What if the same power lines that once kept thousands of Bitcoin miners humming through the night suddenly became the backbone of America’s next wave of artificial intelligence? That question stopped being theoretical for Core Scientific in the second quarter of 2026. The company that filed for bankruptcy with barely four million dollars in cash has now reported roughly 1.1 gigawatts of leased customer capacity and more than $24 billion in potential long-term revenue. On paper the transformation looks spectacular. In practice it arrives with a heavy price tag and a growing mountain of debt that will decide whether this comeback sticks or stalls.
From Bankruptcy Court to AI Powerhouse
I’ve watched a lot of crypto companies try to reinvent themselves. Most fail. Core Scientific is different because it never fully lost the one asset that mattered: energized sites with real grid connections. When Bitcoin prices collapsed and electricity costs spiked in late 2022, the company had already poured capital into land, substations, and facilities across five states. Those assets could not generate enough cash to stay solvent, but they remained standing. A Texas bankruptcy court confirmed the reorganization plan in January 2024. Core Scientific walked out lighter by about $400 million in debt after converting equipment financing and convertible notes into equity, and it kept 724 megawatts of operating capacity.
That capacity suddenly looked valuable the moment large language models started demanding denser racks and steadier power. Building a new data center from scratch can take years of studies, permits, and construction. Core Scientific already owned energized sites designed for power-hungry hardware. Management simply changed the sales pitch. Instead of selling hash rate, they began selling high-density colocation to AI operators who needed speed to market.
The CoreWeave Catalyst That Changed Everything
The commercial shift started with one customer. In June 2024 Core Scientific signed 12-year agreements covering about 200 megawatts of high-performance computing infrastructure. The company estimated more than $3.5 billion of cumulative revenue under those initial contracts. Expansions followed quickly. CoreWeave’s contracted capacity grew to roughly 590 megawatts, and Core Scientific now ties those deals to approximately $10.2 billion in potential revenue over their full terms.
What made the arrangement smarter than a simple hosting contract was the funding structure. CoreWeave agreed to cover portions of the construction cost, with those amounts later credited against future payments. In the first six months of 2026 alone, that customer funding reached $180.9 million. The relationship even survived a failed acquisition attempt. Core Scientific rejected a $5.75 cash offer in 2024, later agreed to an all-stock deal valued near $9 billion, then watched shareholders vote it down in October 2025 over concerns about the fixed exchange ratio and exposure to CoreWeave’s own share price. The commercial contracts stayed intact. That resilience matters more than any single deal announcement.
AMD Arrives and the Pipeline Expands Dramatically
July 28, 2026 brought the second major pillar. Core Scientific announced infrastructure agreements with AMD covering approximately 530 megawatts across five U.S. sites under 15-year terms. The company links that initial capacity to more than $14 billion in potential base contract revenue. Deployments are scheduled to begin in 2027 and will support customers running AMD Instinct accelerators, EPYC processors, and the ROCm software stack.
AMD also received reservation rights on another 1.925 gigawatts. If every reserved megawatt eventually converts into a signed lease, the partnership could reach 2.5 gigawatts. That larger number remains conditional. Reservation rights are not executed leases. Realizing them will depend on customer demand, available grid capacity, construction progress, and additional financing. Still, the warrants AMD received—up to 30 million shares at $23.47—show both sides have skin in the game. About 6.5 million of those warrants already vested after the July leases closed.
Some of the early capacity went to an AI infrastructure operator called Neocloud. AMD provided a form of credit support tied to the equipment, though the filings stop short of calling AMD an unconditional guarantor of every payment. The net effect is clear: Core Scientific has reduced its reliance on a single customer at the contracted-capacity level, even if concentration risk remains material.
When Colocation Overtook Mining
Numbers from the second quarter of 2026 make the shift impossible to ignore. Total quarterly revenue hit $164.2 million. Colocation contributed $136.7 million. Self-mining brought in only $21.5 million. A year earlier, colocation had produced just $10.6 million. The segment now generates roughly $80 million in quarterly gross profit and a margin near 59 percent. Self-mining, by contrast, recorded a gross loss of about $12.2 million.
