For years, Core Scientific filled enormous buildings with specialised
computers competing to produce Bitcoin. The machines changed, the
cryptocurrency price moved and the company’s fortunes rose and fell with
them.
The story nearly ended during the cryptocurrency crash: Core
Scientific filed for Chapter 11 protection in December 2022 and did not
emerge from bankruptcy until January 2024. Its
official emergence announcement
Today, those buildings are being given a different job: housing the
computers that train and run artificial intelligence.
The transformation is more than a fashionable change of description.
In the second quarter of 2026, Core Scientific reported $136.7
million in data-centre colocation revenue, compared with $21.5
million from its own Bitcoin mining. One year earlier, the balance was
almost the reverse: colocation produced only $10.6 million, while
self-mining generated $62.4 million. In other words, more than
four-fifths of its latest quarterly revenue came from providing
infrastructure rather than producing cryptocurrency. Core
Scientific’s second-quarter results
Revenue is not the same as profit. The colocation figure included
$35.1 million of customer electricity costs passed through at no markup,
according to the company’s regulatory filing.
Now a major agreement with chipmaker AMD could make that reinvention
far larger. The result is one of the clearest examples yet of how the AI
boom is reshaping businesses that already control a resource technology
companies urgently need: electricity.
Why an old
Bitcoin mine can interest an AI company
Artificial intelligence may feel like software, but large AI systems
have a very physical appetite. Thousands of advanced chips need
buildings, cooling systems, fibre connections and extraordinary amounts
of dependable electricity.
Getting that electricity is increasingly difficult. A new data centre
can be designed relatively quickly, but securing land, substations,
transmission equipment and permission to connect to the grid can take
years.
Bitcoin miners arrived early at this problem. Their facilities were
deliberately built near large power supplies because mining is an
energy-intensive, around-the-clock activity. That does not make a mining
site instantly ready for AI—high-density GPU systems demand more
sophisticated cooling, networking, reliability and construction—but it
gives the owner a valuable head start.
Core Scientific’s advantage is therefore not the Bitcoin computers
themselves. It is the portfolio beneath and around them: powered land,
electrical connections and teams accustomed to operating large computing
facilities.
The company controlled roughly 2.1 gigawatts of gross utility
power across 11 US data centres at the end of June 2026. About
1.3 gigawatts was considered leasable customer capacity at that point.
Its
regulatory filing explains how it defines and measures that
capacity.
It expanded again in August, paying approximately $444 million in
cash for Polaris DS and securing about 440 megawatts of grid-connected
power in Oklahoma. Core Scientific says it intends to enlarge that
campus further, but the proposed expansion to 1.5 gigawatts remains a
plan rather than operating capacity. Core
Scientific’s Polaris acquisition announcement
CoreWeave proved the
landlord model
Core Scientific’s first major AI customer was CoreWeave, a
specialised cloud company that rents access to powerful chips.
The relationship works somewhat like an industrial landlord and
tenant. Core Scientific supplies the site, electricity, cooling and
physical data-centre infrastructure. CoreWeave brings AI computing
systems and sells access to that computing power to its own
customers.
By June 2026, Core Scientific had contracted approximately
590 megawatts of customer power capacity with CoreWeave
across five sites. Of that total, 395 megawatts had started billing by
quarter-end; the company said the figure reached 437 megawatts by
mid-July. The remaining contracted capacity was still being constructed
or commissioned.
That distinction matters. “Contracted” megawatts describe signed
commitments. “Billable” megawatts describe infrastructure that has
actually been delivered and has begun producing revenue.
The model replaces some of Bitcoin’s volatility with long contracts.
But it also creates a new concentration risk. Core Scientific’s
quarterly filing says CoreWeave supplied all of its colocation
revenue and approximately 77% of total revenue during the first
half of 2026—about 83% in the second quarter alone. If one customer
dominates the business, delays or financial trouble at that customer can
quickly become the infrastructure provider’s problem too.
