Shareholders just said no to $9 billion. That's a statement. But is it a smart one?
Core Scientific's board wanted to sell. The offer was real. The shareholders blocked it. Their justification? The new AMD partnership. A pivot from mining to AI. The market cheered. But I've been in this game long enough to know that press releases don't compute. The $9 billion offer was a yield that was too good to be true. So they didn't take it.
Let's rewind. Core Scientific is a Bitcoin miner. They filed for Chapter 11 in 2023. They emerged with a plan: convert their massive power infrastructure into AI data centers. They signed deals with CoreWeave, a pure AI cloud player. Now they announce a partnership with AMD. Instinct GPUs. The narrative is clear: Core Scientific is becoming an AI infrastructure powerhouse. But the narrative is not the data.
Context: The Infrastructure Mirage
Core Scientific owns 700+ megawatts of operational power capacity. Most of it is still running ASICs for Bitcoin. The conversion to AI requires liquid cooling, high-density racks, InfiniBand or RoCE networking, and a software stack that supports GPU clusters. That's not a simple upgrade. It's a rebuild. According to my work with hedge funds tracking institutional crypto flows, the key metric is delivered megawatts for AI compute. Converted megawatts. Not total capacity. Not partnership announcements.
The AMD deal is a strategic move, sure. But the article that broke this news—the one I'm responding to—contains zero technical details. No wattage numbers. No utilization rates. No performance benchmarks. Just a line: 'Core Scientific partners with AMD.' That's not a technical milestone. That's a press release.
Core: The Technical Reality Check
I've audited mining facilities. I've seen the engineering challenges. Converting a warehouse full of S19s to a GPU cluster is not plug-and-play. The cooling requirements alone are a different beast. ASICs can run at 70°C. GPUs need liquid cooling to stay under 30°C for sustained AI workloads. That means retrofitting piping, installing CDUs, upgrading electrical panels. Every megawatt of AI capacity costs $2-3 million in capital expenditure. Core Scientific will need to raise more money. The shareholder vote against the $9B sale implies they believe management can create more value than that. But the cost of capital is high. Interest rates are still elevated. The stock market is volatile.
And then there's the AMD factor. AMD's Instinct MI300X is a capable GPU. But the software ecosystem is still chasing CUDA. ROCm is improving, but it's not mature. For AI workloads, that means compatibility issues, lower performance, and higher engineering overhead. Core Scientific, if they deploy AMD, will be taking on that risk. The pivot button is a lever, not a purchase. AMD chips are not a guarantee of revenue.
Compare to CoreWeave, which uses NVIDIA. Or to the big cloud providers. Core Scientific's competitive advantage is cheap power. They have long-term PPAs locked in. That's real. But cheap power doesn't make up for an inferior software stack. The clients that need AI compute want performance, not just cost savings. If Core Scientific can't deliver the same throughput as a NVIDIA cluster, they'll lose deals.
Contrarian: The Overhyped Partnership
The market is treating this AMD deal as a validation. The contrarian view: it's a sign of constraint. Core Scientific couldn't secure enough NVIDIA GPUs. AMD is a second choice. The partnership might be a joint engineering effort to optimize ROCm for their power infrastructure. That's fine. But it's not a revenue guarantee. The article mentions that the AMD deal 'exceeds the value of the rejected acquisition.' That's an opinion, not a fact. The acquisition offer was $9 billion. Core Scientific's current market cap is around $4 billion. The stock needs to double to justify that claim. That's a huge bet.
Volatility is just fear wearing a disguise. The stock price will tell the real story. Right now, the market is optimistic. But the fundamentals haven't changed. Core Scientific still has debt from the bankruptcy. They still need to spend billions on retrofitting. The AMD deal doesn't change that timeline. The shareholder vote sets a high bar. Management now has to deliver more than $9 billion in value. That's a lot of megawatts.
Takeaway: Watch the Watts, Not the Headlines
The next twelve months are critical. Core Scientific needs to deliver at least 300 MW of AI-ready capacity by the end of 2025. If they can do that, the bet pays off. If not, the shareholders will regret saying no to $9 billion. The real metric is not the partnership announcement. It's the power-on date. The utilization rate. The revenue per watt. Those are the numbers that matter. I'll be watching the data feeds, not the press releases. The $9 billion offer was a yield that was too good to be true. Now we'll see if the real yield is better.