By the end of the quarter, 437 megawatts were already generating billable revenue. Management calculated that capacity as roughly $635 million in annualized GAAP hosting revenue. Adjusted EBITDA reached $41.1 million. Those are encouraging operating metrics, yet the GAAP bottom line still shows a net loss of approximately $1.16 billion and an operating loss of $78.5 million. Nearly $1.05 billion of the net loss came from non-cash fair-value changes on warrants and contingent value rights—accounting noise driven mainly by movements in the share price rather than cash leaving the building.
I’ve found that investors sometimes overreact to those non-cash charges. The more useful signal sits in the operating numbers and the cash position. Core Scientific held about $1.8 billion in cash, cash equivalents, and digital assets at quarter-end. That war chest is real. The question is how long it lasts once construction spending accelerates.
The $24 Billion Figure Needs Context
Let’s be clear about what the $24 billion represents. It is not cash in the bank. It is not guaranteed profit. It is the company’s estimate of potential revenue across contracts that can stretch as long as 15 years. Recognition depends on finishing construction, delivering capacity on schedule, maintaining uptime, and keeping customers paying. Any slippage in those variables shrinks the eventual number.
The turnaround rests on a genuine shift in revenue, but its completion remains forward-looking. Core Scientific has exchanged direct exposure to Bitcoin prices and mining difficulty for construction, financing, and customer-credit risks.
That trade-off sits at the heart of the story. The company sold $208.3 million of Bitcoin in the first quarter alone to help fund the AI transition—a decisive break from the old accumulate-and-hold mindset. The same sites that once mined coins now host high-density racks. The economic logic is straightforward: AI operators will pay more for reliable power and faster deployment than Bitcoin mining currently returns.
Capital Spending and the Debt Mountain
Transforming sites is expensive. Core Scientific spent $954.2 million on property and equipment in the first half of 2026. It also completed a roughly $232.5 million acquisition of land and development rights for a proposed 430-megawatt site in Hunt County, Texas. To fund the expansion, the company issued $3.3 billion of senior secured notes in May. The notes carry a 7.75 percent interest rate and mature in 2031. Long-term debt climbed to approximately $4.3 billion by June 30, up from about $1.06 billion at the end of 2025.
First-half operating cash flow looked healthier than pure operations alone would suggest. Bitcoin sales, customer construction funding, and working-capital swings all helped. Those sources are useful, but they do not prove that recurring colocation income can yet carry the full development program on its own. The company still needs to bring more CoreWeave capacity online and start the AMD-related deployments in 2027. Delivery schedules, construction costs, customer performance, and the conversion of reservation rights into actual leases will determine how much of the advertised backlog turns into recognized revenue.
What the Market Is Pricing In
On August 20 the shares traded near $18.72, giving Core Scientific a market capitalization of roughly $6.1 billion. That price sits several times above the $3.44 close on the first day of its January 2024 relisting. The market is clearly assigning value to the contracted pipeline. Yet the same market is also watching the debt load and the execution timeline. Any meaningful delay in bringing megawatts online, or any deterioration in customer credit quality, would test that valuation quickly.
Similar economics are pushing other miners to repurpose energized sites. Weakening mining margins make the AI alternative look more attractive. Core Scientific simply moved earlier and secured larger contracts. Whether that first-mover advantage holds depends on execution more than on further deal announcements.
The Real Risks That Still Sit on the Table
Four issues stand out. First, construction risk. Converting former mining sites and developing new campuses requires capital, skilled labor, and timely grid upgrades. Second, the debt itself. Interest expense on $4.3 billion of long-term debt is no small matter, especially at a 7.75 percent coupon on the recent notes. Third, customer concentration. Even with AMD in the mix, a limited number of counterparties still support most of the colocation revenue. Fourth, the gap between reserved capacity and executed leases. The 2.5-gigawatt AMD vision remains partly aspirational until more contracts are signed and financed.
- Construction spending must stay on schedule and within budget
- Debt service has to be covered by rising operating cash flow
- Customer performance under long-term contracts remains critical
- Reservation rights need to convert into actual leased capacity
None of these risks is fatal on its own. Together they form the test that the $24 billion figure must pass. I’ve seen companies with impressive backlogs still struggle when the cash conversion cycle stretches longer than expected. Core Scientific’s cash balance provides a cushion, yet the company has already signaled that construction spending will continue at a high level.
