Summary

  • Nscale announced $3.36bn of pre-IPO convertible loan-note financing on 25 September. The amount consists of a $2.36bn initial tranche at closing and a further $1bn NVIDIA commitment expected to fund in mid-November; the full headline amount was not funded at the first closing.
  • The notes have a defined legal conversion point, but Nscale’s $103.4bn of active-and-contracted total contracted value does not. Only $2.6bn was active at 31 August. The remainder still has to pass financing, power, construction, equipment, acceptance and service-availability gates before it can become operating revenue.

The word “convert” is doing two different jobs in Nscale’s latest financing story. The first is legal and compact. Investors supply debt-like capital; if Nscale completes its initial public offering, the unsecured notes automatically become ordinary shares, or non-voting shares in NVIDIA’s case. The second is physical and prolonged. A signed commitment for AI capacity becomes revenue only after land, electricity, buildings, cooling, networks, GPUs, software and operators produce an accepted service and keep it available.

One conversion can occur at a corporate event. The other happens rack by rack and month by month.

That distinction is easy to lose because the two headline amounts are designed to meet each other in the same sentence. Nscale’s 25 September announcement says it raised $3.36bn through convertible loan notes and has more than $103bn in total contracted value, or TCV. The first figure is financing. The second is a measure of contracted consideration across service terms. Neither is current revenue, and they do not share a recognition rule.

The first closing is $2.36bn, not the whole headline

The financing has a sequence. Nscale says an initial $2.36bn tranche closed. A further $1bn commitment from NVIDIA is expected to fund in mid-November 2026. Its release describes the notes as automatically convertible when the IPO completes and identifies Goldman Sachs as placement agent.

The sequence matters. “Raises $3.36bn” is the transaction label; $2.36bn is the amount the company says arrived at the first closing. The remaining $1bn is a commitment with a later expected funding date. Treating all $3.36bn as same-day cash would erase the only condition the announcement makes explicit.

The Form S-1 filed a week earlier shows the financing while it was still being assembled. It describes a 15 September subscription agreement for a minimum of $3.1bn: $2.1bn of unsecured convertible loan notes plus a $1bn NVIDIA sale. The later announcement increases the initial component by $260m. The dated records support the conclusion that the disclosed size grew between filing and announcement; they do not, by themselves, disclose every amendment or allocation behind that increase.

NVIDIA’s legal route also depends on timing. The S-1 says that if its sale closes before the registration statement becomes effective, NVIDIA receives additional unsecured convertible notes that become non-voting shares at the offering. If closing comes on or after effectiveness, the purchase is instead satisfied through non-voting shares using the assumed IPO price mechanism described in the prospectus. The company expected that closing around 16 November.

These details keep three events separate: subscription, funding and conversion. They may occur on different dates. A commitment is not cash; cash is not yet equity; conversion is not the same as a completed IPO.

The IPO is a gate, not a timetable

The preliminary S-1 leaves the number of shares and the price range blank. Nscale says it intends to apply for a New York Stock Exchange listing under NSCL, but an intention and a filed registration statement do not provide a completion date or valuation. The notes mature on 15 June 2028 unless earlier converted or repurchased. That long-stop matters precisely because an IPO is not automatic simply because the notes convert automatically if it happens.

For existing owners, conversion creates dilution whose final size depends on the offering price and the note terms. For note investors, the instrument bridges a private-company credit exposure into prospective public equity. For Nscale, it supplies capital before the public market has set a price. The parties have made the form change mechanically legible while leaving market timing and valuation to the offering process.

This is a rational bridge for a company whose physical build runs ahead of revenue. It also moves the financing clock forward without moving every delivery clock at the same speed. The notes can finance an organization and its contribution to projects; they cannot make utilities energize a site, suppliers deliver systems or customers accept capacity by legal conversion.

$103.4bn is a term value, not a cash balance

Nscale’s definition of TCV deserves more attention than the size of the number. The S-1 defines it as aggregate revenue contracted across the full committed term of signed customer agreements at the time of signing. It excludes optional extensions and renewals, capacity not yet exercised and significant financing components that arise from upfront payments. As of 31 August, the company reported about $2.6bn of active TCV and $103.4bn of active plus contracted TCV, with a weighted-average contract life of roughly 5.7 years.

The difference between $2.6bn and $103.4bn is the article. Only a small portion of the total was associated with active service at that date. The rest represents contracted service value that has not yet all entered operation. The 5.7 years describe the average period over which services are delivered and revenue recognized, not a promise that the amount arrives evenly or without execution conditions.

Several familiar labels would misstate the measure. It is not cash because customers may pay at different times and because part of any prepayment can have a financing component. It is not revenue because accounting recognition follows performance. It is not identical to remaining performance obligations, because Nscale’s TCV definition and accounting perimeters differ. It is not a non-binding sales forecast because the company says unexercised capacity and extensions are excluded. “Contracted” is therefore substantive, but it is not “delivered.”

