what bitcoin treasury companies actually were. A shortcut to BTC exposure for equity accounts that couldn't buy spot. That was the whole trick. When the premium dies, so does the reason to hold the coin instead of the building that produces cash flow. Hyperscale's CEO, Will Horne, made that trade explicit. Bringing Michigan capacity online is the fastest route to profitability, he said, and selling the reserve cuts exposure to price swings while freeing cash for debt and construction.
Hard to argue with on a spreadsheet. Brutal on a narrative.
The AI trade is eating the bitcoin trade
Anon, let me save you some gas fees. The bitcoin-for-AI swap is the most crowded trade in public crypto right now, and it's probably not done. Anyone with a power contract and a delisted mining rig is pivoting to compute. The ones sitting on a coin stack are staring at the easiest funding source they've got.
So who wins here? Depends on which side of December you're measuring.
Short term, Hyperscale gets $16 million of dry powder and less volatility on the balance sheet. Long term, it's betting the company on executing a data center build in a state that isn't exactly known for fast-grid interconnects or cheap power. That's a real operational risk, and it's one the coin never carried.
The people who lose are the shareholders who bought GPUS as a bitcoin proxy. They got a real estate and infrastructure buildout instead, wrapped in a company that also owns hotels. If you're holding this expecting BTC beta, you're holding the wrong thing.
Traders I've talked to keep circling the same two numbers. What's the debt balance now, and what does the remaining debt load cost per quarter? The release doesn't say how much the bitcoin proceeds retired, and it doesn't say what's still outstanding. That's the hole in the story, and it's a big one.
What to watch on Nov. 23
Two catalysts land on the same day, and both matter.
Hyperscale said it expects to file its third-quarter 10-Q on or before Nov. 23. It also pushed its virtual Investor Day off Nov. 12 to that same date, citing a newly proposed exchange offer for its Class A common stock. Management is expected to give real numbers on debt reduction and the December energization target. Watch the 10-Q for the debt schedule before you watch the deck. Filings don't spin.
Then there's December itself, when the company aims to energize the first 10 megawatts of critical IT capacity in Michigan. Ten megawatts is a start, not a business. If that slips a quarter, the whole AI thesis slips with it, and you're left holding a company that sold the bottom of its best asset to fund a construction site.
The broader tell is MARA. When the largest miner in the room says it'll keep selling to fund data center builds, every smaller treasury holder gets a permission slip. Expect more announcements in this mold through the fourth quarter. The bitcoin treasury era on public balance sheets is winding down, and the trenches don't sleep on this kind of rotation.
My read, and it's a bold one: the companies selling BTC for AI capacity in 2025 are going to look smart if power contracts hold and compute demand stays tight. They're going to look like the people who sold Amazon in 1999 if it doesn't. I aped into the AI-power side of this trade, so yes, I'm biased. But the debt number on Nov. 23 is the tell. If Hyperscale shows the coin sale knocked down a big chunk of what it owes, this pivot is live. If the debt line barely moves, they sold 1,113 bitcoin to build one building and it wasn't enough.
Not financial advice, but I'm watching the filing, not the press release.