The shareholders of Core Scientific did not just reject a $9 billion sale. They issued a statement embedded in the ledger of corporate action: the market's valuation of this company is a lie. The lie is not in the numbers—$9 billion is a concrete figure—but in the assumption that the company's pivot from Bitcoin mining to AI data center hosting is a straightforward upgrade. It is not. It is a complex, capital-intensive, and technically risky transformation that the AMD partnership announcement only obfuscates.
Context: The Ghost in the Corporate State
Core Scientific, listed on Nasdaq as CORZ, exited Chapter 11 bankruptcy in early 2024 after a brutal crypto winter that exposed the fragility of pure-play mining operations. The company survived by restructuring debt and pivoting to high-performance computing (HPC) and AI cloud services. The idea is elegant: repurpose the vast power infrastructure—low-cost electricity locked in long-term power purchase agreements (PPAs)—from running ASIC miners to powering GPU clusters for AI workloads. This is not a new concept. CoreWeave, a competitor that recently raised $1.2 billion, has done exactly this. But Core Scientific's competitive edge is supposed to be its existing fleet of 745 megawatts of operational capacity across five data centers, built for mining but now partially converted to host Nvidia H100 and AMD Instinct GPUs.
The AMD partnership, announced as the shareholders voted down the $9 billion acquisition by a consortium including private equity firm Blue Owl Capital, was framed as a strategic win. The deal signals supply chain diversification away from Nvidia, access to AMD's MI300X and future Instinct accelerators, and potential joint engineering optimization. But the announcement lacked any technical specifics: no committed power capacity, no GPU count, no timeline for deployment, no revenue guarantees. It was a press release, not a contract. The market, however, reacted positively, with CORZ shares rising 8% on the news. This is the classic trap of narrative over substance.
Core: Dissecting the Code Reveals the True Owner
Let me be clear: I have spent 29 years in this industry, reverse-engineering protocols and tracing transaction flows. I approached this analysis not as a journalist but as an on-chain detective examining the balance sheet of a company that claims to be a technology infrastructure play. The first thing I looked for was verifiable data. It is absent. The AMD partnership is a strategic announcement, not a technical milestone. There is no audit report, no third-party validation of the GPU cluster's performance, no published benchmark results comparing their ROCm deployment against Nvidia CUDA. The silence in the logs is louder than the error.
From a technical standpoint, converting a Bitcoin mining facility to an AI data center is not a matter of swapping ASICs for GPUs. It requires a complete overhaul of the physical infrastructure:
- Cooling: Bitcoin miners are air-cooled or use immersion cooling for ASICs, but modern AI workloads demand liquid cooling for high-density GPU racks. Core Scientific's existing sites may not have the plumbing, heat exchangers, or coolant distribution units required. Retrofitting a 100 MW facility costs tens of millions of dollars.
- Networking: Miners use simple Ethernet for Stratum protocols; AI clusters require InfiniBand or RoCE (RDMA over Converged Ethernet) for low-latency inter-GPU communication. This means installing new switches, cables, and network topology—again, a multi-million dollar investment per site.
- Software Stack: Nvidia's CUDA dominates the AI ecosystem with a mature toolchain, libraries like cuDNN, and widespread developer adoption. AMD's ROCm is catching up, but still lacks parity in performance for certain workloads, especially in training large language models. Core Scientific will need to either hire a team of AI infrastructure engineers or rely on AMD's support, which is a bottleneck.
And then there is the supply chain risk. AMD's Instinct MI300X, while promising, has faced production delays and higher-than-expected power consumption in real-world deployments. If AMD cannot deliver the promised volume or if the chips underperform, Core Scientific's committed capacity to customers like CoreWeave—with whom they signed a multi-year contract in 2024—could be at risk. This is not speculation. It is a pattern. In 2023, AMD's MI250X fell short of performance targets for HPC workloads, leading to delayed deployments. The company's earnings call in January 2025 revealed that Instinct revenue grew 200% year-over-year, but the base was small. The ramp is real, but it is not yet proven at scale.
Tracing the ghost in the corporate balance sheet state, I see a company that is trying to balance two conflicting capital expenditure demands: maintaining the Bitcoin mining fleet (which still generates cash flow, albeit diminished by the halving) and investing in AI infrastructure. The 2023 bankruptcy restructuring imposed strict covenants on debt and capital spending. The company's latest 10-K filing (February 2025) shows $1.2 billion in long-term debt, with $400 million due in 2026. The AI pivot requires at least $500 million in additional CapEx over the next two years, based on industry benchmarks of $1.5 million per megawatt for GPU cluster buildout. Where will this money come from? Equity dilution? The shareholders just rejected a $9 billion sale, but they may accept a dilutive secondary offering. The irony is that the same shareholders who voted against the acquisition are now betting on the company's ability to raise capital at a higher valuation. The market is pricing in a successful pivot, but the data does not support it.
Contrarian: What the Bulls Got Right
I am not a cynic for the sake of being one. There are valid reasons to be optimistic about Core Scientific's AMD partnership, and I must acknowledge them to maintain forensic integrity. The bulls argue that the company's power cost advantage is real. Core Scientific locked in PPAs at an average of $0.03 per kWh in 2021, before the energy crisis. This is significantly lower than the $0.06-$0.08 per kWh that typical AI data centers pay. Over a 5-year contract, this spread can generate tens of millions in additional profit margin. Additionally, the company's existing fleet of ASIC miners is not a liability; it is a hedge. If AI demand softens, they can simply switch capacity back to Bitcoin mining, which still has a positive hashprice. This optionality is unique among pure-play AI data center operators.
Furthermore, the AMD partnership is not just about hardware. It includes a joint engineering agreement to optimize ROCm for Core Scientific's specific workloads. If this yields measurable performance improvements, Core Scientific could become a reference architecture for AMD-powered AI clouds, attracting more customers. The company's management, led by CEO Adam Sullivan, has a track record of operational execution—they successfully navigated bankruptcy and kept the lights on. The market is betting on the execution, not the announcement.
But the bulls are ignoring a critical blind spot: the AMD partnership is a defensive move, not an offensive one. Core Scientific relies on CoreWeave for a significant portion of its AI revenue. CoreWeave, in turn, is heavily invested in Nvidia hardware. If Core Scientific moves too aggressively toward AMD, it risks alienating its largest customer. The partnership may be designed to give Core Scientific bargaining power with Nvidia, not to replace Nvidia entirely. This is a delicate balance, and one misstep could trigger a cascade of contract renegotiations.
Flash loans don't forgive, and neither does the market. The strategic ambiguity of the AMD deal is a feature, not a bug. It allows Core Scientific to signal innovation without committing to specifics. But the market will eventually demand proof. The next catalyst will be the Q1 2025 earnings call, where the company must disclose how many AI-capable megawatts are operational, utilization rates, and revenue per megawatt. If those numbers lag expectations, the stock will correct, and the $9 billion offer will look like a missed opportunity.
Takeaway: The Only Audit That Matters
Core Scientific's shareholders have placed a bet that the company can transform itself from a Bitcoin miner into an AI infrastructure leader worth more than $9 billion. The AMD partnership is a necessary step, but it is not sufficient. The real work is in the data center floors, the cooling pipes, the ROCm optimizations, and the capital markets. I will be watching the on-chain data—not the blockchain, but the corporate ledger: the 10-K filings, the capital expenditure disclosures, the debt covenants. Because logic is immutable; intent is often malicious. The intent here is survival and growth. The logic, however, is still being written. The ghost in the smart contract state is the company's own financial health. Until the numbers are auditable, the AMD deal is just a warm lie if the power fails.