We have spent years convincing ourselves that Bitcoin miners are nothing more than energy arbitrageurs—vampires on the grid, sucking up cheap power to produce digital gold. Then Core Scientific, a company that emerged from Chapter 11 bankruptcy with the stigma of over-leverage, signs a 2.5 gigawatt partnership with AMD to build high-performance computing infrastructure for AI. The market cheered with a 40% stock spike, but the silence from traditional cloud providers was deafening. Why? Because this is not a deal about chips or watts; it is a deal about sovereignty—over power, compute, and the permission to build outside the cathedral of hyperscalers.
Context: The Genesis of a Second Act Core Scientific’s story is one of near-death and reinvention. In 2022, the company filed for bankruptcy after a brutal bear market and a failed merger with a SPAC. Its fleet of ASIC miners was sold off or idled. Creditors took control. The narrative was that mining as a business model had failed. But beneath the wreckage lay a network of high-voltage substations, cooling towers, and long-duration power purchase agreements—assets that would be worth billions in the age of AI. In 2024, Core Scientific emerged from restructuring with a new focus: hosting. It began offering colocation services for other miners, keeping its own mining fleet lean. Now, with the AMD deal, it is taking the final leap: from digital gold mine to compute foundry.
AMD’s motivation is equally clear. For years, it has played second fiddle to NVIDIA in the AI GPU market, hamstrung by a less mature software ecosystem (ROCm vs. CUDA) and slower adoption among cloud giants. A deal with a nimble, hungry partner like Core Scientific allows AMD to bypass the long sales cycles of Amazon and Microsoft, and instead prove its hardware in a high-volume, cost-sensitive environment. The 2.5 GW target—enough to power roughly 2-3 nuclear reactors—signals intent to build not just a cluster, but a new industrial standard for AI compute.
Core: The Architecture of Trust vs. The Architecture of Control Let me pause here and offer a personal anchor. In 2017, during the ICO madness, I withdrew from a lucrative token sale for a centralized exchange to spend three weeks auditing the 0x protocol’s relayer architecture. That decision—choosing permissionless design over immediate liquidity—shaped everything I do today. It taught me that architecture is the only true expression of values. When I look at Core Scientific’s plan, I ask: what architecture are they building?
From my work with the UK pension fund in 2024, where I pushed them to see Bitcoin mining as a grid stabilizer rather than a speculative bet, I learned that infrastructure carries an ethical dimension. Power is not neutral; it is political. A 2.5 GW compute footprint in West Virginia or Texas gives that region outsized influence over the direction of AI. If Core Scientific builds a closed, proprietary cloud, it replicates the power dynamics of AWS and Azure. If it builds an open, permissionless compute platform—perhaps with on-chain resource accounting and verifiable fair scheduling—it could become the backbone of decentralized AI.
Technically, the challenges are monumental. Scaling to 2.5 GW requires not just chips, but a complete rethink of data center design. Air cooling becomes impossible at that density; immersion cooling is mandatory. Networking fabrics must handle exabytes of data moving between GPUs. And the software stack—CUDA vs. ROCm—remains the elephant in the room. My 2020 analysis of Compound and Aave taught me that even elegant protocols fail if they exclude the poor on purpose. Here, the risk is that Core Scientific becomes a slave to NVIDIA’s ecosystem, or that AMD’s software immaturity leads to 30% lower utilization, killing the economics.
Yet, there is a hopeful path. Imagine a DePIN layer built on top of Core Scientific’s hardware: a token that represents a slice of compute, tradable on a secondary market, with smart contracts that allocate resources based on proof-of-stake voting. That would be the true liberation—not just selling compute to the highest bidder, but democratizing access. I saw this vision partially realized in 2026 when I led a project to build a blockchain provenance layer for AI content. We partnered with media houses to verify human-created work on-chain. The lesson: the infrastructure must be trustworthy, and trust is not given; it is verified.
Contrarian: The Cathedral’s Revenge But I must hold myself to the same standard I apply to others: the contrarian test. The market is pricing this deal as if Core Scientific has already won. It has not. The three cloud hyperscalers—Amazon, Microsoft, Google—control 67% of the AI compute market. They have decades of operational expertise, proprietary networking stacks (e.g., AWS Nitro), and sticky customer relationships built on services like SageMaker and Azure ML. Core Scientific enters as a commodity supplier, competing on price alone. That is a race to the bottom, especially when the hyperscalers can subsidize compute with their cloud revenue.
Furthermore, the capital requirement is staggering. To build 2.5 GW of HPC infrastructure, Core Scientific will need $10-15 billion, depending on chip prices and construction costs. The company’s balance sheet, even post-bankruptcy, is not pristine. It will need to issue equity, take on debt, or structure off-balance-sheet financing. Given the current interest rate environment (2026), debt is expensive. Equity dilution could crush existing shareholders. This is the same trap that snared many miners in 2022: they overleveraged during a boom and paid the price.
And then there is the regulatory axe. The US Department of Energy has begun scrutinizing data center power use, especially in regions with grid constraints. A 2.5 GW facility is a prime target for new energy taxes or carbon mandates. Core Scientific’s competitive advantage—cheap, often fossil-based power—could become a liability. My experience with the Scottish Highlands retreat in 2022 taught me that the market forgets its promises the moment the environment turns hostile. The protocol remembers what the market forgets.
Finally, consider the demand side. AI compute demand is currently surging, but it is not infinite. If a recession hits, if AI model improvements plateau, if the regulatory environment tightens—any of these could cause a sudden glut of compute capacity. The "blue chip" mining stocks of today, riding the AI narrative, could become the teraLUNA of tomorrow: when liquidity dries up, nothing remains.
Takeaway: The Signal Beneath the Noise So where does this leave us? I believe Core Scientific’s partnership with AMD is a genuine signal—not of immediate success, but of a structural shift in how we value electrical infrastructure. The mining industry, born from a cypherpunk dream of permissionless money, has accidentally built the physical backbone for permissionless compute. The question is whether Core Scientific will remain a service provider to the existing power structure, or whether it will open its gates and let anyone run code on its machines.
Patience is the validator of true intent. The market is pricing in three years of flawless execution. I am more skeptical. But I also know that the best systems are built in silence, by people who care more about the architecture than the applause. If Core Scientific builds with integrity, if it resists the temptation to lock its infrastructure behind a walled garden, it may earn the permission it seeks—not from AMD, but from the network itself.
Stillness reveals the signal beneath the noise. The signal here is that the future of compute is not about the fastest chip or the cheapest watt. It is about who controls the doors. Code is the only permission we truly need. And that permission is still being written.