On March 14, Core Scientific shareholders rejected a $9 billion acquisition offer. The same day, the company announced a strategic partnership with AMD. The stock closed down 2.3%. Data doesn't lie: the market is pricing in skepticism, not euphoria.
Context. Core Scientific is a Nasdaq-listed bitcoin miner (CORZ) that emerged from Chapter 11 bankruptcy in January 2024. Its core asset: 500+ megawatts of fully energized power infrastructure across Texas, Kentucky, and North Carolina. Originally built for ASIC racks, these facilities are now being retrofitted for GPU clusters targeting AI and HPC workloads. The company already hosts for CoreWeave under a multi-year contract. The AMD deal adds a second chip supplier, reducing dependency on Nvidia.
But the announcement is a strategy statement, not a technical delivery. No power capacity allocated to AMD chips, no performance benchmarks, no GPU count. The press release reads like a political memo, not a technical specification. That is a red flag for anyone who has audited infrastructure projects.
Core. The technical challenge is monumental. Converting a bitcoin mining site to an AI data center requires more than swapping ASICs for GPUs. Three bottlenecks dominate:
- Cooling: Bitcoin miners are air-cooled, running at 40-60°C. Modern AI GPUs like AMD's MI300X require liquid cooling, immersion or direct-to-chip, to maintain performance under sustained compute loads. Retrofitting existing facilities demands new piping, coolant distribution units, and heat rejection systems. Cost per megawatt: $1-2 million for liquid cooling alone.
- Networking: Bitcoin mining uses simple UDP-based protocols with low bandwidth. AI training requires high-speed interconnects like InfiniBand or RoCE v2, with sub-microsecond latency. The existing network backbone in most mining sites is a flat Ethernet fabric. Upgrading to a non-blocking GPU fabric increases capital expenditure by 30-50% per rack.
- Software Stack: AMD's ROCm is catching up to CUDA but still lacks full support for popular frameworks like TensorFlow's distributed training, and its memory management is less mature. Core Scientific will need to provide a validated software environment for tenants—this is not a hardware swap; it is a full-stack engineering problem.
Based on my experience auditing the Ethereum Classic supply shock scripts in 2017, I learned that infrastructure announcements without verified code or recent operational data are often priced in prematurely. The same applies here. The AMD partnership is a supply agreement, but the article discloses no minimum purchase commitment, no revenue share, no timeline for GPU deployment. The only concrete number is the rejected $9 billion valuation.
Verify the hash, ignore the hype. Until Core Scientific publishes audited power capacity allocated to AI, GPU utilization rates, and tenant contracts with committed SLAs, this deal remains a narrative. The company's own 2024 annual report shows AI hosting revenue accounted for only 18% of total revenue, with bitcoin mining still the majority. The pivot is real, but the scale is early.
Now, the contrarian angle. The common narrative is that Core Scientific is undervalued and the AMD partnership is a catalyst. But there is a blind spot: AMD needs Core Scientific as much as Core Scientific needs AMD. AMD is desperate for credible data centers to deploy its Instinct series against Nvidia's dominance. By partnering with a former miner, AMD gets a real-world test environment and a reference customer—but it also inherits the operational risk of a company that just emerged from bankruptcy. If Core Scientific stumbles on the conversion, AMD's credibility in the AI hardware race takes a hit.
Furthermore, the $9 billion rejection implies that shareholders believe the company is worth more. But the implied valuation is a double-edged sword. It sets a high bar for execution. If Core Scientific fails to deliver 100+ MW of AI compute capacity within 12 months, the stock will trade below the rejected offer price. The market is likely pricing in a 30-40% probability of execution failure, which explains the muted reaction.
On-chain metrics > Twitter polls. In this case, the 'on-chain' is the company's balance sheet and operational disclosures. The next quarterly report will reveal capital expenditure on AI infrastructure, depreciation schedules, and cash burn. Those numbers will tell the true story.
Takeaway. The next six months are critical. Watch for two specific signals: (1) a power purchase agreement that allocates at least 50 MW to a single AI tenant, and (2) a partnership announcement that includes a minimum revenue guarantee. If neither appears by Q3 2025, the AMD deal will be perceived as a marketing exercise, not a transformation. Investors should demand delivery, not declarations.