Core Scientific, Inc. CORZ

Key data

The business

Core Scientific builds and runs big warehouses full of computers. For years those computers were its own, mining bitcoin. The plan now is to pull the miners out and rent the buildings, power, and cooling to companies running artificial intelligence, a business it calls high-density colocation, and that is now most of what it earns. The asset is what it always was: fenced land, a substation, and hundreds of megawatts of contracted electricity, the scarce thing everyone building AI wants and cannot get quickly. Core's edge is that it already owns the power and the shells and is converting them faster than a rival could permit and build from scratch. The catch is who rents them. Nearly all the new space sits under one hyperscale tenant on long contracts, and that tenant's warrants sit on Core's own balance sheet as a liability worth about $2.0 billion. The thesis needs two things true at once: AI demand for power stays hot, and that one tenant keeps writing checks.

The numbers

The last five quarters show the pivot happening in real time. Revenue more than doubled year on year and jumped 43% in the single step from March to June.

Read the bottom line with suspicion. Net income swings from a $937M loss to a $216M profit to a $1.2B loss across quarters where revenue barely moved, because it is dominated by non-cash marks on the tenant warrants. When Core's stock moves, that liability is remarked and runs straight through earnings, and the same accounting flips a positive net income into a negative per-share figure. The cash and the segments are the real story, not the loss per share.

Across the years this is a company that earned money once, in 2021, then spent a bankruptcy and a bitcoin winter working through roughly $4B of cumulative losses. The 2026 half-year loss is again almost entirely warrant marks; the operating loss for the June quarter was $78M, not $1.2B. Underneath, one table carries the whole thesis.

A year ago colocation was a rounding error at 11% margins; today it is $137M a quarter at 59%, while mining has shrunk by two-thirds. Consensus has revenue reaching $1.65B by 2028, roughly triple the current run rate, on capacity coming online. What this memo believes that the tape does not is narrow: the re-rate does not come from the existing tenant, already contracted and paid for, but from the first large lease signed with a second tenant. The print that settles it is a new colocation contract, not next quarter's revenue.

Management

The people running Core have done the hard operational thing well, standing up a real colocation business with sector-leading margins in under two years out of a bankrupt miner, funded by raising long-term debt from $1.1B to $4.3B and building cash to $1.77B. Judge them on one capital decision above all: they chose to stay independent and lever up rather than sell, which makes every holder an owner of that debt and the buildout under it. Insiders have leaned the other way with modest conviction, selling $4.8M over the last year against $658K bought, the largest sales by Todd Duchene; plan status not disclosed. There are no buybacks, and stock compensation runs at a fifth of revenue, rich for an infrastructure business.

How it fails or surprises you

One tenant is the whole book. Nearly all colocation revenue and the backlog that justifies $4.3B of debt sit under a single hyperscale customer whose own funding depends on continued AI capital. If that customer slows, delays a phase, or renegotiates, Core's revenue and its ability to service 11.5% notes crack at once. The print to watch is any 8-K amending a customer contract.

The balance sheet leaves no room (left tail). Core carries $4.3B of debt against negative equity, a 0.89 current ratio, and $294M of negative working capital while still burning cash on the buildout. Operating cash flow is positive but capex dwarfs it. A construction slip, or the rate on the next raise stacked on notes already at 11.5%, is how a growth story becomes a solvency one.

A second large tenant re-rates everything (right tail). Core has leasable power not yet contracted, and the roughly 82 MW going to a Muskogee customer in late 2027 shows a second name is possible. Each new hyperscale lease at 59% margins both diversifies away from the one-tenant risk and proves this is a platform, not a single deal. The market pays nothing for it today; one signed contract changes the story.

Closing thoughts

Nothing in the next quarter's revenue settles this, and the earnings line stays noise as long as the warrants swing. What matters is slower and narrower: that the debt is survivable through the buildout, that the one tenant keeps paying, and then whether a second tenant ever signs. The left tail is real and permanent, a levered balance sheet against a single counterparty, and it is fatter than a 15-times-sales price implies; the right tail, a diversified power platform earning 59% margins across several hyperscalers, is worth multiples of today's price but rides on contracts not yet signed. Call it a survival-gated option: you lose badly if the tenant or the debt breaks, you win big if a second tenant arrives, and judgment, not a percentage, says both tails are live.

The bet is still that Core can rent its converted warehouses to AI companies and that its one giant tenant keeps paying. It breaks the day that tenant stops, or the day the debt costs more than the buildings earn. Watch two numbers together: leased versus unleased megawatts, and cash against the interest bill. When a second tenant's name appears on the leased line, the bet changes from survival to compounding.