Inside the Korea AI factory deal: Nvidia, Brookfield and NAV

Brookfield Asset Management is widening its Korean digital-infrastructure position beyond the Sejong data-centre project, extending a partnership with NAVER and Nvidia that has already scaled from 55 megawatts to 200 megawatts and now targets 1 gigawatt. The Korea AI factory at Sejong is funded by up to $9 billion arranged by Brookfield as exclusive capital partner, a $1 billion equity commitment from Nvidia and the balance from NAVER, against a project the three companies value at $10 billion.

The buildout arrived in two steps. A 55-megawatt deployment was announced in June 2026, then more than tripled to 200 megawatts in late July, unveiled during South Korean President Jae Myung Lee's AI Summit visit to San Francisco. The partners describe the result as multi-tenant AI cloud infrastructure aimed at enterprises and governments in Korea and the United States, which moves NAVER from consuming compute to selling it.

The Korea AI factory ladder is the real scoreboard

Gigawatt targets are cheap to announce. Energising a site is not. Sejong's schedule runs through four rungs: 55 megawatts operating in the first half of 2027, 100 megawatts by the end of that year, 200 megawatts in 2028, and a longer-term push toward 1 gigawatt with no published date. Each rung roughly doubles the load, and each doubling needs another power purchase agreement, another grid interconnection and another round of substation and cooling construction.

Grid interconnection, not accelerator supply, is emerging as the binding constraint on how fast sovereign AI sites can switch on. Nvidia has said grid queues will not deliver power to all of its national partners on the timetables those partners have published, even as it positions Australia as a 2-gigawatt sovereign hub. Korea's advantage in that contest is an industrial electricity base and a single national champion willing to anchor the equity.

Who carries which risk

The capital structure splits three ways, and the risk splits with it. Brookfield's contribution is capped at $9 billion and framed as the exclusive capital partner role, which makes the figure a ceiling and leaves the drawn amount lower. Nvidia's $1 billion is equity, turning a hardware supplier into a shareholder in demand for its own accelerators. NAVER funds the remainder and carries the commercial risk of filling 200 megawatts of capacity before the next rung is financed.

Phase	Capacity	Target	Financing
First	55 MW	First half of 2027	Announced June 2026
Second	100 MW	End of 2027	Interim step toward the 200 MW plan
Third	200 MW	2028	Up to $9B Brookfield, $1B Nvidia, balance NAVER
Long term	1 GW	No date published	Stated intention, not contracted capacity

The hardware layer is already fixed. The 200-megawatt factory is expected to run on Nvidia's Vera Rubin and Blackwell platforms, and NAVER holds Nvidia Cloud Partner status. That ties the site's bill of materials to one vendor roadmap, so a slip in a platform generation becomes a slip in Korea's sovereign capacity. The second and third rungs are committed to Nvidia silicon before the power contracts behind them are signed.

Sejong also fits Nvidia's wider sovereign-AI playbook. The vendor is pairing with national champions across Korea, Australia and other markets, supplying a reference design for AI factories alongside the accelerators that fill them. For governments that want domestic compute without building a hyperscaler, that package is faster to procure than a bespoke design, and it concentrates the technical standard in one supplier.

The financing has a circular quality that infrastructure investors will scrutinise. Nvidia is putting equity into a customer that will spend heavily on Nvidia accelerators, while Brookfield supplies the debt layer. That is vendor financing in all but name, and it holds as long as tenant demand materialises. If it does not, the equity write-down lands on the supplier and the debt sits with the asset manager.

Demand is the least visible variable in the plan. NAVER is building multi-tenant capacity for Korean enterprises and government buyers, and the announcement also names US customers, which suggests domestic consumption alone will not fill 200 megawatts. The Korea AI factory is being sold as national infrastructure, a designation governments use to justify permitting priority and grid access.

NAVER's stake in the outcome runs deeper than the arithmetic of the funding split suggests. The company is buying a position in the compute layer of its home market, and it will need tenants with workloads heavy enough to justify dedicated capacity. Korean banks, manufacturers and public-sector buyers are the obvious candidates, and each of them is also being courted by the US hyperscalers.

Policy support is the fourth input alongside silicon, capital and power. The July announcement was staged around a presidential AI Summit visit to San Francisco and framed as Korea's national AI factory buildout. None of that policy weight appears in the financing numbers, and all of it determines whether the 2028 date is reachable.

The template NAVER wants to export

Sejong is the first full test of the model, and NAVER is already shopping it abroad. Earlier this month the company pitched French counterparts on building European AI-factory capacity, using the Nvidia-Brookfield structure as its reference and proposing to pair France's low-carbon energy base with NAVER's own AI stack. The logic is portable: an operating champion supplies the software layer, an asset manager supplies the debt, and the host state supplies the power.

Portability is where the model gets tested. Korea's version works because NAVER is large enough to anchor demand and absorb the residual funding gap, and because the state treats the site as national infrastructure. France has the electricity but no direct NAVER equivalent, and European permitting timelines run longer. The June 2026 announcement also framed the original arrangement as covering more than a gigawatt of global AI-factory capacity, which makes Sejong one tranche of a wider programme in which Nvidia pairs with national champions instead of hyperscalers.

Brookfield's September move is the clearest signal about how the template scales. The firm says it is expanding Korean exposure across data centres, energy assets and the NAVER partnership together, a different posture from a single-asset infrastructure bet. Owning generation and interconnection alongside the compute hall shortens the distance between a megawatt promise and an energised rack.

Compare that with hyperscaler construction in the United States, where one company funds the shell, the power contracts and the silicon from its own balance sheet and depreciation schedule. Sejong spreads those functions across a national champion, an infrastructure lender and a chip vendor. Spreading lowers the entry cost for the state and introduces three separate approval processes where a hyperscaler has one.

What to watch

Three markers will show whether the ladder holds. The first is whether 55 megawatts actually energises in the first half of 2027, because a delay there pushes every later rung. The second is the size and structure of Brookfield's new Korean capital, since a drawn $9 billion and a ceiling of $9 billion are different propositions for the 1-gigawatt plan. The third is France, where a signed agreement would confirm the structure travels outside Korea.

The more useful comparison is Korea against other mid-sized economies trying to buy sovereign compute without a hyperscaler balance sheet. Australia carries 2 gigawatts of ambition and a strained grid. Europe has cheap power and fragmented demand. The Korea AI factory at Sejong is the first site where silicon, capital and electricity are contracted together against a published schedule.

Why this matters

The Sejong deal recasts sovereign AI capacity as a finance and power problem. Governments that can assemble an operating champion, an infrastructure lender and a power surplus will move faster than those waiting on hyperscaler investment. For the three partners the payoff is contracted demand; for everyone else the number to track is the megawatt ladder, not the headline gigawatt.

Sources

NAVER, NVIDIA and Brookfield to Expand Korea’s National AI Factory Infrastructure Buildout | NVIDIA Newsroom

✔Human Verified

Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.