The scale difference is visible in the income statement. Nscale reported $140.6m of revenue for the first half of 2026, up from $10.4m a year earlier; it reported $33.0m for all of 2025 and $19.1m for 2024. Those figures establish rapid growth from a small base. They do not supply a shortcut from the entire TCV book to an annual run rate.

Three ledgers describe three different things

At 30 June, Nscale reported $56.4bn of remaining performance obligations under the service-contract accounting model. That amount included a $7.1bn significant financing component associated with customer prepayments and was expected to be recognized over the next seven years. Separately, it disclosed about $2.0bn of minimum lease payments under lease accounting.

TCV, remaining performance obligations and minimum lease payments should not be stacked into one larger number. TCV is the company’s contractual-value measure at signing. RPO is an accounting measure for unsatisfied or partly unsatisfied performance obligations under ASC 606. Lease payments sit under ASC 842. The financing component within customer prepayments reflects the time value created when cash and service delivery are far apart; it is not an extra unit of compute to be delivered.

These ledgers are useful precisely because they disagree in controlled ways. They show that one commercial relationship can contain service consideration, financing economics and lease consideration. A reader who calls all of it backlog loses the boundaries that determine timing, margin and risk.

Contract value must pass a capital stack

The new convertible is only one layer of Nscale’s funding architecture. In July, the company announced a $900m revolving credit facility for flexible liquidity. In August, it announced about $3bn of senior secured delayed-draw facilities: up to $1.85bn for Ward County, Texas, and $1.2bn for Madison, North Carolina. In February, it signed a $1.4bn GPU-backed delayed-draw loan for European clusters.

The instruments do different work. A parent-level unsecured convertible can fund corporate needs before becoming equity. A revolver supplies flexible liquidity. Project and GPU facilities can match borrowing to equipment and sites, often subject to conditions before each draw. Customer prepayments can reduce the external capital needed for a project while creating financing expense and an obligation to perform later.

Adding every announced facility as if it were freely available cash would be as misleading as treating TCV as revenue. Drawn amounts, collateral, eligible expenditure, conditions precedent, subsidiary boundaries and customer milestones determine which dollars can solve which problem. Vertical integration does not make capital fungible; it gives Nscale more of the delivery chain to coordinate and finance.

The customer contract is not a passive purchase order

The Anthropic agreements make the delivery gate concrete. Nscale says four agreements could generate aggregate payments of up to about $44.6bn for dedicated GPU infrastructure at the Monarch campus. Service begins by tranche after acceptance. The agreements require specified delivery and service availability, and Nscale said at the S-1 date that it had no binding commitments for the financing required to fund performance under them.

That is not evidence that the contracts are weak. It is evidence that their economic value is conditional on execution. Nscale must use best efforts to obtain qualifying financing for GPUs and data-centre infrastructure within a specified period. Missed milestones or service failures can bring contractual remedies including credits, penalties, liquidated damages, refunds or termination. A take-or-pay structure can protect the provider once agreed capacity is available; it does not excuse the provider from making that capacity available.

Microsoft’s public description of the Narvik project supplies a customer-side example. Microsoft, Nscale and Aker described a five-year agreement worth an estimated $6.2bn, with service arriving in stages from 2026. Renewable power and secured grid capacity are not decorative details. They are prerequisites to the product being sold.

The conversion chain is therefore observable. Financing closes. Conditions are satisfied. Equipment is ordered and delivered. Power and cooling become available. Systems are commissioned. A customer accepts a tranche. Service starts. Availability is maintained. Only then does a portion of multi-year contract value appear as revenue, and only after operating costs does it become margin.

The two clocks allocate risk to different holders

Before an IPO, note investors carry issuer credit, duration and offering risk. At conversion, some of that exposure becomes equity price risk. Existing shareholders carry dilution. If the IPO takes longer, the maturity date and repurchase provisions become more important, while the company continues to finance an expanding physical programme.

On the delivery side, Nscale carries integration and operating risk across a wide surface. Lenders can withhold draws if conditions are not met. Utilities and power developers influence energization. GPU and network suppliers influence commissioning. Customers control acceptance and may exercise remedies. None of those parties alone controls the whole chain, which is why the company’s claim of vertical integration should be tested through dated delivery receipts rather than treated as a synonym for certainty.

The risk is asymmetric. The company can receive financing before it has activated the corresponding proportion of contract value. That is the purpose of growth capital. Yet financing creates fixed claims, dilution or both even if a site runs late. Contracted TCV may remain legally valuable while its revenue moves rightward. The spread between the two clocks is where capital cost accumulates.

An accumulated deficit of about $1.86bn at 30 June is not a forecast of failure. It is a reminder that past investment and losses already sit beneath the new round. The relevant question is not whether Nscale can announce more capital or more TCV. It is whether each new dollar shortens the path from contracted capacity to accepted, available service faster than financing cost and execution obligations compound.