Meta Data Center Junk Bond Draws Four Times Its Deal Size

Meta’s first data center junk bond attracted about $10 billion of orders for a $2.276 billion offering, despite carrying an 8.25% yield.

The deal finances CleanSpark’s conversion of its Sandersville, Georgia, campus into a 175-megawatt AI data center leased to a Meta subsidiary. Investors received a large yield premium, while CleanSpark obtained most of the capital needed to complete the project.

That combination makes the sale more than another oversubscribed technology bond. It shows how high yields and long hyperscaler leases can move speculative-grade infrastructure debt, even as investors question the scale of AI spending.

The bond market is therefore sending two messages at once. Demand for contracted AI infrastructure remains strong, but investors want more compensation when construction and operating risk sit outside Meta’s balance sheet.

Meta Data Center Junk Bond Prices After Orders Reach $10 Billion

The offering cleared because investors received a high yield alongside lease payments supported by Meta.

CleanSpark’s CSDC Finance I subsidiary priced $2.276 billion of senior secured notes due in 2031. The notes carry a 7.875% coupon and were sold at 98.5% of their face value.

That discount increased the yield to approximately 8.25%. Bloomberg’s bond demand reporting said the yield was about 1.75 percentage points above the average for BB-rated issuers it tracks.

Money managers submitted roughly $10 billion of orders. That was more than four times the final offering, allowing CleanSpark to increase the transaction from the initially proposed $2.227 billion.

The final size matters because it closely matches the project’s planned debt funding. CleanSpark’s September materials showed $2.227 billion of debt in its financing model before the transaction was priced.

The company’s pricing announcement said the offering should close on September 25, subject to customary conditions. The securities were offered privately to eligible institutional buyers and certain non-US investors.

CleanSpark plans to use the proceeds for three purposes. It will fund remaining Sandersville construction, reimburse earlier equity contributions, and establish debt-service reserves.

The bonds are secured at the project level rather than issued directly by Meta. Investors therefore rely on the facility’s assets, its contractual cash flow, and protections contained in the financing documents.

Meta’s role remains central to the credit story. A wholly owned subsidiary, Anviran LLC, has agreed to lease the entire facility for an initial 20-year term.

Meta will guarantee rent and operating expenses, according to CleanSpark’s investor presentation. That support connects the project’s revenue to an investment-grade technology company, although the borrower itself remains speculative grade.

The structure separates several risks that often appear together in corporate borrowing. Meta supports the lease, CleanSpark develops the campus, and the project subsidiary owes the bondholders.

That separation explains why the debt offers a much higher yield than bonds issued directly by Meta. Bondholders receive access to Meta-linked cash flow, but they also accept project completion and structural risk.

The order book does not erase that distinction. It shows that investors found the offered compensation attractive after considering it.

CleanSpark’s original yield discussion reportedly centered near 8.5%. Stronger orders allowed the deal to price at a lower yield, though it remained expensive relative to ordinary BB debt.

That movement provides the clearest measure of demand. Investors did not merely express interest in another AI financing. They accepted a tighter return after reviewing the lease, construction plan, and collateral package.

The result also establishes a public benchmark for future projects connected to Meta. Developers and lenders can now compare new speculative-grade facilities against an 8.25% starting yield and a heavily oversubscribed order book.

A Long Meta Lease Makes the Risk Financeable

The bond sale converts a future Meta rent stream into construction capital available today.

CleanSpark signed the Sandersville lease in July 2026 without initially identifying its technology tenant. A September SEC filing later disclosed that Anviran is a wholly owned Meta subsidiary.

The agreement covers 175 megawatts of critical IT load. Critical IT load measures the electricity available to computing equipment, excluding much of the supporting facility overhead.

CleanSpark expects the first rent to begin during the fourth quarter of 2027. Its model assumes the first completed network hall starts generating revenue on November 30, 2027.

The lease has a 20-year base term and two five-year extension options. A separate 12-month option also appears in CleanSpark’s September project materials.

The initial contract represents approximately $6.6 billion of expected lease payments. CleanSpark estimates average annual net operating income near $330 million after the facility reaches its operating phase.

Its Sandersville presentation also shows a 3% annual rent escalator. That provision raises contractual payments over time rather than leaving revenue fixed for two decades.

The agreement is structured as a triple-net lease. Under that arrangement, the tenant generally carries operating costs that might otherwise fall to the property owner.

CleanSpark expects the project to operate with a net operating income margin near 100%. That figure is a company projection, not an independently verified operating result.

The facility must still be built and delivered before those economics become real. Its expected development cost is about $11.9 million for each megawatt of critical IT capacity.

