The press release said 'strategic partnership.' The 8-K filing said 'non-binding.' The metadata said 'no commitments.' Someone needs to audit the difference.
Core Scientific shareholders just rejected a $9 billion acquisition. The stated reason: belief in the AMD partnership. But the partnership details are a black box. No power capacity commitments. No GPU deployment numbers. No software stack benchmarks. Just a press release.
That's not a strategy. That's a PR stunt.
Context: The Physical Infrastructure Pivot
Core Scientific is a publicly traded Bitcoin miner (CORZ) that survived a 2023 bankruptcy. Its core asset: long-term power purchase agreements (PPAs) secured at low rates. Bitcoin mining is a commodity business—post-halving, margins are razor-thin. The pivot to AI data center hosting is a logical move: repurpose the same power infrastructure to run GPU clusters for AI workloads. The narrative is clear: miners become AI cloud providers.
The $9 billion acquisition offer was a liquidity event. The board said no. The CEO said the AMD partnership is worth more. The market is now pricing that bet.
Core: The Systematic Teardown
Let's examine the AMD partnership from a technical, not marketing, perspective.
First, the hardware. AMD's Instinct MI300X series is a credible Nvidia H100 competitor on paper. But credible on paper is not credible in production. The software ecosystem—ROCm—is still playing catch-up to CUDA. Every AI framework, every inference optimization, every training pipeline is built on CUDA first. Porting to ROCm is not trivial. It requires engineering time, and, more importantly, it requires verified reference architectures.
Core Scientific has no publicly disclosed reference architecture. No test results. No latency benchmarks. No utilization rates. The partnership announcement was a handshake, not a deployment.
The code spoke, but the metadata lied. The press release screams 'transformation.' The data screams 'missing.'
Second, the infrastructure conversion. A Bitcoin mining facility is a simple setup: power in, ASICs in, heat out. AI data centers require liquid cooling, high-density racks, InfiniBand or RoCE networking, and precise power distribution. The conversion is not plug-and-play. It's a multi-year, capital-intensive engineering project. Core Scientific has experience with large-scale operations, but AI hosting is a different beast. The cooling, the networking, the software stack—these are new failure modes.
Based on my experience auditing smart contracts during the ICO boom—I reviewed over 40 ERC-20 tokens in three weeks, finding critical integer overflows in supposedly 'audited' projects—I know the gap between a press release and a working product. The AMD partnership smells the same: a logo slapped on a PowerPoint, not a working system.
Third, the financial engineering. The $9 billion offer was in cash. Shareholders rejected it. They are now betting that the AMD partnership generates more than $9 billion in future value. That's a high bar. The AMD partnership, as described, is a supply agreement. It doesn't guarantee revenue. It doesn't guarantee customers. It guarantees a supply of chips that may or may not find demand.
Volatility is the product; loss is the feature. The conversion costs are undisclosed. The capital budget is unknown. The balance sheet is still recovering from bankruptcy. If the AMD partnership requires upfront payments for chips, Core Scientific may need to raise capital—diluting existing shareholders. The very shareholders who just rejected a $9 billion cash offer.
DeFi doesn't scale; it just renames the same liquidity. The same applies here: AI doesn't scale; it just renames the same power. The power assets are real. The AI expertise is not. Core Scientific is a power company, not a software company. The AMD partnership does not address the software maturity gap. Who will manage the cluster orchestration? Who will optimize the model inference? The press release is silent.
Contrarian: What the Bulls Got Right
To be fair, the bulls have a point. The power assets are real. Core Scientific's PPAs give it a cost advantage that pure-play AI cloud providers cannot match. Power is the bottleneck for AI training. Access to low-cost, reliable power is a strategic asset.
The AMD partnership also diversifies the supply chain. Nvidia's dominance is a single point of failure. If AMD can deliver competitive performance, Core Scientific could offer lower-cost AI compute. The trend is real: hyperscalers are moving to custom chips and multi-vendor strategies.
The contrarian view is that the infrastructure is the moat, not the technology. The bull case is that Core Scientific is a power play, not a tech play. And in that context, the AMD partnership is a hedge. It's a bet that power will be more valuable than software in the long run.
Takeaway: The Accountability Call
The rejection of the $9 billion bid is a high-stakes bet. The AMD partnership is the hammer. But the market is still waiting for the nails. Where are the customer contracts? Where are the utilization numbers? Where is the evidence that the software stack works?
The press release is not the product. The infrastructure is the product. And until we see actual GPUs humming in those converted mining sheds, the only truth is the missing data.
Shareholders bet against the offer. Now they need to prove the bet. The clock is ticking. The next quarterly report will either validate the play or expose the gap.
I'll be watching the metadata.