NAVER Corporation (KRX:035420) first disclosed a partnership with Nvidia on June 8. On July 27 it amended that filing, and the stated reason for the amendment is worth quoting: to reflect Nvidia's equity investment, and to correct the cooperation structure and the expected investment amount.
All three changes matter, and the equity one is the headline. Nvidia will participate in a third-party allotment of newly issued NAVER common shares worth about $1bn.
At the ₩1,402.50 exchange rate a Korean filing used on August 20, that is roughly ₩1.40tn. Against NAVER's market capitalisation of ₩32.92tn at the August 27 close of ₩216,500, it would give Nvidia something in the region of 4% of the company.
Read the two versions side by side and the shift in Nvidia's position is unmistakable.
The June text described the arrangement this way: NAVER leads data centre site acquisition, construction and operation, and Nvidia participates as the supplier of GPUs.
The July text reads: NAVER leads data centre site acquisition, construction and operation, and Nvidia participates as a business partner that supplies GPUs and also jointly develops global customers and shares revenue and business risk.
A chip vendor that sells you hardware has been paid regardless of whether you fill the racks. A partner that shares revenue and business risk has not. Nvidia moved from the safe side of that line to the exposed one in seven weeks, and it put $1bn of its own money into the equity at the same time.
For anyone trying to judge how seriously to take this, that revision is the most informative sentence in either document.
The filing gives a schedule, which is more than most AI infrastructure announcements manage.
Fifty-five megawatts in the first half of 2027. One hundred megawatts cumulative by the end of 2027. Two hundred megawatts cumulative in 2028. And then, in the filing's words, a plan for gigawatt-scale infrastructure, with no date attached.
The first site is Gak Sejong, NAVER's hyperscale data centre, and the stated expansion path runs from Asia-Pacific to Europe and the Middle East in stages. NAVER also says it will adopt Nvidia's next-generation infrastructure platform, DSX, to run the facilities.
The filing is careful to flag that the timetable is provisional and still under negotiation. Take it as an ambition with numbers attached rather than a commitment. But 200MW by 2028 is a real and checkable claim, and it is small enough to be credible. A single large US hyperscaler campus can exceed that on its own.
The word to notice in the expected effects section is "sovereign." NAVER says the goal includes responding to sovereign AI demand, which is Nvidia's own term for governments that want national AI compute inside their own borders, on their own language models, under their own jurisdiction. That is a market where being Korean, and being a company that already runs the search engine and the language model for a country of 50 million people, is an advantage rather than a limitation.
The financial architecture is the part most readers will skip and shouldn't.
The filing states that the computing infrastructure required for the AI factory, at a scale of $9bn, has been proposed to be arranged by Brookfield Asset Management as project financier, and that NAVER is negotiating a contract under which it would use that infrastructure. NAVER plans to appoint Brookfield as exclusive preferred negotiating partner for twelve weeks.
That structure matters enormously. Nine billion dollars is roughly ₩12.6tn. NAVER spent ₩1.23tn on property, plant and equipment in all of FY2025 and ₩809.8bn in the first half of 2026. It held ₩6.38tn of cash at June 30 against total equity of ₩31.03tn. It cannot fund ₩12.6tn of GPUs from its own balance sheet without transforming what kind of company it is.
Under the proposed arrangement it does not have to. An infrastructure fund owns the assets and takes the residual value risk on hardware that depreciates fast; NAVER contracts for the use of it and pays as it goes. The capital intensity sits with Brookfield, and NAVER's exposure becomes a contractual commitment rather than a balance sheet full of chips.
The trade is that NAVER captures less of the upside if the AI factory works, and pays a rental spread that a self-funded operator would keep. It is the same calculation every airline makes about leasing versus owning aircraft, and for a company with NAVER's cash generation it is defensible.
Note the word "proposed." Nothing is signed. Twelve weeks of exclusivity from late July puts the negotiation into October.
From Nvidia's side the logic is straightforward and worth stating plainly, because it also identifies the risk.
Nvidia's constraint is not demand for chips, it is the availability of places to put them: sites with power, cooling, permits and an operator. Taking equity in a national champion that already owns hyperscale data centre capacity in a country with a sovereign AI agenda secures a destination for GPUs and a partner who can sell compute to customers Nvidia does not reach directly.
Which means Nvidia's $1bn is partly a customer acquisition cost. That is not a criticism. It does mean an investor should not read the equity stake as a pure endorsement of NAVER's valuation. Nvidia has made a number of investments of this shape, and the strategic motive and the financial motive are not the same thing.
To its credit, NAVER lists them, and they are the right ones.
Uncertainty in securing data centre sites and regulatory approvals in each country. Delays in securing power infrastructure. Volatility in global AI infrastructure supply and demand. And the possibility that terms change or timelines slip during detailed contract negotiation.
Power is the one that would worry me most. Every megawatt in that ladder needs a grid connection, and grid connection queues are the binding constraint on data centre construction in most of the world right now. Korea's own power situation is tight, and Sejong is not near the coast where most new generation lands.
The commercial risk deserves equal weight. NAVER is proposing to build compute capacity and sell it, which is a business it has never run at scale, against hyperscalers with far deeper pockets and against neoclouds that already do exactly this. Owning the racks is not the same as filling them.
There is also a straightforward financial observation. NAVER's revenue grew 16.3% in the first half of 2026 while operating income grew 3.4%, and the operating margin fell from 18.01% to 16.02%. Capital spending doubled in FY2025 and doubled again in the first half. The AI investment is already compressing margins before a single megawatt of the Nvidia programme is live.
December 31, 2026. NAVER committed in the filing to disclose the status of this project, including the conclusion of major contracts by stage, by that date, and to disclose immediately if anything material is signed sooner.
Three specific things should appear by then. Whether the Nvidia share subscription actually closed and at what price, since the terms sit in a separate capital increase filing and a third-party allotment price is set by formula against a reference period. Whether the Brookfield exclusivity converted into a signed agreement or lapsed. And whether NAVER's own direct investment figure, which the filing still records as undetermined, has finally been quantified.
Until those three exist, this is a well-specified plan between three credible parties, with a schedule, a financing structure and a named list of things that could break it. That is considerably more than most AI infrastructure announcements offer. It is still not a contract.
kstock reads DART every morning and writes up what moved — the contract, the buyback, the number that does not add up. The daily post and a Saturday roundup, by email.