Core Scientific pitches 1.1 GW AI colocation pipeline tied to >$24B in potential revenue
Q2 2026 results show colocation is now the profit engine, but the buildout is being financed with roughly $4.3B of long-term debt.
Core Scientific is marketing its shift from Bitcoin mining to AI data center colocation as a 1.1-gigawatt leased-power pipeline tied to more than $24 billion in potential long-term contract revenue. The setup is already visible in Q2 results, but the trade still hinges on whether a debt-funded construction ramp converts contracted megawatts into billable capacity on schedule.
Key Takeaways
- Core Scientific disclosed about 1.1 GW of leased customer power capacity tied to more than $24 billion in potential contract revenue over terms as long as 15 years, which it does not frame as cash received or guaranteed profit.
- Q2 2026 revenue was $164.2 million, with high-density colocation contributing $136.7 million versus $21.5 million from digital asset self-mining.
- Colocation produced roughly $80 million of segment gross profit at about a 59% margin, while self-mining posted an approximate $12.2 million gross loss.
- The AI buildout has been paired with heavy capex and leverage, including $954.2 million of property-and-equipment spend in 1H 2026 and long-term debt of about $4.3 billion as of June 30, 2026.
Core Scientific’s AI Pivot Hits Scale: 1.1 GW Leased, >$24B in “Potential” Contract Revenue
Core Scientific is positioning its miner-to-AI rotation as a capacity business: about 1.1 gigawatts of leased customer power capacity that the company associates with more than $24 billion in total potential contracted revenue under long-term agreements lasting up to 15 years. Mechanically, that “leased power” figure is the electrical envelope customers have committed to, which is a proxy for how much high-density compute can be hosted once sites are built out and energized for AI workloads.
The catch is in the company’s own framing. The >$24 billion number is described as potential revenue spread over long-dated contracts, not cash received and not guaranteed profit. It only turns into recognized revenue if Core Scientific completes construction, brings capacity online, meets service requirements like uptime, and customers keep operating and paying across the contract life.
Q2 Numbers Show Colocation Took Over—Even as GAAP Losses Stayed Deep
The income statement shows the pivot is no longer theoretical. Q2 2026 revenue totaled $164.2 million, led by high-density colocation at $136.7 million, while digital asset self-mining contributed $21.5 million.
Colocation’s year-over-year step-up was steep, rising to $136.7 million from $10.6 million a year earlier. The segment produced about $80 million of quarterly gross profit, implying a margin near 59%. Self-mining, by contrast, recorded a gross loss of approximately $12.2 million in the quarter.
Operational progress is also being measured in billable capacity, not just contracted megawatts. Core Scientific said 437 MW were generating billable revenue by the end of Q2 2026, which it calculated as about $635 million in annualized GAAP hosting revenue.
GAAP profitability stayed negative despite the colocation ramp. Core Scientific reported an operating loss of $78.5 million and a quarterly net loss of roughly $1.16 billion. Most of that net loss, about $1.05 billion, was attributed to non-cash fair-value changes involving warrants and contingent value rights, driven mainly by movements in Core Scientific’s share price.
The company also reported Adjusted EBITDA of $41.1 million, a non-GAAP measure that excludes several expenses. It is useful for tracking operating trajectory, but it is not equivalent to net income or unrestricted cash flow, which matters when the business is simultaneously trying to finance a large construction program.
The Customer Map: CoreWeave’s Expanded Footprint and AMD’s 530 MW Deal
Core Scientific’s AI colocation story is concentrated in a small set of counterparties, with CoreWeave as the original anchor tenant. In June 2024, the companies signed 12-year agreements covering about 200 MW of high-performance computing infrastructure, which Core Scientific estimated at more than $3.5 billion of cumulative revenue under the initial contracts. Subsequent expansions increased CoreWeave’s contracted capacity to roughly 590 MW, and Core Scientific now associates those agreements with about $10.2 billion in potential revenue over their terms.
Those contracts also shaped the financing mechanics. During the first six months of 2026, CoreWeave funded $180.9 million of Core Scientific’s capital expenditures, with the amounts credited against future hosting payments. That reduces near-term cash burden, but it also ties part of the buildout economics to the durability of a single customer relationship.
On July 28, 2026, Core Scientific announced a broader infrastructure partnership with AMD. The initial package covers about 530 MW across five U.S. sites under 15-year agreements, which the company associates with more than $14 billion in potential base contract revenue. Deployments are scheduled to begin in 2027 and are described as supporting customers using AMD Instinct accelerators, EPYC processors, and ROCm software.
The headline upside is larger but conditional. AMD also received reservation rights for another 1.925 GW that could bring the partnership to about 2.5 GW if converted, but reservation rights are not executed leases. Core Scientific tied conversion to customer demand, available grid capacity, construction progress, and additional financing.
The AMD arrangement also includes equity-linked economics. AMD received warrants to purchase up to 30 million Core Scientific shares at $23.47 each, and about 6.5 million warrants vested after related leases were executed in July, per the quarterly filing. Some initial capacity was leased to AI infrastructure operator Neocloud, and AMD entered a credit-support arrangement connected to equipment installed for Neocloud, though the filed contracts do not describe AMD as an unconditional guarantor of every Neocloud payment.
Milestones Traders Can Track Into 2027: Billable MW Ramp, Financing Needs, and Concentration Risk
The near-term scoreboard is whether leased customer power capacity converts into billable megawatts beyond the 437 MW generating revenue at the end of Q2 2026, and whether the company updates its annualized GAAP hosting revenue run-rate as more sites come online.
Financing is the other gating item. Core Scientific spent $954.2 million on property and equipment in 1H 2026 and long-term debt rose to about $4.3 billion by June 30, 2026 after the company issued $3.3 billion of senior secured notes in May at 7.75% interest due 2031. The company reported about $1.8 billion in cash, cash equivalents, and digital assets at quarter-end, but it also flagged that first-half operating cash flow benefited from Bitcoin sales, customer construction funding, and working-capital changes.
On the customer side, the AMD timeline is explicit: deployments are scheduled to begin in 2027. The market will likely treat any conversion of AMD’s 1.925 GW of reservation rights into executed leases as a separate catalyst, because it would move conditional capacity into contracted capacity.
Concentration risk remains the quiet variable. CoreWeave’s footprint is still large at roughly 590 MW and about $10.2 billion in associated potential revenue, even after the AMD expansion reduces reliance at the contracted-capacity level. Any new large-customer signings that diversify revenue would matter as much as another backlog headline.
My Read: The Backlog Is the Story—But the Conversion Rate Is the Trade
The threshold that matters is the gap between “leased customer power capacity” and “billable revenue capacity.” Core Scientific has already proven demand and pricing power well enough to flip colocation into the dominant revenue line, but the >$24 billion figure is only as real as the construction schedule that turns megawatts into invoices.
If the billable MW ramp keeps pace while incremental financing needs stay contained relative to liquidity, the setup starts to look structural rather than narrative-driven. If debt and capex keep rising faster than billable capacity, the backlog reads less like contracted revenue and more like an execution obligation with a clock attached.