AMD brings a second
route into the AI boom
On 28 July 2026, Core Scientific and AMD announced a partnership
anchored by 529 megawatts of US capacity, with the
first deliveries scheduled for 2027 and the full amount expected by the
end of 2028. The companies said they would collaborate on facilities
designed for AMD Instinct AI accelerators, EPYC processors and AMD’s
ROCm software platform. Core
Scientific and AMD announcement
The signed leases run for 15 years. Regulatory filings show that AMD
directly leased 377 megawatts across three sites, while an unnamed
AI-cloud operator leased another 152 megawatts at two sites with AMD
providing credit support. Importantly, AMD is securing infrastructure
for customer deployments; the announcement does not mean Core Scientific
is purchasing and operating 2.5 gigawatts of AMD chips itself. The
larger figure includes a conditional right to reserve up to another
1,925 megawatts through December 2028, rather than capacity already
under firm lease. Core
Scientific’s regulatory filing gives the contract structure, while
Reuters’
independent report explains the commercial rationale.
This could diversify Core Scientific beyond CoreWeave while giving
AMD an answer to Nvidia’s enormous data-centre ecosystem. A chip is
useful only when a customer can find somewhere to power and cool it. By
pairing hardware with reserved infrastructure, AMD can make large
deployments easier to organise.
Core Scientific says the AMD agreements could represent more than
$14 billion of potential base contracted revenue, and
that its total leased portfolio could represent more than $24 billion.
Those are company estimates spread over long contract lives—not cash
already collected, guaranteed profit or a prediction of the company’s
market value.
The difficult
part starts after the announcement
The opportunity is enormous, but so is the construction job.
Core Scientific spent $797.5 million on capital expenditure during
the second quarter alone. It also issued $3.3 billion of senior secured
notes carrying a 7.75% interest rate and reported $1.82 billion of
liquidity at quarter-end. Those numbers show both sides of the
transformation: it has substantial financing, but building AI
infrastructure is capital-intensive and debt is not free.
The company’s reported quarterly net loss of $1.16 billion also needs
context. About $1.05 billion came from a non-cash remeasurement of
warrants and related rights as the share price rose, rather than an
equivalent amount of cash disappearing from operations. Core Scientific
was nevertheless still loss-making under standard accounting. Adjusted
EBITDA, a non-GAAP measure that excludes several items, was positive at
$41.1 million. Neither figure should be read alone.
Execution now matters more than announcements. Core Scientific must
finish construction on schedule, bring electrical and cooling systems
online, manage rising equipment and labour costs, and satisfy customers
that demand extremely high reliability. It must also navigate growing
political concern about data centres’ effects on local grids,
electricity prices and water use. The company says its newer cooling
technology uses relatively little water and that its Texas operations
pay their own electricity and infrastructure costs, but those are claims
communities and regulators will continue to examine. Core
Scientific’s Texas infrastructure statement
A transformation
that is real—but unfinished
Core Scientific is no longer merely promising to enter AI
infrastructure. Its revenue has already shifted decisively toward
colocation, and hundreds of megawatts are already billing.
What remains uncertain is the scale and profitability of what comes
next. The AMD expansion could make Core Scientific one of the more
important independent owners of AI-ready power in the United States. It
could also encounter the familiar hazards of a construction boom:
delays, cost overruns, heavy debt and customers whose own economics
depend on continuing AI demand.
The simplest way to understand the company is no longer as a miner
chasing digital coins. It is becoming a landlord for machines—one whose
most valuable property feature is not the building, but the electricity
already connected to it.
Reporting note
Capacity planned for 2027 and the possible expansion to 2.5 gigawatts
are forward-looking. The $14 billion and $24 billion figures are company
descriptions of potential contracted revenue over long agreement terms;
they are not current revenue, profit or guaranteed receipts. Financial
figures in this article refer to Core Scientific’s quarter ended 30 June
2026.
For an Australian view of the race to build and power AI facilities, read about NEXTDC’s data-centre expansion plans.
Sources and further reading
- Core
Scientific: second-quarter 2026 results - Core
Scientific and AMD: infrastructure partnership announcement - US
SEC: Core Scientific quarterly report for the period ended 30 June
2026 - Reuters:
AMD signs AI data-centre agreement with Core Scientific - Core
Scientific: Texas data-centre infrastructure statement - Core
Scientific: $1 billion development financing facility - Core
Scientific: emergence from Chapter 11 in January 2024 - Core
Scientific: completed acquisition of Polaris DS
FutureTechDose covers biotechnology, AI, data-centre and
energy-sector research and industry progress for a general audience.
This article is informational and does not provide medical or investment
advice.
Leave a Reply