Reported 12 Nov 2026 · Compute A one-minute read

Naver sold 2 trillion won of AI compute to one American customer.

On November 12 Naver's chief executive said it signed compute supply contracts worth about 2 trillion won with a United States AI inference company. Caveat: the customer is not named.

Until now

Korea buys American compute to run its own AI.

Now

It also sells compute back, on contracts it signs first.

Why it matters

The usual direction is Korean money going to American data centres. Here an American inference company has booked 20 megawatts in Korea for four to five years, and Brookfield put up the equity so Naver carries little of the cost. Compute capacity is becoming an export.

Established

Naver says it signed about 2 trillion won of compute supply contracts.
The volume is given as 20 megawatts over 4 to 5 years.
Revenue starts when service launches in the first half of 2027.

Not established

Who the customer is. Naver did not name it.
That the chips are secured. Naver is still negotiating GPU orders.
The margin. No price or cost per megawatt was disclosed.

The five-minute readThe record, the timeline, the weight and 5 sources

The record

Reported, not verified. The contract value, the 20 megawatts and the 4 to 5 year term come from a transcript of Naver's third-quarter call and from Korean trade coverage, not from a document we read. Naver's investor relations page lists a 3Q 2026 release, but its file and webcast were not readable to us, and no exchange disclosure names the contract. Verification date: December 31, 2026.

Naver chief executive Choi Soo-yeon told the third-quarter 2026 earnings call on November 12, 2026 that "we signed a compute services supply agreement worth approximately KRW 2 trillion with a U.S.-based AI inference company," and called it "a meaningful contract that validates real market demand for the AI factory business." The transcript gives the volume as about 20 megawatts over 4 to 5 years, on a mix of Nvidia B300 and Vera Rubin chips, with revenue recognised once service launches in the first half of 2027. The customer is not named. Finance chief Kim Hee-cheol said the asset company's capital "is funded entirely by Brookfield's equity contribution rather than borrowings," and that "by securing demand first and then expanding infrastructure, NAVER will preserve its financial soundness." Seoul Economic Daily, publishing at 09:22 Korea time on November 12, reported that Brookfield, through Brookfield Computing, and Nvidia both completed third-party share allotments to Naver, which Choi said minimised its initial capital burden; that the contracts secured exceed the volume Naver first targeted, in her words "the contracts secured so far are above our target volume"; that Naver plans long-term contracts with several large clients to avoid dependence on one; that sites with the power and liquid cooling Vera Rubin needs are scarce and Naver has secured them; and that Naver is pursuing what it calls Asia's largest GPU order. The quarter: revenue of 3.4724 trillion won, up 10.7% on a year earlier; operating profit of 451.3 billion won, down 20.9%; margin of 13%; net income of 402.5 billion won, down 45.2%; and free cash flow down 353.4 billion won, on higher capital spending and about 300 billion won of income tax. Chairman Lee Hae-jin announced the Nvidia and Brookfield AI factory partnership on July 24, 2026. Not disclosed: the customer, the price per megawatt, termination terms, allotment sizes, total project cost, or GPU count.

How it got here

Jul 24
Naver's chairman Lee Hae-jin announces the AI factory partnership with Nvidia and Brookfield. At that point there is no customer and no contract.
Oct 29
Korea's small business ministry commits more than 5 trillion won to AI and robots in small factories. The state is paying to put compute to work. Edition 065.
Nov 9
Nvidia agrees to let a Korean chip from Rebellions do half the work inside its own racks, with SK Telecom. Korea moves up the compute stack. Edition 082.
Nov 11
CoreWeave says it expects about $9 billion of customer prepayments and deposits over 3 months. Buyers of compute are now paying before delivery.
Nov 12
Naver says it signed about 2 trillion won of compute supply contracts with an American inference company, about 20 megawatts over 4 to 5 years, revenue from the first half of 2027.

Who has to react

The unnamed American customer, whose capacity plan now runs through Seoul; CoreWeave, Nebius and the other rented-compute firms; Brookfield, which owns the equity in the asset company; Nvidia, both supplier and shareholder; Korea's power utility, which must deliver the load; Samsung and SK hynix; and Naver's shareholders, told profit fell as the build continues.

What to watch

Would strengthen itA second Asian operator discloses a compute supply contract with an American AI company worth more than $500 million, or Naver names its customer, by September 30, 2027.
Would weaken itNaver's AI factory revenue does not begin in the first half of 2027, or the contract is reduced or cancelled, by December 31, 2027.
Would strengthen itBrookfield or Nvidia repeats the asset company and share allotment structure with another operator outside Korea, by December 31, 2027.

Why it weighs this much

Signal strength

CostMedium

CommitmentHigh

ReversibilityHard

PersistenceYears

Cost is medium rather than high, and that is the point of the structure: Brookfield's equity funds the asset company, Nvidia took shares, and Naver says it carries little of the initial burden. Commitment is high. A signed multi-year supply contract disclosed on an earnings call by a listed company's chief executive is a statement the market will price and the auditors will test. Reversibility is hard: the counterparty is buying 20 megawatts of liquid-cooled capacity that barely exists elsewhere in the region, with chips ordered against it. Persistence is years: the term is 4 to 5, and revenue does not start until 2027.

The words

AI factory: Nvidia's term for a data centre built only to train and run AI models. Naver is building one and selling its capacity by contract.
Inference company: A firm that runs other people's models for them rather than training its own. Its main cost is rented compute.
Third-party share allotment: New shares sold to a chosen investor rather than the market. Brookfield and Nvidia both took shares, so Naver built without borrowing.
Megawatt: The unit compute is now sold in, because power, not floor space, is the limit. 20 megawatts is mid-sized, not hyperscale.

Sources

Your call

Which explanation do you believe?

A call is final. Other people's calls show after yours.