Volta’s $10 billion compute deal: the rent and duration behind 121MW

A new compute company can lease a site, raise capital, secure a customer and buy GPUs. The difficult part is aligning the timing and obligations of those contracts. Volta’s first Norwegian project combines a long-term Bitdeer colocation lease, a company-announced $10 billion AI-lab partnership and planned NVIDIA Vera Rubin systems.^[1][2]

Media identify the customer as Anthropic and describe a six-year term. Volta’s release does not name the laboratory or specify that duration. The reported term can support a clearly conditional calculation, but it cannot be subtracted from a sixteen-year property contract to manufacture a profit margin.

Follow the payment chain

Bitdeer’s Tydal Data Center AS signed the lease and services agreement with Volta Tydal AS. Bitdeer affiliates retain the campus, while Volta contracts for 121 IT MW supported by an estimated 133 gross MW. The AI laboratory receives computing services through Volta; Dell is named as technology provider.^[1]

Bitdeer receives colocation and service payments. Volta funds those obligations and organizes the compute platform. The laboratory pays for computing. These revenue streams belong to successive layers of one transaction and cannot be added as independent end-market demand.

Volta disclosed a $5 billion infrastructure financing program with Azora and completed seed and Series A rounds at a $2.4 billion valuation, with investors including NVIDIA and Michael Dell’s family office. The approximately $300 million raised is media-reported. Financing capacity, equity valuation and the $10 billion customer partnership measure different things.^[2][4]

Institutional funding can support durable assets and strategic investors can help execution. Their participation is not an unconditional delivery guarantee. If equipment payments precede customer receipts, Volta still needs construction and commissioning liquidity.

Gross power is not IT capacity

The 121 IT MW is contractual critical load; 133 gross MW includes facility overhead. Dividing 133 by 121 gives approximately 1.10, consistent with the company’s approximate design PUE of 1.1. Design efficiency is not an observed annual operating result.^[1]

The project has two equal phases across four halls, with target commencement on December 31, 2026 and March 31, 2027. A schedule is disclosed; the question is whether it is achieved. A separate 47 gross MW expansion for other AI/HPC uses in the second half of 2027 should not automatically be assigned to the same customer.

Equal phasing implies approximately 60.5 IT MW per phase. Even timely first-phase commencement does not justify applying the full 121MW to an entire year. Equipment arrival, power availability, lease commencement and customer acceptance can occur on different dates.

Norwegian hydropower, low temperatures and the company-described dual grid connection can support operating economics. Actual advantage still depends on energy contracts, network charges, power constraints and system efficiency. Geography does not replace a procurement agreement or remove equipment-delivery risk.

The lease contains more than its headline value

Bitdeer describes approximately $4.7 billion of contracted payments over an initial sixteen-year term, rising to a potential $8 billion with an eight-year extension. The tenant has a no-fee termination right at year ten, while rent and services escalate 3% annually. Full contractual value assumes performance; extended value also assumes renewal. Neither is current GAAP revenue.^[1]

The stated sixteen-year average price is about $202 per kW per month. Multiplying 121,000kW by $202 and twelve months gives approximately $293.3 million annually, close to $293.75 million from dividing $4.7 billion by sixteen. Rounding explains the small difference. These are lifetime averages, not first-year rent, given escalation and phased commencement.

Electricity is passed through to the tenant. Volta therefore owes power costs in addition to colocation fees. The mechanism reduces direct energy-price exposure for Bitdeer’s property income, but does not eliminate collection, metering, settlement or interruption risks.

Management estimates approximately 90% NOI margin. Applying it to average annual revenue gives roughly $264 million of illustrative property net operating income. NOI excludes items including corporate overhead, depreciation, amortization and share-based compensation. It is not net profit and does not pay financing or reinvestment automatically.

Remaining capital expenditure is approximately $500 million, or about $4 million per IT MW, and Bitdeer plans additional borrowing. Dividing that illustrative NOI by remaining capex produces roughly 53%, but is not a valid project return: it compares income from the whole asset with only the unfinished investment. Historical land and facilities expenditure, maintenance, interest and taxes remain relevant.

Six years of customer revenue and sixteen years of rent

Assuming the media-reported six-year term and full realization of $10 billion, average annual customer value is approximately $1.667 billion. If the entire arrangement covers the same 121 IT MW, this equals roughly $13.77 million per IT MW annually. Using 133 gross MW instead produces $12.53 million. These denominators are different, rather than evidence of a price change.