CleanSpark had previously estimated landlord costs between $10 million and $12 million per megawatt. The final project plan sits close to the top of that range.

The company has said it prepaid for long-lead equipment required to meet the delivery schedule. It also reported that the anticipated equity portion of the project had been funded.

Those steps reduce two common construction pressures, but they do not eliminate execution risk. Equipment integration, power availability, regulatory approvals, labor, and contractor performance remain essential.

CleanSpark disclosed these uncertainties in the lease filing. It warned that delays, power interruptions, financing constraints, and changing AI requirements could affect the project.

The long lease matters because construction lenders need predictable future income. A developer with no committed tenant would face a far harder financing process and potentially more expensive capital.

Meta’s guarantee improves that equation. It allows investors to evaluate expected rent against scheduled interest and principal payments.

CleanSpark’s illustrative model showed lease-support coverage beginning around 2.9 times after rent starts. The ratio compares the supporting lease income with required project debt service.

Coverage was projected to decline gradually before rising as principal amortizes. Those calculations remain forecasts that depend on timely completion and continuing lease performance.

The bonds mature in five years, well before the base lease expires. That difference gives bondholders a long contracted revenue tail behind a shorter financing instrument.

However, maturity before lease expiration does not automatically make the notes safe. The project must generate cash on schedule, preserve collateral value, and refinance or repay according to its final documents.

The lease therefore turns Meta’s demand for capacity into financeable revenue, but the developer still carries the difficult middle step. CleanSpark must transform a former Bitcoin mining campus into infrastructure suitable for Meta’s AI workloads.

CleanSpark Is Trading Bitcoin Exposure for Data Center Execution Risk

The financing tests whether a Bitcoin miner can become an institutional AI infrastructure developer without losing control of its capital needs.

CleanSpark built its public identity around Bitcoin mining, a business exposed to cryptocurrency prices, network difficulty, energy costs, and equipment efficiency. Sandersville marks a major expansion into contracted data center revenue.

The shift has a logical foundation. Bitcoin miners already control power agreements, land, substations, and operating sites that can be valuable during the AI capacity shortage.

Yet a mining campus is not automatically ready for hyperscale AI. Large AI deployments demand different cooling, networking, redundancy, security, and construction standards.

The $2.276 billion borrowing reflects the cost of that conversion. CleanSpark is funding a purpose-built facility rather than simply placing Meta servers inside existing mining buildings.

The Sandersville development carries 175 megawatts of contracted critical IT load. That represents a substantial facility, but it is smaller than Meta’s multigigawatt campuses in Louisiana and Texas.

Its narrower scope can help execution. CleanSpark can deliver the project in phases, collect rent after individual milestones, and focus on one fully committed tenant.

The project also concentrates risk. A single tenant supports the facility, while one development schedule determines when the rent stream begins.

Meta’s guarantee strengthens payment quality once lease obligations apply. It does not construct the buildings or guarantee that every completion milestone will arrive on time.

CleanSpark offers a limited parent completion guarantee during development. That support gives creditors recourse related to finishing the project, but it also creates exposure for CleanSpark beyond the project subsidiary.

The company’s transition depends on balancing those obligations with its existing operations. Heavy secured borrowing can finance growth, yet it also limits room for mistakes.

High-yield investors were paid for this uncertainty through the 8.25% yield. The premium acknowledges that a creditworthy tenant does not turn a developing facility into a completed asset.

This distinction matters for other Bitcoin miners pursuing AI customers. Control of electricity has become valuable, but credible delivery requires engineering capacity and large amounts of capital.

Some miners may secure leases and financing. Others could discover that their sites lack suitable fiber, grid reliability, cooling options, or expansion space.

CleanSpark has additional ambitions beyond Georgia. Its July announcement said the same unnamed tenant had entered an exclusivity arrangement covering the company’s Texas portfolio.

That portfolio included 718 acres and up to 885 megawatts of secured and planned power capacity. Sandersville could therefore become the first transaction in a much larger Meta relationship.

An exclusivity arrangement is not the same as a signed lease. Future Texas development still requires final contracts, technical planning, financing, and construction decisions.

Investors should resist treating the full Texas portfolio as committed Meta capacity. The Georgia lease is binding, while the broader relationship remains less certain.

The distinction captures CleanSpark’s central trade. Contracted AI revenue offers greater visibility than Bitcoin mining, but the conversion demands leverage and long execution schedules.

Its debt investors appear comfortable with that exchange at the offered yield. Equity investors face a broader question because CleanSpark must manage both its existing business and its new construction obligations.

If Sandersville arrives on time, the project gives CleanSpark a reference facility backed by Meta. That could improve its credibility with other tenants and financing partners.

