A partnership between a Bitcoin miner and a chip giant sounds like a fairy tale for the bull market. But the code in the contract holds a bug that most headlines ignore.
Core Scientific, one of North America's largest Bitcoin miners, announced a strategic collaboration with AMD. The headline: AMD gains access to over 500 megawatts of computing capacity, with a vision to expand to 2.5 gigawatts. The sweetener: AMD receives warrants to buy Core Scientific stock at market price.
The market cheered. AI meets mining. The narrative practically writes itself.
But I read the fine print, and I see something else. A silent dilution engine, camouflaged by buzzwords.
Context: The Pivot That Isn't a Pivot
Core Scientific emerged from bankruptcy in early 2024 with a renewed focus: diversify beyond Bitcoin. The plan was to convert some of its massive energy infrastructure into AI data centers. This partnership with AMD is the culmination of that strategy.
AMD, for its part, is hungry for capacity. Its MI300X GPUs compete with Nvidia's H100, but AMD lacks the same integration with cloud providers. By locking in power and physical space via Core Scientific, AMD secures a scalable fabric for its chips.
The deal gives AMD warrants to buy CORZ stock. The exercise price is the market price at the time of warrant issuance—no discount. That sounds fair. But “market price” is a moving target, and warrants are leverage.
Core: Breaking Down the Code in the Spreadsheet
Let's treat this like a smart contract audit. I will examine the three critical variables: capacity, warrant terms, and electricity cost.
Capacity: 500 MW now, 2.5 GW eventually. For perspective, 1 GW can power roughly 300,000 US homes. Bitcoin miners typically operate at 30-50% overhead for cooling and infrastructure. For AI workloads, that overhead can double. The revenue per MW for AI hosting is higher than Bitcoin mining, but the capital expenditure is also higher. Core Scientific must spend heavily on liquid cooling, networking, and hardware procurement before a single dollar of AI revenue flows.
Warrants: AMD receives warrants to purchase up to an undisclosed number of shares at the average closing price of CORZ over the 30 days preceding the deal. This is the bug. By tying the exercise price to a short-term average, AMD effectively gets a free lookback option. If the stock rises, they exercise low; if it falls, they wait. The potential dilution is not capped in the press release. Based on comparable deals, warrant coverage often ranges from 5% to 15% of outstanding shares. For Core Scientific (roughly 150M shares outstanding), that could mean 7.5M to 22.5M new shares. That's a 5-15% dilution—equivalent to a subtle token inflation in a DeFi protocol.
"Code is law, but bugs are the human exception." The corporate code here has a bug: the warrant mechanics reward AMD for a share price increase they can influence through AI narrative pumping. It's a classic principal-agent misalignment.
Electricity Cost: Core Scientific reports average power cost of $0.04/kWh. For AI data centers, the all-in cost (power + cooling + maintenance) can exceed $0.10/kWh. AMD's MI300X draws up to 750W per chip. Scaling to 500 MW means hundreds of thousands of GPUs. At $0.10/kWh, that's $1.2M per day in electricity alone. Even with a bulk discount, margin is thin. The narrative says AI yields high margins, but the reality is that compute arbitrage is a razor-thin business when everyone chases it.
Based on my audit experience—remember the 0x protocol integer overflow that cost the team months of patches?—this partnership has a similar silent vulnerability. The overflow here is in the warrant dilution and the assumption that AI compute demand will remain inelastic to price.
Contractarian: What the Hype Misses
The contrarian angle is not that the deal is bad. It's that the market is ignoring the structural risk hidden in plain sight.
First, the demand for AI compute is not infinite. Current GPU shortage is real, but a 2.5 GW expansion takes 2-3 years to build. By then, hyperscalers like AWS and Google will have deployed their own custom chips (TPUs, Trainium). The market could be oversupplied by 2027. Core Scientific is betting on a tidal wave that may crest earlier than expected.
Second, the warrants create a perverse incentive for AMD to keep the stock price artificially high via positive announcements, rather than focusing on operational excellence. This is the "bug in the governance code"—a vulnerability that an auditor would flag as lack of time locks or vesting cliffs.
Third, there is the regulatory angle. MiCA and US energy regulations are tightening. A 2.5 GW facility in Texas could face power curtailment during peak summer. Core Scientific already experienced forced shutdowns during winter storms. ``The ledger remembers what the wallet forgets." The market has a short memory for blackouts.
I recall the Curve Finance liquidity audit I performed in 2020. The amp coefficient precision loss was minor, but under stress, it broke the invariant. Similarly, here the minor assumption of uninterrupted power can break the entire business model when a heat wave hits. The AI narrative is the amp coefficient—boosting returns when all goes well, but amplifying losses when the underlying assumptions fail.
Takeaway: A Forward-Looking Vulnerability Forecast
This partnership is not a moonshot. It's a defensive move—a hedge against Bitcoin price risk. But the hedge itself carries execution risk.
The true vulnerability is the dilution unwind. If Core Scientific fails to transition fast enough, the warrants become a deadweight loss. Investors will discover that the "AI premium" was just future dilution masked as revenue growth.
Watch for two signals: (1) the amount of shares AMD ultimately exercises, reported in the next 10-K; (2) the power cost per MWh reported in quarterly earnings. If power costs rise faster than AI revenue, the fairy tale turns into a tragedy.
"Code is law, but bugs are the human exception." The bug in this code is overconfidence in linear extrapolation—assuming today's AI demand will grow indefinitely. The correction will come when the market recalculates the true cost of compute.
As I wrote in my 2022 post on the Terra collapse: the fundamental chain is only as strong as its weakest external dependency. Here, the dependency is on power grids and market sentiment. Both are fragile.
For those of us who audit blockchain projects, the methodology transfers directly: verify every assumption, model the worst case, and never trust the narrative without reading the source. The source here is the warrant prospectus and the power purchase agreements. Everything else is noise.
The ledger remembers what the wallet forgets. In six months, the market will remember the dilution. When it does, the price will correct. Until then, trade the narrative if you must, but know the bug is hiding in the depths.
