Trust nothing. Verify everything.
The data shows a familiar pattern. A Bitcoin miner announces a partnership with a chip giant. The press release emphasizes synergy, scale, and the AI frontier. The market briefly cheers. But the real story is buried in the warrant structure. And the numbers do not lie.
On the surface, Core Scientific’s agreement with AMD looks like a textbook infrastructure play. AMD gains access to 500 megawatts of U.S.-based computing capacity—a rare, low-latency resource for AI inference workloads. Core Scientific gets the right to issue warrants for approximately 50 million shares of its common stock, exercisable at market price on the issuance date. The partnership also sets a target to expand Core Scientific’s total capacity to 2.5 gigawatts. A compelling narrative, if you ignore the technical details.
But I do not ignore details. I disassemble them. This is a forensic audit, not a summary. Let me walk through the contract mechanics, the implied risks, and the blind spots that most coverage misses.
Context: The Miner’s Pivot
Core Scientific emerged from Chapter 11 bankruptcy in early 2024 with a new identity: a hybrid Bitcoin mining and AI compute provider. The company operates several large-scale data centers in the U.S., originally designed for ASIC miners. The pivot to AI hosting is logical—AI workloads require high power density, low latency, and robust cooling, all of which mining facilities already deliver. But the economics differ: AI hosting demands guaranteed uptime (SLA of 99.9% vs. mining’s variable hash power), different hardware (GPUs over ASICs), and a different customer base (enterprise vs. miners).
AMD is the world’s second-largest GPU manufacturer, behind Nvidia. Its Instinct MI300X accelerators target AI workloads, but AMD lacks its own data center capacity. Partnerships with hosting providers like Core Scientific are a strategic necessity. The warrants—50 million shares, roughly 15% of Core Scientific’s current diluted shares—serve as both a financial incentive and a long-term lock-in. AMD is betting on Core Scientific’s ability to convert its mining infrastructure into a competitive AI compute platform.
The announcement itself is short on specifics. No timeline for the 2.5 GW expansion. No details on power procurement. No disclosure of the warrant exercise price beyond “market price on issuance date.” This lack of transparency is a red flag for anyone who has audited tokenized incentive structures. Based on my experience architecting a DeFi yield aggregator in Zurich, I know that vague contractual terms often hide execution risk.
Core Analysis: The Warrants and the Hidden Leverage
Let us audit the warrants. A warrant is a derivative that gives the holder the right to buy shares at a predetermined price. In this case, the price is the market price on the date of issuance. This is unusual. Typically, warrants include a premium (e.g., 120% of market) to protect existing shareholders from dilution. The absence of a premium means AMD can immediately exercise and sell for a zero-cost arbitrage—if the stock remains stable. But the real risk is dynamic.
Assume Core Scientific’s stock trades at $5.00 per share on the warrant issuance date. AMD can purchase shares at $5.00. If the stock drops to $4.00, the warrants are worthless unless the terms allow reissuance. If the stock rises to $6.00, AMD gains a 20% profit per share. This structure incentivizes AMD to keep the stock price stable or rising—but only if they intend to exercise. However, the warrants are not immediately exercisable; there is usually a vesting period (e.g., 6 months). During that period, any dilution from the warrants is a shadow overhang.
Let me calculate the potential dilution. Core Scientific currently has approximately 330 million shares outstanding (pre-warrant). 50 million warrants represent a 15% dilution. If fully exercised, the company receives $250 million at $5.00 per share—funding for the 2.5 GW expansion. But this dilutes existing shareholders by 15%. The market caps this dilution into the stock price immediately. The net effect: a capital raise disguised as a partnership.
Complexity is the enemy of security.
The warrants also grant AMD a significant stake—possibly enough to demand a board seat. This is not disclosed, but the pattern is clear. In my work designing a regulatory compliance framework for a Swiss RWA tokenization platform, I learned that equity-linked warrants often include governance clauses. If AMD gains board representation, Core Scientific’s strategic independence becomes compromised. The miner’s capital allocation will be influenced by AMD’s product roadmap. This is a blind spot that most market analysts miss.
