Core Scientific's $14bn AMD Deal Vindicates Its Rejection of CoreWeave

The headline numbers, a $10.5bn market capitalization, $24bn in total contracted revenue across the company, a $1.16bn quarterly net loss, are not really the story. The story is what those numbers are actually made of.

What the CoreWeave contract actually looks like

Nine months ago, Core Scientific shareholders rejected CoreWeave's $9bn all-stock buyout for the second time, on advice from proxy advisers that the company was worth more independent. Today, colocation revenue is still derived entirely from that one customer. Per the company's own 10-Q, CoreWeave accounted for approximately 77% of total revenue for the six months to 30 Jun.

The CoreWeave lease is closer to a piece of structured real estate finance than a typical technology contract. It runs 12 years with two five-year options, take-or-pay, meaning CoreWeave pays whether or not it uses the capacity, at a fixed cost with an annual escalator, and CoreWeave itself pays for capex, power and utilities, funding up to $1.5mn per megawatt of buildout, credited back against hosting fees until repaid. Core Scientific holds liens on the infrastructure assets, excluding the GPUs themselves. Revenue from the contract is paid directly into a lockbox tied to the $3.3bn of senior secured notes due 2031, according to the company's own investor materials, meaning CoreWeave's rent effectively services Core Scientific's bondholders before it does anything else. That is the arrangement that let a company with roughly $4.4bn of debt and a single paying customer finance a data-centre buildout at all.

AMD deal is not the same kind of deal

The AMD leases cover 377 megawatts (MW) at Pecos, Texas; Muskogee, Oklahoma; and Hunt County, Texas. A further 152MW at Auburn, Alabama and Dalton, Georgia is leased not to AMD directly but to a Neocloud operator, with AMD holding only a Credit Support Agreement giving it the right, not the obligation, to step in if that operator defaults. None of the five sites delivers before 2027, and full delivery of the 530MW runs to the end of 2028. Nothing here has generated a dollar of revenue yet.

In place of a straightforward rent negotiation, AMD received warrants for up to 30mn Core Scientific shares at $23.47, vesting at 12,222 shares per megawatt leased, with roughly 6.5mn shares already vested. Core Scientific's own quarterly filing says plainly: "The Company is evaluating the accounting treatment, including measurement and classification." In the same filing that discloses the deal, the company that issued the warrants has not yet decided whether to book them as equity or as a liability.

The net loss is a warrant story, not new

Core Scientific's $1.16bn net loss this quarter was driven, by the company's own account, by the rising fair value of its warrant liabilities as its own share price climbed, an accounting mechanism rather than a cash cost. It is tempting to treat that as new, a side effect of the AMD warrants specifically. It isn't. The comparable quarter a year earlier carried a warrant revaluation charge of roughly $910mn, on a stock that was also rising. This is a recurring feature of Core Scientific's capital structure, not a one-off consequence of the AMD deal, and given the company has not yet settled how to classify the new AMD warrants, there is no reason to expect this quarter's pattern won't repeat once those shares are on the books, too.

The Bitcoin miner that mostly isn't, any more

Core Scientific's own 16-slide earnings deck does not contain a single figure on Bitcoin held, mined, or produced. Management describes itself as "repurposing remaining bitcoin mining facilities" and operates a self-mining fleet at two sites, down from a business that once ran mining operations across the country. The deck's own choice of peer group, TeraWulf, Cipher, Galaxy, Applied Digital, Hut 8 and Iren, names the same cohort of former miners chasing AI infrastructure revenue that Sandmark has tracked signing more than $100bn of comparable deals over the past two years. Core Scientific is simply the furthest along.

What hasn't changed is the risk shareholders bet on when they said no to CoreWeave. The company's own quarterly filing says its colocation revenue still comes from a single customer, and that its strategy depends on winning others. AMD is that additional customer, on paper, worth more in headline dollars than the entire company was offered for outright 18 months ago. But none of it bills until 2027, some of it isn't even AMD's own credit risk, and it arrived paid in stock the company itself hasn't finished deciding how to count.

(Corrected, clarifying that Core Scientific disclosed it had not yet decided how to classify the AMD warrants in the same filing that disclosed the deal itself, not nine months afterward.)