Average annual colocation payments of approximately $294 million would represent about 17.6% of that customer value. The remaining 82.4% is not profit. GPUs, servers, networking, energy, staff, software, insurance, financing and replacement still need funding. Different commencement dates and escalators make the ratio a scale comparison only.

Duration is the more consequential mismatch. If the customer leaves after six years but the property lease is freely terminable only after ten, Volta may need replacement workloads in between. Continuing for sixteen years increases the importance of equipment refresh and future customers. Renewal, termination, equipment ownership and transfer provisions are not fully public.

Six years may itself span several compute generations. A performance-based commitment could require operator-funded upgrades; a contract for specified hardware may leave the customer assessing the later-life cost of older systems. Neither arrangement is assumed here, but each would produce different economics from the same headline value.

An unnamed customer limits public credit and concentration analysis. It does not establish that a contract is non-binding. Refusal to comment is also not evidence of deterioration. Payments, amendments and delivery performance are the meaningful tests.

Credit support cannot finish construction

Bitdeer anticipates approximately $1.3 billion in letters of credit arranged by J.P. Morgan affiliates and another major institution, subject to customary conditions. It can terminate if Volta misses specified credit-support milestones. Anticipated support is not proof that every instrument is already effective.^[1]

Against roughly $294 million of average annual colocation payments, $1.3 billion is about 4.4 years of rent in scale. It does not guarantee unconditional coverage for that duration. Drawing conditions, expiry, reductions, beneficiary scope and dispute treatment remain decisive.

Credit support also does not deliver missing hardware or make a system pass acceptance testing. Construction overruns and commissioning still need an allocated payer. Financing and physical completion must be assessed together.

As a separate illustration, $1 billion of equipment funding at an assumed annual cost of 8% incurs roughly $40 million over six months. Actual equipment spending and financing terms are unavailable. Rent or minimum energy payments could add to the waiting cost. Staged procurement and alignment of rent with acceptance can reduce it.

One campus contains two contracts with different margins

The Tydal hosting lease and Volta’s computing agreement occupy different levels of the same operating chain. Bitdeer receives facility payments; Volta must also fund computing equipment, software and customer delivery. The laboratory’s payments must cover those costs and financing before becoming Volta profit. Adding $4.7 billion to $10 billion would double-count part of the same economic chain as independent final demand.

A reported six-year customer term and a sixteen-year landlord lease are therefore not conflicting descriptions. They describe different commitments, with the no-fee termination right after year ten adding another boundary. Volta needs both initial equipment recovery and a plan for customers and refresh beyond the first service contract.

Mismatch is not inherently unacceptable if initial contractual cash recovers equipment investment quickly. It becomes more consequential when repayment relies on renewal at favorable prices. Equipment cost, amortization, customer minimum payments and service deductions are needed to evaluate that distinction; they remain unavailable.

Software and operational capability also matter. Volta’s announced acquisition of Genesis Cloud capability forms part of its service strategy but does not prove future utilization or margin. A customer buys usable computing, requiring electricity, hardware and scheduling to work together. A delay in any layer can allow fixed facility payments to begin before service receipts.

Bitdeer investors should therefore examine construction, lease credit and retained capital, while Volta investors need equipment recovery and service contribution. The same MW does not give both businesses the same return, and a prominent customer cannot erase their different costs and responsibilities.

Two different investment propositions

For Bitdeer, the opportunity is to develop a long-duration colocation asset. Expected NOI, escalation and energy pass-through are attractive features, but historical capital, the year-ten break, completion cost and debt terms determine equity returns. Borrowing against the project to accelerate other developments also increases overall financial exposure.

For Volta, the proposition is lower-cost funding combined with dependable compute delivery, leaving a sufficient spread to cover hardware life cycles. Its company-described pipeline above 1GW and multi-gigawatt ambition for 2030 are not orders with identical credit quality. Each location needs its own alignment of customers, funding and power.^[2]

The favorable case is timely phased delivery, effective credit support, customer payment and equipment and energy costs consistent with the expected spread. The counter-case combines shorter customer commitments with longer asset obligations and additional funding needs when hardware or customers change. Platform scale can rise faster than cash returned to equity.

The first observations are phased commencement in late 2026 and early 2027, subsequent Bitdeer revenue and capital expenditure, and operating evidence from the compute platform. Confirmation of Anthropic would improve credit analysis. Delivery and cash collection are what convert partnership value into an operating business.