Core Scientific and AMD: The Fine Print of a Pivot That Isn't About Mining

The numbers are straightforward, but the narrative requires disassembly. On September 4, 2024, Core Scientific announced a multi-year partnership with AMD. The headline: AMD acquires warrants to purchase Core Scientific common stock at the market price, and in return, Core Scientific makes available over 500 MW of its U.S. compute capacity, with an expansion target of 2.5 GW. The press release frames this as a strategic alignment of computing power demand and supply. The immediate market reaction was muted—Core Scientific's stock (CORZ) moved less than 2% in the following session. That silence is the first data point worth investigating. Context is everything. Core Scientific emerged from Chapter 11 bankruptcy in January 2024, having restructured $450 million in debt. The company, once the largest publicly traded bitcoin miner by hash rate, had been hammered by the 2022 credit crunch and the post-FTX collapse in asset prices. Its recovery plan hinged on diversifying revenue beyond bitcoin mining—specifically into high-performance computing (HPC) and artificial intelligence (AI) hosting. This is not a secret. Core Scientific’s investor presentations since early 2023 have consistently flagged AI hosting as a growth vector. The AMD deal is the concrete step. But the details matter. The cooperation is not about bitcoin mining. Bitcoin ASICs are application-specific; AMD’s GPUs are general-purpose. The 500 MW to 2.5 GW of compute capacity will be used to host AMD Instinct GPUs for AI training and inference workloads. Core Scientific is effectively converting its existing miner hosting sites—built for high power draw and cooling—into AI data centers. This is a sector pivot, not a technology upgrade. The company retains its mining operations, but the incremental capacity is allocated to AMD’s ecosystem. The economics are simple: AI hosting yields higher per-megawatt revenue than bitcoin mining at current hash prices (approximately $50–55 per petahash per day as of Q3 2024). However, the capital expenditure to retrofit these sites is non-trivial. Cooling systems, network infrastructure, and power stability compliance for AI workloads surpass mining requirements. Core Scientific has not disclosed the capital budget for this conversion—a yellow flag. The core of the analysis lies in the warrant structure. AMD will receive warrants to purchase shares of Core Scientific common stock at the prevailing market price. The number of warrants is tied to the compute capacity milestones: initial tranches for the first 500 MW, additional for expansion. The exercise price is the volume-weighted average price at the time of issuance. This is a standard industry arrangement—similar to the 2022 partnership between CoreWeave and an undisclosed GPU supplier—but with a twist. AMD is not paying cash upfront for the capacity; it is paying with the potential future equity dilution. The on-chain data doesn't lie, but the off-chain cap table tells the story. If AMD exercises all warrants, it could hold a 5–10% stake in Core Scientific, depending on the final capacity and share price. That dilution is a direct cost to existing shareholders, masked by the narrative of "strategic partnership." The press release avoided specifying the exact dilution percentage—a deliberate omission. Furthermore, the contract term is five years, renewable, with AMD having right of first refusal on additional capacity. This locks Core Scientific into a single-chip supplier. AMD’s Instinct MI300X and future MI400 series are competitive products, but sole-supplier dependency introduces supply chain risk—especially given AMD’s own production constraints and the geopolitical volatility of TSMC fabs. Core Scientific’s management likely negotiated performance guarantees, but no details were provided. Silence from the team speaks volumes. Let’s examine the feasibility of the 2.5 GW target. Core Scientific currently operates approximately 500 MW of mining capacity across five sites in the U.S. (Georgia, Kentucky, Texas, North Carolina, and North Dakota). The company has land and power purchase agreements for an additional 1.2 GW, but those are primarily allocated for future mining expansions. Diverting that to AI hosting requires new transformer installations, cooling retrofits, and—in some locations—grid interconnection upgrades. Texas, for example, is attractive for its deregulated market but suffers from peak summer demand that leads to curtailment. AI workloads require 99.99% uptime, unlike bitcoin mining which can pause. Reliability will be a key operational risk. The company has not published a timeline for the capacity ramp. Based on my audit experience