The chart lies. The crowd feels.
And right now, the crowd is feeling a wild rush of optimism. Core Scientific’s shareholders just rejected a $9 billion acquisition offer. The reason? They’re betting the company’s pivot into AI infrastructure, powered by a new partnership with AMD, is worth more. More than nine billion dollars. More than a guaranteed exit. More than the safety of a buyout.
But here’s the thing no one’s saying out loud: that bet is built on a press release, not a delivered megawatt.
Smile while the liquidity drains.
Let me take you inside the numbers, the tech, and the unspoken risks. I’ve been watching this space since 2017, back when I was a junior dev in Nairobi, chasing EtherDelta rumors. I’ve seen miners pivot to AI before. I’ve seen the hype drown out the hard realities of hardware integration. And I’ve seen the crowd smile while the liquidity drains.
This is the story of Core Scientific’s big gamble — and why you need to look past the headline.
Hook: The $9B Rejection That Shook the Street
On a quiet Tuesday, Bloomberg broke the news: Core Scientific shareholders had rejected a $9 billion acquisition proposal. The offer was from an unnamed consortium — likely a private equity firm looking to scoop up the bankrupt miner turned AI host. The rejection was swift. The board, backed by major institutional holders, said no. They pointed to the AMD partnership as the reason.
But here’s the kicker: the AMD partnership was announced just days before the rejection. No terms. No volume commitments. No delivery timelines. Just a handshake and a press release.
The market cheered. CORZ stock jumped 12% in after-hours trading. Analysts rushed to upgrade their price targets. The narrative was set: Core Scientific is no longer a miner; it’s an AI infrastructure play. And AI infrastructure plays trade at multiples miners can only dream of.
Yet I can’t shake the feeling that this is a story we’ve seen before. In 2021, every miner with a warehouse claimed they were becoming an AI cloud provider. Few succeeded. Most burned cash on GPUs that sat idle. The ones that survived were the ones that already had deep relationships with hyperscalers — not those signing press-release partnerships with GPU vendors.
Context: From Bankruptcy to AI Hype
Core Scientific isn’t new to the crypto infrastructure game. Before the 2022 bear market, it was one of the largest Bitcoin miners in North America, operating over 200,000 ASICs across multiple sites. Then the crypto winter hit. Energy prices surged. Bitcoin dropped. The company filed for Chapter 11 in late 2022.
Emerging from bankruptcy in early 2024, the company had a new strategy: convert its mining sites into AI data centers. The logic was simple. Mining sites already have power — lots of it. They have cooling, security, and access to cheap, long-term power purchase agreements. Why not plug in GPUs instead of ASICs?
By late 2024, Core Scientific had signed a multi-year hosting deal with CoreWeave, a pure-play AI cloud provider. The deal was for over 200 MW of capacity. That was the first real signal that the pivot had legs.
Then came the AMD partnership in early 2025. The press release was short on details. It mentioned “collaboration” and “deployment of AMD Instinct GPUs” for AI workloads. No word on how many GPUs, which models, or when the first cluster would go live.
But the market didn’t care. The news was enough to trigger the $9 billion rejection.
Core: The Technical Reality Check
Let’s strip away the hype and look at the technical foundations.
The Mining-to-AI Conversion Is Not Plug-and-Play
I’ve audited data center conversion plans for a few mining firms. Here’s the ugly truth: Bitcoin mining sites are built for ASICs, not GPUs. ASICs are low-power, air-cooled, and don’t need high-speed interconnects. GPU clusters — especially for AI training — require liquid cooling, high-density racks, InfiniBand or RoCE networking, and a completely different power distribution architecture.
A single NVIDIA H100 GPU can draw 700 watts. A rack of 8 GPUs can pull 6 kW. Now multiply that by hundreds of racks. The power density per square foot jumps 10x compared to ASIC mining. Most mining sites aren’t built for that. Retrofitting costs millions per megawatt.
Core Scientific has experience retrofitting — they did it for the CoreWeave deal. But that deal was with a hyperscaler that provided the GPUs and software stack. The AMD partnership is different. Core Scientific is likely buying the GPUs themselves. That means they’re taking on inventory risk.
