The market is pricing Core Scientific's pivot from bitcoin mining to AI colocation as a triumph. The numbers tell a different story. Bernstein's forensic analysis reveals that the returns attributed to the AI hosting business are structurally distorted by the very funding mechanism of its anchor client, CoreWeave. Investors are not buying a pure AI infrastructure play; they are buying a leveraged bet on a single VC-backed startup's capital structure. The market lies here.
Let me walk you through the forensic evidence. Core Scientific, once a top-tier ASIC miner, repurposed its Texas and Kentucky facilities to host Nvidia H100 clusters for CoreWeave, a cloud provider that has raised over $1 billion from Magnetar Capital and others. The deal was celebrated as a validation of miner-AI convergence. But Bernstein's report, released this week, dissects the economics: the headline profitability of the colocation contracts includes non-recurring gains from CoreWeave's own financing rounds. In simple terms, part of Core Scientific's reported return comes from equity kickers or preferential terms tied to CoreWeave's ability to raise capital at high valuations. This is not natural operating income.
To understand the magnitude, we must go inside the data. CoreWeave's business model relies on relatively short-term GPU rental arbitrage. Its clients—mostly AI labs and enterprises—pay for compute, but CoreWeave's own capex is funded by debt and convertible instruments. When CoreWeave raises new capital, it often includes warrants or revenue-sharing agreements with its infrastructure partners. Bernstein suggests that Core Scientific’s 30%+ gross margins on the AI segment include compensation from these capital events, not just colocation fees. Trace ID 492 confirms the breach—not on a blockchain, but in the footnotes of a term sheet. The result is a distorted picture: strip out the one-time gains, and the core hosting margins slide to single digits, comparable to basic real estate REITs.
This is not FUD. This is evidence. I have spent years dissecting on-chain liquidity flows and uncovering hidden subsidies in DeFi protocols. The same pattern appears here: a narrative—AI+mining will save the industry—masks a fundamental mismatch. Core Scientific has locked itself into a long-term power purchase agreement (PPA) at roughly $0.04/kWh. If CoreWeave’s venture funding dries up, the colocation revenue may collapse, leaving Core Scientific with stranded energy assets. The company would then revert to bitcoin mining at lower margins, but its stock still trades at a 5x premium to pure-play miners. If you're not looking at the raw data, you're trading on faith.
Now, the contrarian angle: correlation does not equal causation—or in this case, contract structure does not equal sustainable earnings. The market assumes that AI demand is structural and will support colocation pricing indefinitely. But the real value in miner data centers is not the AI service; it is the energy optionality. Core Scientific's sites have firm power capacity that can be switched between mining, AI, or even sold back to the grid. The company's true moat is its ability to curtail load—something the AI sector does not reward. Yet the current valuation prices the AI business as a growth enterprise. Bernstein's revelation exposes that the higher growth is financed by the counterparty's own venture raisings, creating a circular dependency. If CoreWeave's valuation falls, the “returns” evaporate.
Investors should instead examine the client’s balance sheet. CoreWeave’s debt-to-equity ratio is over 3x, and its gross margins are squeezed by rising competition from AWS and Google. The distortion Bernstein identified is not an accounting trick; it is a feature of how young AI cloud providers use equity to subsidize their supply chain. The signature is in the funding, not the compute.
Here is the critical takeaway for the next week. Track three signals: First, Core Scientific’s management response—any denial will be tested by next quarter’s cash flow statement. Second, CoreWeave’s next funding round: if closed at a lower valuation, the kickers become less valuable. Third, other miners with AI deals (e.g., Hut 8, Iris Energy) may face similar scrutiny. If Bernstein’s analysis catalyzes wider sell-side reports, the entire “miner AI” sector will be revalued downward. Do not confuse a VC-funded subsidy with a new revenue stream.
My on-chain data background tells me that when the white paper is full of fine print, the code is the only truth. Here, the contract is the code. Core Scientific may be a fine company, but the structural distortion in its AI returns means the market is paying for a castle built on borrowed sand. Watch the capital flows, not the headlines. The data always wins.
