We didn’t see the shareholder revolt coming. On paper, a $9 billion acquisition offer from CoreWeave seemed like a lifeline for a company that emerged from bankruptcy just last year. Core Scientific, the Nasdaq-listed (CORZ) Bitcoin mining giant turned AI infrastructure hopeful, had every reason to take the money and run. But the shareholders said no. And then, just days later, the company announced a partnership with AMD to supply GPUs for AI workloads. The market cheered. I frowned.
Let me be clear: I’m not a cynic by nature. But after 29 years in open-source and blockchain infrastructure, I’ve learned that the biggest risks hide in the most celebrated press releases. We didn’t see the 2017 ICO insider allocations until my audit team exposed them. We didn’t see the 2020 DeFi liquidity mirages until the crash. And today, I’m worried that the crypto community is missing the real story behind Core Scientific’s pivot.
Context: The Infrastructure Layer’s Identity Crisis
Core Scientific operates at the physical layer of crypto. They run Bitcoin mining rigs in massive data centers, consuming cheap power secured through long-term contracts. But the mining industry has been brutal since the 2022 bear market and the 2024 halving. Margins are squeezed. Hashrate is consolidating. So, like many miners, Core Scientific is trying to repurpose its infrastructure for AI—specifically, hosting high-performance computing (HPC) and GPU clusters for AI workloads. This isn’t new. Bit Digital, Hut 8, and others have done the same. But Core Scientific’s scale is different: they have over 1,000 megawatts of capacity, mostly in the U.S., with access to low-cost power.
The AMD partnership announcement was the capstone of this narrative. Core Scientific would integrate AMD’s Instinct GPUs into its data centers, offering AI cloud services to enterprise clients. The stock jumped 15% on the news. The narrative was simple: “We’re no longer just a miner; we’re an AI infrastructure play.” But as an open-source evangelist and a financial engineer, I know that narratives are cheap. Execution is everything.
Core: The Technical Reality Behind the AMD Partnership
Based on my audit experience with blockchain infrastructure projects, I’ve learned to scrutinize three things: power, networking, and software stack. The AMD partnership, as described, addresses none of these in detail.
First, power. Core Scientific’s mining sites are designed for high-density ASIC rigs, which are air-cooled and consume 3-5 kW per unit. AI GPUs like the AMD Instinct MI300X require 700W per GPU, but a full rack of eight GPUs plus networking can exceed 15 kW. That’s a 3x to 5x increase in power density. Without liquid cooling or significant infrastructure upgrades, those racks will melt. Core Scientific has said it will retrofit its sites, but the cost is substantial—millions per megawatt. The company’s balance sheet, still recovering from bankruptcy, may not support this without additional debt or equity dilution.
Second, networking. Bitcoin mining is compute-intensive but not latency-sensitive. You can batch blocks and send them over the internet. AI training, on the other hand, requires low-latency, high-bandwidth interconnects like InfiniBand or at least high-speed RoCE (RDMA over Converged Ethernet). Core Scientific’s existing network infrastructure is designed for relatively simple data flows. Upgrading to a GPU-cluster network requires new switches, fiber, and expertise. The company hasn’t disclosed its networking plans. We didn’t get a technical whitepaper or a pilot deployment result. We got a press release.
Third, the software stack. AMD’s ROCm is the open-source alternative to Nvidia’s CUDA. I’ve used both. ROCm is improving, but it still lags in library support, debugging tools, and community adoption. Many AI frameworks (PyTorch, TensorFlow, JAX) have CUDA as the default backend. ROCm requires custom builds or containers. For an enterprise client, the cost of switching is not just the GPU price difference—it’s the engineering time to port and optimize models. Core Scientific is effectively betting that they can make ROCm easy enough for their clients. That’s a big bet.
The Hidden Value: The Power Contract Arbitrage
There is one genuine advantage Core Scientific has: long-term power purchase agreements (PPAs) at below-market rates. In a bear market, power is the most critical input. AI training is energy-intensive, and cloud providers like AWS, Google, and Azure charge premium prices for compute. If Core Scientific can offer a 20-30% discount on GPU compute by leveraging cheap power, they could attract price-sensitive customers, such as academic research labs, startups, or even decentralized AI projects.
But even this advantage is fragile. Power prices are volatile. During the 2022 energy crisis, many miners had to curtail operations. More importantly, the AMD chips themselves are not cheap. The partnership likely involves a volume purchase agreement, not a revenue-sharing deal. Core Scientific has to buy the GPUs upfront, then amortize the cost over time. If utilization is low, the losses could be significant.
Contrarian: The $9 Billion Rejection Was a Trap
Here’s the counter-intuitive angle: The shareholders might have made a mistake by rejecting the $9 billion acquisition. I know that sounds heretical in a market that values independence. But consider the numbers.
Core Scientific’s market cap before the rejection was around $1.5 billion. The offer was $9 billion—a 6x premium. By rejecting it, the shareholders are essentially saying, “We can build a company worth more than $9 billion.” But building an AI infrastructure business from a mining base is capital-intensive. The company will need to raise billions in debt or equity to retrofit its sites, buy GPUs, and hire engineering talent. Dilution is almost certain. The 15% stock jump after the AMD announcement was a short-term reaction, not a fundamental valuation change.
More importantly, the AMD partnership is a signal of desperation. Nvidia controls over 80% of the AI GPU market. AMD is a distant second. Core Scientific could have partnered with Nvidia, but they chose AMD. Why? Because Nvidia might have demanded exclusivity or high margins. AMD, on the other hand, needs real-world deployments to prove its chips are viable. Core Scientific is essentially a guinea pig for AMD’s enterprise ambitions. That’s a risky position.
We didn’t think about the execution risk. We didn’t ask about the software stack. We didn’t question the timing—why announce a partnership with no technical details during a bear market when every dollar counts?
Takeaway: The Only Metric That Matters Is Delivered Power
If I’ve learned anything from the 2022 bear market, it’s that survival matters more than gains. In this environment, investors should not be swayed by partnership announcements. They should look for concrete milestones: How many megawatts of AI-capable data center capacity has Core Scientific actually delivered? What is the utilization rate? Are there any reference customers? Have they published any benchmark results for AI training on their AMD clusters?
Until we see those numbers, the AMD partnership is just a story. And in my experience, stories without data are the most dangerous assets in crypto.
I’ll be watching Core Scientific closely. Not because I don’t believe in the pivot—I do believe that mining infrastructure can be repurposed for AI. But I’ve seen too many projects promise the future and deliver only hype. We need to demand transparency, not just press releases. We need to champion protocols that prioritize human well-being over market narratives.
The question is: Will Core Scientific become a bridge between mining and AI, or will it be another casualty of the hype cycle? We don’t have the answer yet. But we know what to look for. And we won’t stop asking.