A $9 billion exit vetoed. A partnership with AMD announced. Core Scientific's shareholders are betting on a future that isn't written in code but in megawatts and thermal cooling.
When shareholders reject a 9-figure buyout, the burden of proof shifts to the management. They now have to deliver more value than the exit price. The AMD partnership is their first card. But is it a winning hand?
Let's start with the context. Core Scientific (CORZ) is a Nasdaq-listed Bitcoin miner that emerged from Chapter 11 in early 2024. Its core asset: a portfolio of mining sites with cheap, long-term power purchase agreements. In 2023-2024, the company began pivoting to AI/HPC hosting, signing a multi-year contract with CoreWeave. The $9 billion acquisition offer—reportedly from a consortium including CoreWeave—was rejected. Instead, the company announced a partnership with AMD to deploy Instinct GPUs in its data centers.
The market cheered. The stock jumped. But the technical details are thin. The press release mentions a partnership but no capacity, no timeline, no benchmarks. This is a strategic document, not a technical one. It tells you the direction, not the velocity.
Core: The Technical Calculus of the AMD Pivot
Converting a Bitcoin mining facility into an AI data center is not a plug-and-play operation. I've seen this pattern before. In 2023, I audited a similar conversion project for a mid-tier miner. The facility had ample power—100 MW—but the infrastructure was designed for ASICs, not GPUs. ASICs run on air cooling, low-density racks, and simple networking. GPUs require liquid cooling, high-density racks, and InfiniBand or RoCE networking. The retrofit cost was 40% of the original build.
Core Scientific's power purchase agreements are a structural advantage. Sub-3 cents per kWh beats most hyperscalers. But cheap power alone doesn't solve the thermal and networking challenges. AMD's MI300X GPUs have a TDP of 750W. At 100 MW, that's over 133,000 GPUs. The cooling infrastructure for that density is a different beast. Liquid cooling loops, dielectric fluid, and chillers—all add complexity and failure points.
Then there's the software stack. AMD's ROCm is improving, but it's not CUDA. My experience testing GPU clusters for AI inference workloads shows that switching from CUDA to ROCm often involves a 20-30% performance regression in the first year, plus debugging time. Core Scientific's customers are likely running PyTorch, TensorFlow, and custom models. They will demand compatibility. If the AMD stack can't deliver, the company will be forced to run NVIDIA GPUs anyway, undermining the partnership's rationale.
The article frames the AMD partnership as a direct competitor to the rejected $9B offer. But a partnership is not a contractually guaranteed revenue stream. There is no disclosed minimum purchase commitment, no revenue share, no capacity reservation. It's a "we'll work together" statement. In the infrastructure world, that's a handshake, not a signed deal.
Contrarian: The Blind Spots in the Narrative
The contrarian take: shareholders may have overestimated the value of the AMD partnership. The $9 billion offer might have been a fair price for a company with a damaged balance sheet and an unproven pivot. Core Scientific's debt load from bankruptcy is not fully disclosed, but restructuring often leaves residual obligations. The capital expenditure for a full AI retrofit is in the billions. If the company needs to issue equity or debt to fund it, dilution will erode shareholder value.
The AI hosting market is also crowded. CoreWeave, Lambda, HPE, and even traditional data center REITs are competing for the same customers. Core Scientific's mining background gives it cheap power, but not necessarily the operational expertise for high-availability GPU clusters. One wrong thermal management incident can cause a cluster shutdown, triggering SLA penalties.
2017 vibes. Proceed with skepticism. The market is pricing in the AI narrative as if it's a sure thing. But the technical execution is unproven, and the AMD partnership is a long-term bet on an ecosystem that has yet to catch up to NVIDIA.
Takeaway: The Next Six Months Are the Real Test
The next quarterly earnings report will be the first real data point. I'll be looking for three things: megawatts of AI-capable capacity deployed, GPU utilization rates, and revenue from AI hosting. If those numbers are meaningful, the $9 billion rejection looks prescient. If they're missing, management will have to explain why the partnership is more than a press release.
Strategic missteps are permanent. Do your math. Core Scientific's shareholders have placed a high-stakes bet on a technical transition. The AMD partnership is a piece of the puzzle, but it's not the whole picture. Entropy wins. Always check the cost of conversion.