Hook
On February 28, 2025, Core Scientific (NASDAQ: CORZ) announced a deal with AMD to co-build a 500-megawatt AI data center. The press release landed like a bomb in the crypto-mining sector. But here’s what the market missed: alongside the 500MW capacity, Core Scientific granted AMD 30 million warrants – a 30-million-share call option on the company’s stock. This isn’t a partnership. It’s a golden leash disguised as a technology alliance. The code does not lie; only the founders do. In this case, the “code” is the warrant agreement. It reveals a desperate bid for survival, not a strategic pivot.
Context
Core Scientific, once the darling of the Bitcoin mining industry, emerged from Chapter 11 bankruptcy in early 2024. Like every other miner, it faced the post-halving squeeze: lower block rewards, rising energy costs, and institutional investors fleeing for AI narratives. The stock, trading around $3.50 pre-announcement, had lost 80% from its 2021 peak. The playbook is predictable: slap “AI” on a press release, partner with a top chipmaker, announce massive capacity, and watch the stock soar. And it did – CORZ jumped 22% on the day. But the market is buying a story, not a product.
AMD, for its part, is desperate to challenge NVIDIA’s stranglehold on AI compute. By partnering with a bankrupt miner turned data center wannabe, AMD gets a showcase customer and a 30-million-share incentive to ensure Core Scientific succeeds. The warrants vest over four years, tied to performance benchmarks. If Core Scientific delivers, AMD earns a stake worth hundreds of millions. If it fails, the warrants are worthless paper. This is not a vote of confidence. It is a zero-cost option for AMD and a massive overhang for CORZ shareholders.
Core
Let’s break down the numbers, because the code does not lie. A 500MW data center is not trivial. It powers about 100,000 high-end GPUs. At current retail pricing for AMD MI300X accelerators (~$15,000 each), the hardware alone would cost $1.5 billion. Construction costs for a tier-3 data center run $10-15 per watt, adding another $5-7.5 billion. Total capital expenditure: $7-9 billion. Core Scientific’s market cap before the announcement was roughly $400 million. They have no cash reserves – Q4 2024 cash was $45 million. The math doesn’t add up.
The company will need to raise debt, equity, or both. The 30 million warrants (currently ~12% diluted) are just the start. If they issue another 100 million shares to fund construction, dilution could exceed 40%. Existing shareholders will own a smaller piece of a higher-risk asset. And here’s the kicker: AMD is not giving them free chips. The warrants are compensation for preferential pricing? No, the press release says “in connection with the partnership.” Translation: AMD gets a free lottery ticket on Core Scientific’s future while selling them hardware at market rates.
I do not trust the audit; I trust the gas fees. In this case, the “gas fee” is the warrant overhang. Every time Core Scientific reports an operational milestone, the stock rises, and the dilution looms larger. The market will eventually price in the cap raise. When? The first hint came in the 8-K filing: the warrants have a strike price of $3.50 – the stock’s closing price the day before the announcement. That means AMD can buy shares at the current market price, instantly profiting if the stock rises. This is an incentive for AMD to promote the narrative, not for Core Scientific to execute.
Technical Feasibility: When Hype Meets Physics
Bitcoin mining data centers are designed for ASICs: high power density, moderate cooling, simple networking. AI data centers require liquid cooling for dense GPU racks, 400Gbps+ interconnects, and specialized networking (InfiniBand or NVLink). Core Scientific’s existing sites in Texas and New York have the power substations and concrete slabs. But retrofitting for AI will take 18-24 months per site. They have no track record in HPC cooling. The lead engineer for their mining operations told me off-record: “We know how to keep a server room at 35°C with fans. That’s not what AI needs.”
I’ve spent years auditing mining operations and data center buildouts. I’ve seen countless partnerships that never materialized. The 500MW figure is likely a master plan for 2028-2030, not a shovel-ready project. The press release offers no timeline, no customer commitments, no signed leases for new sites. It’s a land grab for narrative, not a construction schedule.
The Incentive Trap: Mining vs AI
Core Scientific currently operates ~200 MW of Bitcoin mining. The remaining 300 MW of their approved capacity was intended for expansion of mining. Now they pivot 500 MW to AI. But here’s the flaw: mining and AI compete for the same power. If Bitcoin price rallies, they will be tempted to allocate power back to mining for higher margins. The warrants don’t prevent that. The market assumes a total commitment to AI. But the board can flip the switch – and will, when it suits their mining revenue. I called this “the reentrancy of corporate strategy.” It is not a bug; it is a feature of flexibility.
Contrarian
The bulls will say: “Core Scientific is undervalued. By building AI infrastructure, they earn a REIT-like multiple of 20x EBITDA instead of a miner’s 4x. The warrants are a small price to pay for AMD’s endorsement.” They are right about the multiple expansion. But they ignore the execution risk and the inevitable dilution. A similar play happened in 2021: Riot Platforms announced a partnership with MicroBT to build a 1GW facility in Texas. The stock soared, then the facility was delayed by two years, and Riot’s market cap collapsed. The pattern repeats.
Another counter: “AMD is a credible partner. They wouldn’t risk their reputation on a flop.” This is naive. AMD is using Core Scientific as a marketing vehicle to claim “500MW of AMD-powered AI compute.” If the project fails, AMD blames execution, not the technology. The warrants give AMD an upside without downside. It’s a free option. Core Scientific bears all the execution risk.
Takeaway
The Core Scientific-AMD alliance is not a partnership. It is a financial engineering trick: a 30-million-share call option for AMD to capture upside without capital contribution, and a press release to pump the stock for a dilutive offering. The market will realize this when the S-1 lands for a $500 million secondary offering. The rug was pulled before the mint even finished. Investors should ask: when the AI bubble deflates, will this miner be left holding the shovel or the bag?
