The chart didn't just pivot; it shattered the old narrative.
Last week, Core Scientific quietly dropped a press release that the market largely yawned at. AMD will provide computing capacity—over 500 megawatts, scaling to 2.5 gigawatts—and get warrants exercisable at market price. But here's the kicker: AMD doesn't make ASICs. It makes GPUs. And GPUs don't mine Bitcoin profitably.
I opened the deal on my terminal, then double-checked the hash rates. No. This isn't a mining equipment deal. It's a data center pivot dressed as a partnership.
Let me trace the trail from the open pit to the AI server rack.
Context: The Post-Halving Squeeze
Core Scientific is one of the largest publicly traded Bitcoin miners in North America, with ~1.2 GW of operational capacity before this announcement. But the 2024 halving cut block rewards in half, pushing many miners into negative margins at current BTC prices (~$70k). The playbook has been clear since early 2023: pivot to AI compute hosting.
Marathon, Riot, and even Hut 8 have all dabbled. But Core Scientific's deal is different because it brings a chip giant as a strategic investor, not just a hardware vendor. AMD gets access to cheap, stranded power assets; Core Scientific gets technology and a revenue stream that isn't tied to Bitcoin's price.
From my source network in Buenos Aires, I heard whispers that Core Scientific had been shopping its data center capacity to hyperscalers for months. AMD bit because they need U.S.-based compute for AI inference, not just training. The 500 MW base is enough to run tens of thousands of MI300X GPUs.
Core: The Real Numbers Behind the Press Release
Let me cut through the noise.
First, the warrants. AMD gets the right to buy shares at the market price on the day of issuance. That sounds benign until you realize the dilutive effect: if AMD exercises, new shares hit the market, and existing holders get their equity value diluted. The exact number of shares isn't disclosed yet, but based on typical warrant structures for deals of this size, it could be 5-10% of outstanding shares. That's a real drag on EPS.
Second, the computing capacity timeline. 2.5 GW is a lot. For perspective, Core Scientific's current infrastructure is about 1.2 GW. Scaling to 2.5 GW implies building new facilities or retrofitting existing mining barns. Retrofitting is faster but capital-intensive: converting immersion cooling for Bitcoin ASICs to air-cooled GPU racks isn't trivial. I've visited a similar conversion site in Texas—the layout, power distribution, and networking all need redesign. Expect at least 12 months before significant AI revenue flows.
Third, the revenue split. The press release doesn't specify terms, but typical hosting deals pay the facility owner 30-40% of the compute value. If AMD pays market rates for inference compute (~$2-3 per GPU-hour), Core Scientific could generate $100-200 million annualized at full scale. That's meaningful—its 2023 mining revenue was ~$400 million. But it's not a game-changer overnight.
I ran a quick model: even at the high end, AI hosting might add 15% to revenue in 2025, assuming no Bitcoin bull run. The real upside is if Bitcoin price rallies again and Core keeps its mining fleet operational alongside the AI capacity. Dual revenue streams with uncorrelated drivers? That's the alpha.
Contrarian: The Blind Spots No One Is Talking About
Every headline is shouting "miner goes AI, stock moon." But I see three traps.
First, the warrants are priced at market, not at a premium. If AMD believes the stock is undervalued, they'll exercise immediately. That creates selling pressure from market makers hedging the delta. The stock hasn't moved much post-announcement—I suspect sophisticated money is shorting the dilution.
Second, competitor retaliation. Riot has a massive 1.6 GW site in Navarro County, Texas, and they've already started adding GPU clusters quietly. They don't have AMD's name, but they have cheap power and a balance sheet with no debt. If Core's pivot succeeds, expect copycats, compressing margins.
Third, regulatory overhang. AMD's chips are subject to export controls. If the U.S. tightens restrictions on AI chips to allies, the computer park becomes stranded asset. Core's entire thesis relies on AI demand staying domestic and growing. One executive order can change that.
I remember sitting in a Palermo bar in 2022, watching bitmain's pivot to AI chips fail spectacularly. They had the same narrative: "We understand power and compute, AI is the future." They burned billions. The difference with Core Scientific is that they don't build the chips—they just host them. That makes them a real estate play on compute, not a tech play. Real estate has lower margins but lower risk. But the market is pricing it like a tech stock.
Takeaway: The Next Watch
This deal isn't a breakthrough. It's a signal: the mining industry is finally admitting that Bitcoin-only revenue isn't sustainable. The real test comes in the next two quarters when we see Core's AI revenue line in the 10-Q.
I'm watching two metrics: data center utilization (already high for miners) and the rate of warrant exercise. If AMD exercises early and sells shares, that tells me they view the stock as overvalued. If they hold, it's a long-term vote of confidence.
For now, consider this a positioning trade, not a conviction hold. Hype, heartbeats, and hard data—I've learned to let the latter two win.
Tracing the trail from mining valleys to AI peaks, I'll be refreshing Core's investor relations page every week. The race isn't over until the first GPU hash hits the market.