AMD’s ROCm vs. NVIDIA’s CUDA: The Elephant in the Room
AMD’s Instinct GPUs are powerful on paper. But in the real world, AI workloads are optimized for NVIDIA’s CUDA ecosystem. PyTorch, TensorFlow, and most popular AI frameworks run best on CUDA. AMD’s ROCm open-source stack has improved, but it’s still a distant second. Developers don’t want to retrain their models for a different GPU architecture unless there’s a massive price advantage.
Core Scientific’s AMD deal is a bet that ROCm will catch up. Or that the price discount (AMD offers 20-30% cheaper per teraflop) will be enough to lure customers. But I’ve seen this play out in 2023 with the Intel Gaudi accelerators. The promise was there. The adoption wasn’t.
Based on my audit experience, I’d say the real bottleneck isn’t hardware — it’s software ecosystem lock-in. Even if Core Scientific deploys 10,000 AMD GPUs, they’ll struggle to fill them with paying customers if the software stack isn’t seamless.
The $9B Valuation Anchor
By rejecting the acquisition, shareholders implicitly set a floor: Core Scientific is worth more than $9 billion. At current share prices, the market cap is around $7.5 billion. So the stock has room to run — if the AMD partnership delivers.
But here’s the contrarian cold truth: the $9 billion offer was likely structured as a premium for control. Private equity buyers don’t pay retail. They pay a premium because they expect to extract more value by cutting costs, selling assets, or improving operations. The fact that shareholders said no means they believe the current management can create more value than a PE firm.
I’m not so sure. The management team has proven they can run a mining operation. They haven’t proven they can run an AI cloud business. The AMD partnership is their first big test. If they stumble, the stock could drop below $5 billion.
Contrarian: The Unreported Angle — This Is a Bet on Power, Not on GPUs
Everyone is talking about the AMD GPUs. The Instinct MI300X. The 5nm process. The teraflops.
But the real strategic asset Core Scientific brings to the table is power contracts.
Core Scientific has locked in long-term power purchase agreements (PPAs) at prices far below the market average. In Texas, they’re paying around $0.03 per kWh. The average hyperscaler pays $0.08-0.12. That’s a massive cost advantage.
AMD needs partners like Core Scientific to deploy their GPUs in real-world data centers. They need to prove that Instinct can handle AI workloads at scale. They need reference architectures. And they need to do it cheaply.
The partnership isn’t just about buying GPUs — it’s about co-engineering a solution. AMD will likely provide engineers to help Core Scientific integrate ROCm and optimize workloads. That’s a hidden value that the market hasn’t priced in.
But here’s the catch: power contracts are only valuable if you have customers to fill the capacity. Core Scientific’s CoreWeave deal is for 200 MW. The AMD deal is for an undisclosed amount. If they overbuild capacity and demand doesn’t materialize, they’ll be sitting on stranded assets.
The crowd is smiling because they see the AMD logo. But the chart lies. The crowd feels. The real question is: can Core Scientific sell that capacity to AI startups, enterprises, and researchers? That’s a sales and marketing challenge, not a technical one. And I haven’t seen any evidence that they have the sales team to do it.
Takeaway: What to Watch Next
Forget the press release. Here’s what I’ll be watching in the next 90 days:
- MW delivered: How many megawatts of GPU-optimized capacity does Core Scientific actually bring online? Not announced — delivered. The gap between press release and energized racks is where value is lost.
- Customer pipeline: Are they signing new AI hosting contracts? The CoreWeave deal was public. Any new names? If they’re relying on AMD’s customer relationships, that’s a risk.
- Earnings call tone: Management will be grilled on the AMD partnership. Listen for specifics — volume, timeline, revenue recognition. If they dodge, sell.
- AMD’s own execution: AMD has a history of missing GPU delivery targets. If Instinct supply is constrained, Core Scientific’s buildout slows.
Smile while the liquidity drains. The market is pricing in a perfect execution scenario. But infrastructure transformations are messy. The path from $7.5 billion to $10 billion is paved with a thousand operational details.
The chart lies. The crowd feels. Right now, the crowd feels euphoric. That’s usually the time to ask hard questions.
I’ll be watching. And I’ll be here when the next data point drops.