How the Business Model Has Fundamentally Changed
The old model was simple: buy or host miners, consume electricity, sell Bitcoin. Margins swung with coin price and network difficulty. The new model is closer to a traditional data-center operator: secure power and land, build or convert facilities, sign multi-year leases, and collect monthly hosting fees. Gross margins on colocation already look healthy. The challenge is the capital intensity required to reach scale and the time lag between spending and full revenue recognition.
Perhaps the most interesting aspect is the speed of the pivot. In less than two and a half years the company went from Chapter 11 to a contracted AI pipeline that dwarfs its remaining mining operations. That speed was possible only because the physical assets already existed. Most pure-play data-center developers would still be waiting for interconnection studies.
Looking Ahead to 2027 and Beyond
The immediate milestones are clear. More of the CoreWeave capacity needs to come online and begin generating the higher billable revenue already visible in the second-quarter results. AMD-related deployments are scheduled to start in 2027. Between those two programs sits the bulk of the near-term revenue growth story. Longer term, the conversion of AMD reservation rights and any additional customer wins will determine whether the company can push past the current 1.1-gigawatt leased figure.
Management has repeatedly stated that the transition requires heavy spending. Investors will need to track two numbers in particular: the rate at which capital expenditures convert into billable megawatts, and the trajectory of adjusted operating cash flow once the non-recurring funding sources normalize. If those two lines move in the right direction, the debt load becomes more manageable. If they lag, the interest burden starts to bite harder.
In my view the company has already cleared the hardest credibility hurdle. It proved that former mining infrastructure can attract serious AI customers at meaningful scale. The next hurdle is pure execution and balance-sheet discipline. That is a different kind of test, but it is the one that ultimately decides whether a $24 billion pipeline turns into durable shareholder value.
Why This Story Resonates Beyond One Company
Core Scientific’s path is not unique in outline. Other Bitcoin miners face the same pressure from thinner margins and the same opportunity in AI demand for power. What sets this case apart is the size of the contracts already signed and the speed of the revenue mix shift. Colocation moved from a rounding error to the dominant revenue source in a single year. That kind of rapid change forces the market to reassess the entire sector.
At the same time, the debt increase serves as a reminder that growth of this magnitude rarely comes free. The 7.75 percent notes and the jump in total leverage are the visible cost of converting potential into operating capacity. Future filings will show whether the returns on that capital justify the risk. For now the numbers are still early, the contracts are long-dated, and the construction schedule remains the critical path.
The company that once struggled to keep the lights on during a Bitcoin winter now sits at the center of a multi-year AI infrastructure build-out. That irony is hard to miss. Whether the story ends as a textbook turnaround or a cautionary tale about leverage will depend on the next several quarters of delivery, not on any further press releases. The $24 billion figure is impressive. Turning it into reliable, billable computing capacity without letting the debt outrun the earnings is the harder—and more important—task still ahead.
Key Numbers at a Glance
| Metric | Figure | Context |
| Leased customer capacity | 1.1 GW | End of Q2 2026 |
| Potential contracted revenue | >$24 billion | Across long-term deals |
| Q2 colocation revenue | $136.7 million | Up from $10.6 million year earlier |
| Q2 self-mining revenue | $21.5 million | Gross loss of ~$12.2 million |
| Long-term debt | ~$4.3 billion | Up from ~$1.06 billion end-2025 |
| First-half CapEx | $954.2 million | Property and equipment |
| AMD initial capacity | 530 MW | 15-year agreements, five sites |
| Cash and digital assets | ~$1.8 billion | Quarter-end |
Those figures capture the scale of both the opportunity and the challenge. The capacity is real. The contracts are signed. The spending is already underway. What remains is the conversion of those contracts into steady cash flow that can service the new debt load while still funding further growth. That is the debt test Core Scientific now faces, and it is the test that will determine whether this AI comeback becomes a lasting success or a temporary peak.
For anyone following the intersection of crypto infrastructure and artificial intelligence demand, the next set of quarterly results will matter more than the latest deal headline. Watch the megawatts that actually start billing. Watch the cash conversion. Watch the interest coverage. Everything else is still potential.
