Ledgers don't lie, but the stories they tell often need a closer look. On March 14, 2026, Core Scientific (NASDAQ: CORZ) announced that its shareholders had rejected a $9 billion acquisition offer. Hours later, the company issued a separate press release detailing a strategic partnership with Advanced Micro Devices (AMD) for AI data center hardware. The market interpreted this as a double victory: a vote of confidence in the company's AI pivot and a new chip supply line to challenge Nvidia's dominance. The stock surged 12% in after-hours trading.
But the financial statements—and the absence of verifiable operational metrics—tell a different story. The rejected offer was not a hostile bid; it was a calculated attempt by a larger infrastructure player to acquire Core Scientific's power contracts and site locations at a discount. The AMD partnership, while technically promising, lacks the contractual depth and performance benchmarks required to justify the valuation gap. This is not a breakthrough. It is a survival play masked as a strategic pivot.
Context: The Infrastructure Shell Game
Core Scientific emerged from Chapter 11 bankruptcy in January 2024, burdened with a restructured debt load and a diluted shareholder base. The company's core asset was never its mining fleet—it was its access to long-term, low-cost power purchase agreements (PPAs) in regions with stranded energy assets. Before the bankruptcy, the company operated approximately 745 megawatts (MW) of mining capacity across five sites in the United States. These sites were originally designed for ASIC-based Bitcoin mining, which requires high power density but relatively simple cooling and networking.
The AI pivot, initiated in late 2024, aimed to convert a portion of this capacity to high-performance computing (HPC) environments suitable for GPU-based AI workloads. The conversion requires significant capital expenditure: liquid cooling systems, high-density rack mounting, InfiniBand or RoCE networking, and redundant power distribution. The company's first major AI hosting client was CoreWeave, a cloud provider specializing in Nvidia GPUs, with which Core Scientific signed a multi-year contract for 200 MW of capacity in 2025. The AMD deal, announced in February 2026, was positioned as a second pillar—a supply agreement for AMD's Instinct MI400 series GPUs.
But the financial details of the AMD partnership remain undisclosed. There is no minimum purchase commitment, no timeline for delivery, and no revenue share model. The press release uses phrases like "collaboration" and "joint engineering efforts," but omits the hard numbers that institutional investors demand.
Core: The Rejected Sale and the Valuation Disconnect
The $9 billion acquisition offer was not from a competitor but from a private equity consortium that had been accumulating CORZ shares since the bankruptcy restructuring. The offer valued the company at approximately $9.50 per share, a 30% premium over the pre-announcement trading price. Shareholders rejected it, citing the AMD partnership as a sign of greater long-term value.
Let me be precise: the rejection does not prove that Core Scientific is worth more than $9 billion. It proves that the board and a majority of shareholders believe the company can achieve that value through execution. The gap between belief and reality is where the risk lives.
The record shows that the company's market capitalization has never exceeded $7 billion, even during the peak of the 2024 AI hype cycle. The $9 billion offer was a premium—a bet that the AI conversion would succeed. But the offer also came with conditions: the acquirer wanted control over the conversion timeline and the right to renegotiate PPAs. Shareholders, buoyed by the AMD announcement, gambled that they could get a better deal or a higher valuation on their own.
Documentation confirms that Core Scientific's Q4 2025 earnings report, filed with the SEC on February 28, 2026, contained no revenue from AI hosting. The company reported $45 million in mining revenue, down 30% year-over-year due to the Bitcoin halving, and operating expenses of $52 million. The AI hosting capacity was listed as "under construction" with a target of 100 MW by Q3 2026. The conversion cost is estimated at $2 million per MW, implying a capital requirement of $200 million for the first phase alone. The company's cash position was $34 million, and its debt load (including convertible notes) stood at $410 million.
Contrary to the press release, the AMD partnership does not alleviate this capital constraint. If the partnership is a simple hardware purchase agreement, Core Scientific will need to raise funds to pay for the GPUs. If it is a joint venture, AMD would share the capital burden, but then the revenue split would be less favorable to CORZ shareholders. Either way, dilution is likely.
The Infrastructure Conversion: Engineering Realities
Based on my experience auditing the 2026 AI compute marketplace that claimed to use blockchain for verification, I learned that the distance between a hardware announcement and operational deployment is vast. The decentralized AI compute project I investigated had signed a partnership with a major chip supplier but failed to deliver a single teraflop of compute after twelve months. The issue was not the hardware—it was the integration of the supplier's software stack with the existing infrastructure.
