Shareholders said no. A $9 billion exit, rejected. The math looked simple: take the premium, lock in the gain. But the board and majority holders chose a different path. They bet on AMD. They bet on the pivot from mining to AI. The question is whether that bet is grounded in technical reality or just a narrative shift.
Core Scientific is a Bitcoin mining operator. It emerged from bankruptcy in 2023 with a new strategy: repurpose its massive power infrastructure for AI data center hosting. The company signed a multi-year hosting deal with CoreWeave, a pure-play AI cloud provider. Then came the AMD partnership. The press release was light on specifics. No GPU count. No wattage. No revenue guarantees. Just a handshake and a promise.
Context: The infrastructure play
Core Scientific exists at the intersection of two capital-intensive industries: Bitcoin mining and AI cloud. The asset is not the GPUs. It is the power. Long-term power purchase agreements at fixed, low rates. Sites that are already built, with high-voltage transformers, cooling, and security. The engineering thesis is simple: a bitcoin mining site is a data center. It just needs different hardware inside.
But converting a mining site to an AI data center is not trivial. Mining requires ASICs. AI requires GPUs. The power density per rack is higher. The cooling shifts from air to liquid. The networking changes from a simple LAN to InfiniBand or RoCE fabric. The software stack moves from a single mining client to a multi-tenant orchestration layer. Core Scientific has experience running mining, not AI. The CoreWeave deal provided a bridge. The AMD deal is a source of supply.
Core: The AMD partnership under the microscope
The article provided no technical data. No benchmarks. No delivery timelines. No information on which AMD Instinct models are being used. No reference to software integration with ROCm. This is a strategic announcement, not a technical milestone.
From my experience auditing Layer2 protocols, I've learned to distinguish between a whitepaper with formal proofs and a press release with vague promises. The gap is similar here. The AMD partnership is a supply agreement. It does not guarantee that the GPUs will be deployed, that they will achieve target utilization, or that the AI workloads will be competitive against Nvidia's CUDA ecosystem.
AMD's ROCm software stack has improved significantly in the last two years. But it is not parity with CUDA. Many AI frameworks and libraries are still optimized for Nvidia. A large language model that runs on CUDA may require significant engineering effort to port to ROCm. Core Scientific is not a software company. It is a hardware operator. The burden of software integration falls on AMD and the end customers. If customers are not willing to run on AMD, the GPUs sit idle. The hosting revenue never materializes.
The trillion-dollar question: capacity and utilization
Core Scientific's value proposition is its power capacity. The company reported total capacity of around 1.2 gigawatts across its sites. Some of that is already contracted for mining. Some is being converted for AI. The conversion rate is the key metric. How many megawatts are actually live for AI hosting? What is the utilization rate? The press release gives zero guidance.
Contrast with CoreWeave, which operates its own data centers and claims 99% utilization on its Nvidia clusters. Core Scientific is a landlord, not a tenant. The landlord makes money from the lease, not from the compute. The revenue per megawatt is lower than a pure AI cloud provider. The risk is lower as well. But the upside is capped unless Core Scientific moves into the compute layer itself.
The AMD partnership could be a step toward that. If Core Scientific can offer a differentiated AMD-based cloud service, it could capture a share of the growing demand for alternative AI hardware. But that requires a sales team, a software stack, and a support infrastructure. None of that is visible today.
Contrarian: The blind spots in the AMD narrative
The market is pricing in a success scenario. The stock has rallied since the announcement. The rejection of the $9 billion sale implies that shareholders believe the company can create more than $9 billion in value through the AMD pivot. That is a high bar. Core Scientific's market cap before the news was around $3 billion. The acquisition premium was 300%. To justify the rejection, management must deliver significantly more than that.
Risk is a feature, not a bug, until it isn't.
The AMD partnership introduces a single-supplier concentration risk. AMD's GPU supply chain is still maturing. The MI300 series is in high demand, but availability is constrained. If AMD cannot deliver enough GPUs to meet Core Scientific's deployment schedule, the revenue growth will stall. The company's entire narrative is tied to this partnership.
Another blind spot: the timing. The AI boom is happening now. Core Scientific's conversion is not. While hyperscalers are building new data centers from scratch, Core Scientific is retrofitting mining sites. Retrofits take time. Permitting, construction, and commissioning of liquid cooling and high-density racks can take 12-18 months. The market may be pricing in revenue that won't appear until 2026.
History repeats in the ledger, not the news.
Core Scientific has a history of financial distress. It filed for bankruptcy in 2022 after the crypto bear market and the collapse of the Terra ecosystem. The company emerged with a new capital structure, but the debt is still there. The AI pivot requires significant capital expenditure. Core Scientific will need to raise additional funds, either through equity issuance or debt. Equity issuance would dilute shareholders. Debt would increase the risk profile. Neither is priced into the current stock.
The yield is the exit liquidity.
The company's mining operations still generate Bitcoin. But the Bitcoin halving in 2024 reduces the mining revenue per hash. The AI hosting revenue is meant to fill the gap. If the AMD partnership fails to deliver, the company will be left with a shrinking mining business and a high fixed cost base.
Takeaway: The next 12 months will reveal the truth
The market is betting on a successful pivot. But the evidence is thin. The company has not published any audited capacity metrics for its AI hosting. The AMD partnership is a supply agreement, not a revenue contract. The shareholders' rejection of the $9 billion sale sets a performance benchmark that is ambitious.
Investors should watch for concrete milestones: megawatts of AI hosting capacity brought online, GPU utilization rates, average revenue per megawatt, and customer announcements. If the company posts quarterly results showing AI revenue growing linearly, the narrative holds. If the growth is slow or the numbers are masked by one-time gains, the market will reprice.
The math holds until the incentive breaks.
The incentive for management is to paint a rosy picture. The incentive for investors is to verify the details. The Core Scientific story is a test of whether the market can look past the press release and demand the data. So far, the data is missing. The stock is a bet on execution. Execution is hard. History shows that.