Now, the infrastructure expansion. 2.5 GW is a massive scale. For context, a typical Bitcoin mining site uses 50–100 MW. A large AI data center uses 100–200 MW. Core Scientific’s current footprint is around 700 MW. The goal requires tripling capacity. Where will the power come from? The U.S. grid is strained, especially in Texas (Core Scientific’s home state) where ERCOT has faced winter reliability crises. Power purchase agreements (PPAs) for new connections can take 2–4 years. The announcement offers no timeline. This suggests the expansion is aspirational, not immediate.
Let me cross-reference with my scalability benchmarking work on Polygon zkEVM. In that project, I measured proof generation latency and found that scaling physical infrastructure introduces delays that are often underestimated. For a data center, the latency is in power transmission, cooling installation, and regulatory permits. Core Scientific’s 2.5 GW target is similar to a large cloud provider, but they lack the balance sheet of AWS. The warrants provide capital, but $250 million covers only a fraction of the buildout cost—estimated at $5–10 million per MW for AI-ready facilities. So the expansion will likely require debt or further equity sales, which means more dilution.
The ledger does not forgive.
Contrarian Angle: The Commoditization of AI Compute
The partnership presents a counter-intuitive risk: it accelerates the commoditization of AI compute. As more mining farms convert to AI hosting, the supply of GPU-accessible data centers will surge. This pushes down margins. Core Scientific’s competitive edge—low-cost power from mining sites—will erode as others copy the model. Riot Platforms, Marathon Digital, and Hive Blockchain are all pursuing similar pivots. The market may see a race to the bottom on pricing.
Moreover, the warrants create a misaligned incentive. AMD benefits if Core Scientific’s stock price rises, but AMD’s primary interest is securing cheap compute capacity, not maximizing Core Scientific’s equity. If the compute deal offers favorable rates to AMD, Core Scientific’s revenue from AI hosting may be lower than market rates. This is a classic principal-agent problem. The warrants might have been structured to compensate AMD for taking a discounted compute price. But that discount is not visible in the press release. It is a hidden cost to Core Scientific’s shareholders.
Another blind spot: regulatory scrutiny. The SEC has been eyeing crypto mining companies for potential securities law violations related to power offtake agreements and token sales. Core Scientific’s warrants are standard, but the framework is shifting. In my work with the Swiss RWA platform, I mapped MiCA compliance for governance tokens. Here, the warrants could be viewed as a form of compensation that requires shareholder approval under Nasdaq rules. If not properly disclosed, this could trigger a lawsuit. The risk is low but non-zero.
Finally, the AI compute market is not purely a technology play—it is an operational one. Cooling failures, grid outages, and supply chain delays can turn a profitable data center into a liability. During the Terra-Luna collapse audit, I identified 12 failure points in the Anchor Protocol’s rebalancing logic. Similarly, Core Scientific’s pivot has multiple single points of failure: its reliance on AMD for GPU supply, its dependence on ERCOT grid stability, and its need to retain experienced staff. The company emerged from bankruptcy only a year ago; its financial resilience is unproven.
Takeaway: A Binary Outcome
This partnership will be judged by two metrics: AI revenue as a percentage of total revenue within 12 months, and the actual capacity reaching 1.5 GW within 18 months. If AI hosting contributes less than 35% of revenue by Q4 2025, the warrants will be purely dilutive, and the stock will underperform. If the expansion stalls, Core Scientific will face an existential crisis—caught between a declining Bitcoin mining margin and a crowded AI market.
My recommendation for institutional readers: monitor the quarterly filings for AI revenue breakdown. Do not trust the press release. Verify the power PPAs. Audit the warrant exercise pattern. The code of this deal is written in financial contracts, and it will be enforced by the market. The ledger does not forgive. Complexity is the enemy of security. And the data—once fully disclosed—will reveal whether this is a strategic pivot or a dilution trap.
Based on my experience designing a formal verification framework for AI-agent smart contract interactions, I know that any system that relies on optimistic assumptions about execution will fail. Core Scientific’s expansion is an optimistic assumption. The warrants are a hedge. The only certainty is that the burden of proof lies with the company. Until they deliver verifiable results, skepticism is the rational position. Trust nothing. Verify everything.