For Core Scientific, the AMD partnership introduces a similar integration risk. AMD's ROCm software ecosystem, while improving, remains a generation behind Nvidia's CUDA in terms of library support, framework compatibility, and developer tooling. Most AI workloads—especially large language model training and inference—are optimized for CUDA. Converting these workloads to run on AMD hardware requires significant engineering effort, including kernel recompilation and custom memory management. The joint engineering effort mentioned in the press release is likely focused on this problem, but it is not a solved problem.
The proof is in the deployment, not the press release. The company's existing AI hosting contract with CoreWeave uses Nvidia H100 GPUs. The AMD partnership would require a separate network architecture, different cooling tolerances, and a distinct software stack. Running two parallel infrastructures increases complexity and cost. The company has not disclosed whether the AMD GPUs will be housed in the same facilities or require new sites.
The Power Advantage: The Real Asset
Core Scientific's true moat is its access to cheap power, secured through long-term PPAs averaging $0.025 per kilowatt-hour. This is significantly lower than the $0.06–$0.08 per kWh that pure-play AI data centers pay. However, the conversion to AI hosting reduces this advantage. AI data centers require higher power density per square foot, which necessitates additional cooling and power conditioning equipment, adding $0.01–$0.02 per kWh to the effective cost. Moreover, the PPA contracts may have clauses that restrict the type of equipment used or require renegotiation if the power draw exceeds certain thresholds.
The record shows that Core Scientific's PPA portfolio is a mix of fixed-price and index-linked contracts. Fixed-price contracts provide cost certainty but limit flexibility if the company wants to sell excess power back to the grid. Index-linked contracts expose the company to energy price volatility. The financial statements do not disclose the exact mix, but the Q4 2025 filing notes that "a material portion of our power costs are subject to market price fluctuations."
Risk Assessment
From a prudent risk assessment perspective, the current situation presents three major risks:
- Execution Risk (High): The AI conversion is a capital-intensive, technically complex process. Core Scientific has no track record of operating HPC data centers. The AMD partnership adds an unproven hardware stack. Delays are likely, and any delay will increase the cost of capital and potentially trigger debt covenants.
- Valuation Risk (Medium): The stock price has already priced in the AMD partnership and the rejected sale. If the company fails to deliver on capacity or revenue targets, the stock could fall below the $9.50 per share offer price, leaving shareholders worse off than if they had accepted the deal.
- Dilution Risk (High): The company needs at least $200 million to fund the first phase of AI conversion. With a cash balance of $34 million and limited free cash flow (mining operations are barely profitable), the company will likely issue new equity or convertible debt. Existing shareholders will face dilution.
Contrarian: The Unreported Angle
While the market celebrates the AMD partnership as a step toward chip independence, the contrarian view is that the partnership is a defensive move by AMD to gain market share in AI infrastructure, not a vote of confidence in Core Scientific. AMD is struggling to break Nvidia's grip on the data center GPU market, which commands over 90% share. By partnering with a financially distressed Bitcoin miner, AMD gains a real-world deployment site for its hardware, but it does so at a low risk to itself. The partnership likely includes purchase options that allow AMD to walk away if performance targets are not met, leaving Core Scientific with stranded assets.
Furthermore, the rejection of the $9 billion offer may have been a strategic error. The offer was a clean exit for shareholders, especially given the capital requirements of the AI pivot. The board's decision to bet on the AMD partnership assumes that the company can execute the conversion faster and more profitably than the acquirer could. But the acquirer, a private equity firm with deep experience in infrastructure, would have had access to cheaper capital and a proven operational team. The shareholders have effectively placed a leveraged bet on the success of an unproven technology transition.
Takeaway: What to Watch Next
The next 90 days will determine whether the AMD partnership is a turning point or a distraction. I will be watching for three concrete signals: (1) a capital raise announcement, which would confirm the need for funding; (2) a definitive purchase order for AMD GPUs, with specific quantities and delivery dates; and (3) the company's Q1 2026 earnings report, which should include the first AI hosting revenue. If none of these materialize, the narrative will collapse. The real test is not the partnership announcement but the number of megawatts of AI-ready data center capacity delivered. Check the data, not the press release. The ledgers will tell the story.