with data center conversions, a 500 MW AI site takes 18–24 months to bring online from initial planning. Core Scientific’s first 500 MW may not be operational until late 2025 or early 2026—well past the current AI hype cycle peak. Now the contrarian angle. What do the bulls get right? The AI compute demand is real. Large language models and inference are power-hungry, and the existing cloud providers (AWS, Azure, GCP) are capacity-constrained for high-end GPUs. Niche colocation providers like Core Scientific can offer competitive pricing by leveraging lower real estate and power costs. AMD’s Instinct GPUs are gaining traction in inference workloads, especially where price/performance ratios beat NVIDIA's H100. The partnership cements Core Scientific as a preferred host for AMD-based AI clusters. If AMD’s market share grows, Core Scientific rides that wave. Moreover, the warrant structure aligns incentives: AMD profits from both chip sales and potential equity upside if Core Scientific succeeds. This is structurally similar to the NVIDIA–CoreWeave partnership that saw CoreWeave’s valuation surge to $8 billion in 2023. History suggests these deals can create real value. But the counterargument is rooted in execution asymmetry. CoreWeave operated as a pure-play AI cloud from the start, with engineering talent and data center expertise. Core Scientific is a mining company retooling its workforce. Its management team, led by CEO Adam Sullivan, has limited experience in AI hosting. The company’s technical staff consists of electrical engineers and mining technicians, not NVIDIA-certified network architects. Training or hiring for AI data center operations will be costly and time-consuming. Financial transparency is another concern. Core Scientific’s post-bankruptcy balance sheet shows $240 million in cash and $700 million in long-term debt. The debt carries high interest (12–15% floating). Funding the retrofits may require additional debt or equity issuance, further diluting shareholders. The AMD warrants are just the beginning—investors should expect more capital raises. The regulatory angle adds another layer. The U.S. government is increasingly scrutinizing energy consumption of data centers, especially those tied to AI. In May 2024, the Biden administration announced a task force to study AI data center power use. Core Scientific’s facilities in Texas and Kentucky already face community opposition over noise and water usage. Expanding to multi-gigawatt scale invites state-level energy regulation. Bitcoin miners have enjoyed some exemptions as "interruptible load" participants in grid balancing programs. AI hosting does not qualify for that flexibility. The business model shift exposes Core Scientific to stricter compliance costs. Quantify the key metric: Revenue per MW. Current bitcoin mining yields approximately $2,000–$3,000 per MW per day at $60,000 BTC and $0.05/kWh power. AI hosting, based on industry benchmarks, yields $8,000–$12,000 per MW per day for GPU clusters. The spread is substantial—if utilization rates exceed 70%. But utilization is not guaranteed. AI workloads are batch-oriented and often non-continuous. Core Scientific must secure long-term contracts from AI startups or enterprise clients. The AMD partnership provides the hardware supply but not the demand guarantee. The company is essentially building a hotel without confirmed guests. The partnership is a strategic hedge, but the roadmap is opaque. Core Scientific’s stock already trades at 8x forward EBITDA, reflecting some optimism. The warrants add another variable: if AMD exercises, it becomes an influential shareholder, potentially pushing for management changes or board seats. That could be either stabilizing or disruptive. The entire thesis of the "miner-to-AI-cloud" play requires a flawless execution of capital deployment, facility conversion, and customer acquisition. The press release glosses over all three. Follow the liquidity, find the leak. In this case, the liquidity is in the warrant terms and the capital expenditure schedule. I expect the company to file a Form 8-K with the SEC detailing the warrant agreement within weeks. That document will reveal the exact dilution cap and any vesting conditions. Until then, the narrative is incomplete. The takeaway is a call for accountability: regulatory filings are the only truth. The crypto industry has taught us that press releases are marketing, not analysis. Core Scientific and AMD have given investors a story, but the numbers are not yet auditable. Run the numbers, ignore the hype. Trust the code, not the press release. In this case, the code is the SEC filings. Until those are public, the partnership remains an intention, not a transaction.