If delivery slips, the same project could expose weaknesses in the miner-to-data-center strategy. It would also test the protections embedded in Meta’s lease and CleanSpark’s completion guarantee.

Blowout Demand Does Not Mean Investors Ignored the Risk

Fourfold demand signals confidence in the deal’s return, not a judgment that AI infrastructure carries little risk.

Oversubscribed bonds often produce a simple headline: investors want the asset. The Meta data center junk bond requires a more precise interpretation.

Buyers wanted the bonds at an 8.25% yield, supported by a long lease and Meta’s payment guarantee. Demand at a substantially lower return might have looked very different.

The premium was about 1.75 percentage points above comparable BB debt tracked by Bloomberg. That gap represents meaningful additional compensation over five years.

Investors have recently demanded larger premiums for data center financings. A growing pipeline of offerings has increased supply, while concerns about technology spending have made credit selection more important.

The strongest hyperscalers can issue conventional corporate bonds. Developers such as CleanSpark must often rely on secured project debt tied to specific facilities and tenant contracts.

That creates a spectrum of AI credit rather than one uniform asset class. A direct Meta obligation is different from a CleanSpark project bond, even when Meta occupies the building.

The deal’s security package should improve recovery prospects if problems emerge. Still, data centers are specialized assets whose value depends on power, connectivity, tenant demand, and technical relevance.

Technology can also change faster than real estate financing. A facility designed around current hardware must remain useful as rack densities, cooling systems, and network designs evolve.

The 20-year lease transfers part of that obsolescence risk to Meta because the tenant has committed to long-term payments. Contract language, termination rights, and completion conditions determine how complete that transfer is.

Public materials provide the economic outline, but private offering documents contain more detail. Outside investors cannot assume every project risk sits with Meta.

Construction creates the first uncertainty. CleanSpark targets initial delivery in late 2027, leaving more than a year before rent begins.

During that period, the project consumes capital without receiving its full contracted income. Delays can increase interest, labor, and equipment costs.

Power creates another uncertainty. AI data centers require sustained electricity at a scale that can strain regional infrastructure and approval processes.

CleanSpark’s existing relationship with the Sandersville site provides a foundation. However, AI infrastructure requires reliability and redundancy beyond ordinary access to power.

Tenant concentration presents a third risk. Meta’s financial capacity is strong, but the facility has no diversified customer base if the relationship changes.

The agreement’s long duration protects CleanSpark from ordinary vacancy risk. It also ties the project’s value closely to one company’s future infrastructure strategy.

Meta’s own commitments are becoming enormous. Its June-quarter filing reported approximately $278.99 billion of operating and finance leases that had not yet commenced.

In July, Meta entered additional data center leases carrying roughly $68 billion of obligations. Those contracts were expected to begin during 2027 and 2028.

Meta also reported $349.31 billion of noncancelable contractual commitments at June 30. The total included cloud capacity, servers, networks, data centers, and consumer hardware.

These Meta commitments show why developers can secure financing around its leases. They also explain why credit investors are studying aggregate AI obligations more closely.

The company’s resources remain considerable. However, every new lease adds another fixed claim that AI products must ultimately justify through revenue, efficiency, or strategic value.

That does not make the Sandersville debt unsound. It means the bond’s attractive return exists because the market recognizes genuine construction, concentration, and industry risks.

AI Infrastructure Debt Is Moving Beyond Big Tech Balance Sheets

Sandersville shows how the AI buildout increasingly depends on developers and bondholders that sit outside the largest technology companies.

Meta can finance infrastructure from operating cash flow, corporate bonds, leases, joint ventures, and project-level structures. Using several channels preserves flexibility while accelerating construction.

The company used a different model for Hyperion, its massive Louisiana campus. Meta retained a minority stake while Blue Owl-managed funds held the majority position in a joint venture.

That project involved approximately $27 billion of development costs for buildings and long-lived infrastructure. Private debt investors funded part of the capital behind it.

Sandersville is smaller and carries a different credit profile. CleanSpark owns the development platform, its subsidiary issues speculative-grade debt, and Meta leases the completed capacity.

The common element is contractual separation. Meta gains access to infrastructure without placing the entire construction cost into a single direct corporate borrowing.

This approach does not make the economic obligation disappear. Lease payments and guarantees still affect Meta’s commitments, cash flows, and exposure.

It changes who funds construction and how each risk is allocated. Developers take execution responsibilities, lenders provide capital, and Meta commits to use the resulting capacity.

Other hyperscalers have supported comparable structures with specialist developers. Oracle, Amazon, Google, and Microsoft all use combinations of owned facilities, leases, and third-party capacity.

