Naver, Nvidia and Brookfield proposed expanding an AI factory at Naver's GAK Sejong data center from an announced 55 megawatts to 200 megawatts by 2028. Nvidia plans to invest $1 billion in Naver shares, but only if Naver finalizes at least $9 billion of committed financing separate from Nvidia's investment. Brookfield's term sheet to fund up to $9 billion is nonbinding. [1]
Those conditions distinguish the proposal from the paper's July 25 account of Nvidia and SK's $500 billion-plus letters of intent. That initiative sits inside South Korea's inherited $950 billion umbrella: $750 billion of SK scope, including the Nvidia/SK initiative, plus roughly $200 billion of Samsung/Broadcom scope. The Naver project is separate and must not be added to that hierarchy.
Reuters and its authorized X post emphasize the $1 billion Nvidia investment. [2] The primary release makes the more important sequencing clear. Naver must first secure at least $9 billion in committed project financing, Nvidia's investment must satisfy customary closing conditions, and Naver must fund remaining amounts. [1]
Capacity belongs to different tenses
The 55 megawatts describes the initial buildout announced in June. The 200-megawatt figure is the proposed 2028 expansion. One gigawatt is Naver's longer-term intention. [1] None is interchangeable with commissioned capacity serving paying customers today.
The release names the proposed site and technology. The expansion would use Nvidia's DSX platform at GAK Sejong and is expected to include Blackwell and Vera Rubin infrastructure. [1] That specificity makes the project testable. It does not mean equipment has been ordered, installed or accepted.
The financing stack is equally specific about incompletion. Brookfield has entered a nonbinding term sheet, not a final funding agreement. Up to $9 billion is a ceiling, not cash committed. Nvidia's $1 billion is planned and conditional, not a completed share purchase. Naver's remaining contribution is not quantified in the release. [1]
Reuters supplies the market-facing headline that Nvidia will acquire new Naver shares. [2] The authorized Reuters post is useful discourse evidence because it shows which part of the arrangement travels fastest: a famous chipmaker and a round billion-dollar number. It is not independent confirmation that the condition has been met.
A separate Korean project
Keeping the Naver proposal separate from the SK and Samsung scopes is more than bookkeeping. National AI announcements often overlap companies, factories, power and time horizons. Adding every headline number can count one planned facility or financing source more than once.
The Naver release describes a multi-tenant project intended to serve Korean and U.S. AI developers. Nvidia and Naver also describe work on HyperCLOVA X models, an agent platform and a Seoul world model. [1] Those product ambitions may create demand for the factory. They do not establish contracted tenants, utilization, revenue or return.
The next receipts should arrive in order. Naver needs binding financing with named lenders and terms. Nvidia needs to close its investment. The project needs permits, power arrangements, equipment orders and construction milestones. Commissioning must precede customer use, and use must precede financial return.
Power is a particularly important missing receipt. Megawatts in a data-center announcement describe electrical scale, but the release does not publish the contract that would supply the expanded load, the generation mix, grid upgrades or connection schedule. A planned 200-megawatt facility can be commercially financed and still wait on electricity.
The customer side has a similar denominator. Multi-tenant describes the intended business model. It does not say how many tenants have signed, what capacity they reserved or what prices they will pay. Production-scale compute becomes an operating business only when installed systems are available and customers use them.
The proposal's technology names do not close those gaps. Blackwell, Vera Rubin and DSX identify the intended stack. Product availability, system configuration, delivery timing and performance at the Sejong site remain future records. [1]
This sequence is not hostility to investment. It is how a large proposal becomes infrastructure rather than an indefinitely recycled announcement. The companies have identified a site, partners, target capacity and a financing condition. That is meaningful progress. The condition also tells readers exactly what has not happened.
The $1 billion headline makes Nvidia look like the project's anchor. The release makes the missing $9 billion the gate. Until that financing becomes binding, the 200-megawatt expansion remains a named design with sophisticated partners and an unbuilt balance sheet.
-- DAVID CHEN, Beijing