Core Scientific’s $9B Rejection: The AMD Hype Is Code Without Execution

The shareholders voted no. A $9 billion exit—gone. Core Scientific’s board must now prove they can build something worth more than that. Their answer? An AMD partnership. But the code didn't execute. The data didn't verify. The market is still waiting for the proof. Let’s rewind. Core Scientific is a Bitcoin miner. It runs sprawling facilities packed with ASICs, sucking down cheap power under long-term contracts. That’s its core asset: not the chips, but the electrical infrastructure. The pivot to AI hosting is a natural play—repurpose those same megawatts for GPU clusters. It’s the same story we’ve seen from Hut 8, Riot, and others. The difference here is the scale of the rejection. Shareholders looked at a $9 billion offer and said, “We can do better.” That’s a bold bet. It implies the market is undervaluing the company’s AI potential. But potential is not delivery. Now the AMD partnership. On paper, it’s a strategic move: diversify away from Nvidia’s monopoly, secure a supply of GPUs for AI workloads. But here’s where the forensic skepticism kicks in. The announcement is a press release, not a technical report. No mention of how many MW of AI capacity will be deployed. No details on the ROCm software stack integration. No benchmarks for inference or training throughput. The code didn’t run. The data didn’t appear. The “truth” is not mined; it is verified in watts delivered. I’ve spent years tracking the intersection of mining and compute. I’ve seen the engineering challenges firsthand. Converting a mining facility to an AI data center isn’t just plugging in different hardware. It’s a complete re-architecture. Mining rigs are air-cooled, low-density, and tolerant of variable power. AI clusters need liquid cooling, high-density racks, InfiniBand networking, and redundant power with sub-millisecond failover. The AMD Instinct GPUs require ROCm, which still lags CUDA in software maturity. The partnership will likely involve joint engineering—AMD needs real-world deployment sites to validate its hardware against Nvidia. That’s a benefit, but it’s also a risk: if AMD’s chips don’t perform, Core Scientific is left with idle capacity. Volume was a ghost. The market rallied on the announcement, but the on-chain truth—the actual operational metrics—is missing. Core Scientific has not disclosed its current AI hosting revenue, utilization rates, or the specific terms of the AMD deal. Is it a purchase agreement, a revenue share, or a joint venture? We don’t know. The $9 billion rejection sets a valuation anchor: shareholders believe the company is worth more than that. But anchoring is not trading. The market will now watch every quarterly report, every MW of capacity delivered, every GPU order filled. If the company fails to execute, the stock will drift lower than the rejected offer. Here’s the contrarian angle. The real value isn’t the AMD partnership. It’s the power contracts. Core Scientific locked in cheap electricity before the AI boom. Those long-term power purchase agreements are the true moat. Any GPU can be bought, but cheap power is scarce. The market is fixated on the chip supply narrative, but the engineering bottleneck is cooling and power delivery. The AMD partnership is a distraction from the core thesis: the company is a real estate play on stranded energy. I’ve audited similar transitions. In 2022, I analyzed a mining firm that tried to pivot to AI hosting. They spent $50 million on GPUs, but their power infrastructure couldn’t support the density. The project failed. The lesson is simple: kilowatts are truth. Code is law, but logic is justice. Until Core Scientific publishes its power utilization metrics for AI workloads, the AMD partnership is just a story. Takeaway: ignore the hype. Watch the MW. The next quarterly report must show actual AI capacity delivered. If it doesn’t, the $9 billion rejection will look like a mistake. The market is in a sideways chop, waiting for direction. Core Scientific has the potential to break out—but only if the code executes. The data didn’t lie. The code didn’t run. The truth is still waiting to be mined.