The financing market is expanding because AI campuses require more capital than even cash-rich technology companies want to deploy alone. Power generation, transmission, buildings, servers, and networking create separate funding needs.

By July 22, Alphabet, Amazon, Meta, Microsoft, and Oracle had raised nearly $302 billion through debt and equity during 2026. S&P Global Market Intelligence supplied the data cited in an analysis of hyperscaler borrowing.

A separate Goldman Sachs estimate placed global AI-related bond and loan supply at $489 billion for the year at that point. That total had already exceeded its full-year 2025 estimate.

Large issuance can eventually pressure bond prices. Investors have limited capital, and each new transaction must compete with other corporate and infrastructure debt.

CleanSpark overcame that pressure by offering a combination of yield, collateral, and contracted rent. Its order book suggests those features can still attract substantial institutional money.

However, the offering does not prove every proposed data center will find financing. Tenant quality, completion support, power certainty, and lease terms will separate viable projects from weaker ones.

It also does not prove that the industry’s planned capacity will generate adequate returns. Bondholders primarily need timely payments, while the wider AI economy must justify the infrastructure over many years.

That difference can produce a financing boom before the technology’s final economics become clear. Contracted leases move risk away from developers, but they concentrate long-term obligations at hyperscalers.

Meta’s strategy is to secure capacity before computing demand becomes fully visible. Waiting for certainty could leave the company without enough power or space to train and serve future models.

The cost is reduced flexibility. Twenty-year commitments remain economically important even when project debt stays outside Meta’s corporate balance sheet.

Sandersville therefore sits between two competing pressures. Meta needs faster infrastructure delivery, while credit markets want clearer compensation for the capital and uncertainty involved.

The deal resolved that tension through price. Investors accepted the structure after receiving a yield far above the average for similarly rated issuers.

If future projects offer less protection or lower yields, demand might weaken. If competition for AI capacity intensifies, developers may continue securing finance on favorable terms.

Three Signals Will Show Whether the Deal Set a Durable Benchmark

The next test is not whether the notes remain oversubscribed, but whether Sandersville converts its contracts and capital into operating capacity.

The first signal is construction progress before the fourth quarter of 2027. CleanSpark must complete network halls, install supporting infrastructure, and meet Meta’s delivery requirements.

Timely milestones would strengthen the case that experienced power-site operators can move into AI development. Delays would expose the execution gap between Bitcoin mining and hyperscale computing.

Investors should watch whether CleanSpark updates its targeted rent commencement date. Changes to estimated development cost per megawatt would also reveal pressure on the budget.

The second signal is trading in the notes after issuance. The initial 8.25% yield establishes the starting point, but secondary-market pricing will show how investors reassess risk.

A tighter yield would suggest that buyers view Meta’s lease support and project progress favorably. A wider yield would indicate growing concern about construction, supply, or the wider AI debt market.

Future speculative-grade data center offerings will provide another comparison. Strong order books at lower premiums would confirm that Sandersville opened a durable financing route.

Weak demand for similar transactions would suggest this sale succeeded because its yield was unusually generous. That outcome would make CleanSpark’s deal an exception rather than a template.

The third signal is whether the Meta relationship expands into Texas. CleanSpark has described exclusivity across a portfolio with up to 885 megawatts of secured and planned capacity.

A signed Texas lease would show that Sandersville is the beginning of a repeatable partnership. It would also create another large financing need and test CleanSpark’s capacity to manage parallel developments.

If no additional agreement follows, investors should treat the exclusivity as preliminary business development rather than contracted growth. The company’s value would remain more dependent on one Georgia project.

Meta’s own disclosures will provide further context for all three signals. Rising lease commitments would confirm that it continues shifting infrastructure funding toward third-party owners.

Slower commitments could indicate that Meta is moderating expansion or favoring different ownership structures. Either change would affect developers competing to serve its AI plans.

For technology buyers and enterprise users, the bond sale offers a useful reality check. AI services increasingly depend on long-lived physical infrastructure financed years before customers see the resulting capacity.

For investors, the lesson is more specific. Tenant quality cannot replace analysis of construction, collateral, completion support, and lease terms.

The Meta data center junk bond attracted extraordinary demand because it combined a creditworthy lease guarantor with an extraordinary yield. Both sides of that equation matter.

CleanSpark now has the capital to pursue its Sandersville plan, and Meta has secured another block of future computing capacity. The remaining question is whether construction performance validates the confidence embedded in that $10 billion order book.

Watch the late-2027 delivery schedule, the notes’ secondary-market yield, and any binding Texas lease. Together, those signals will show whether this transaction became a financing model or only a well-priced debut.