NAVER’s $10bn AI factory deal brings NVIDIA capital, but the market is already questioning the payoff

NAVER Corporation (KRX: 035420) has expanded its planned sovereign artificial intelligence factory at GAK Sejong from 55 megawatts to 200 megawatts, supported by a financing structure involving NVIDIA Corporation and Brookfield. The project could involve up to $10 billion of capital, with Brookfield agreeing to provide as much as $9 billion as the exclusive capital partner and NVIDIA committing $1 billion through an investment in newly issued NAVER shares. The expanded facility is expected to use NVIDIA Blackwell and Vera Rubin platforms and reach its planned 200-megawatt scale by 2028, giving Korean and United States-based customers access to production-level AI computing infrastructure. The deal potentially transforms NAVER from a major search, commerce and internet platform into a more consequential AI cloud infrastructure operator. The central tension is whether NAVER can fill that capacity with paying customers quickly enough to generate attractive returns while managing dilution, construction risk, energy demand and pressure on operating margins.

How is the $10 billion NAVER, NVIDIA and Brookfield AI factory financing structured?

The project’s headline value should not be interpreted as $10 billion of cash already received by NAVER. Brookfield has agreed to fund up to $9 billion as the exclusive capital partner, NVIDIA will invest $1 billion and NAVER will contribute any remaining amount required to finance the project. The final capital contributed by each party could therefore depend on project development, contractual conditions and how much of Brookfield’s maximum commitment is ultimately deployed.

NVIDIA’s contribution has a direct equity component. NAVER disclosed that NVIDIA would acquire 7,241,564 newly issued shares at KRW 204,500 each, raising approximately KRW 1.48 trillion, or around $1 billion. The subscription price represented a discount of about 1% to NAVER’s July 24 closing price, and payment is scheduled for October 30, 2026. Once completed, the placement would give NVIDIA approximately 4.5% of NAVER and make it one of the Korean company’s largest shareholders.

This is more strategically significant than a conventional customer-supplier agreement. NVIDIA is not merely selling processors to NAVER. It is becoming a substantial shareholder whose economic interests will be connected to the performance of the Korean company’s wider AI strategy.

The structure gives NAVER access to capital, computing technology and infrastructure-financing expertise without requiring the company to fund the entire buildout through its existing balance sheet. Brookfield can contribute experience in data centres, power infrastructure and large-scale project financing, while NVIDIA supplies the computing architecture around which the AI factory will be built.

However, strategic alignment does not eliminate financial risk. New shares dilute existing ownership, while the project must still generate enough revenue and cash flow to justify the capital deployed. Brookfield and NVIDIA may reduce NAVER’s direct funding burden, but they do not guarantee customer demand, high utilisation or commercially attractive pricing.

Why does expanding GAK Sejong from 55 megawatts to 200 megawatts change NAVER’s business model?

The planned expansion would more than triple the capacity announced in June 2026. NAVER said the 200-megawatt facility could contain approximately 100,000 NVIDIA graphics processing units and would provide a dedicated resource pool for companies developing AI models, autonomous agents and physical AI applications. The project is scheduled to reach the enlarged capacity by 2028.

For NAVER, the opportunity is to build a new infrastructure business alongside its existing search, advertising, commerce, payments, content and cloud operations. Instead of using AI computing capacity exclusively for internal products such as HyperCLOVA X, the company plans to make production-scale infrastructure available to external businesses, government organisations and AI developers.

That creates several potential revenue streams. Customers could pay for computing capacity, cloud services, software tools, model development support, storage and managed infrastructure. NAVER could also sell industry-specific AI systems that combine its cloud platform, language models, search technology and Korean market knowledge.

A multi-tenant facility can produce better economics than infrastructure reserved for a single internal workload, provided NAVER can maintain high utilisation. Expensive processors, cooling systems and power infrastructure create costs whether customers are using them or not. The commercial objective must therefore be to secure enough long-term contracts to keep the facility productive across different workloads and customer groups.

NAVER’s established relationships with Korean businesses and public institutions may provide an initial customer base. Its experience operating domestic internet services could also appeal to organisations that require Korean-language capabilities, local technical support and control over where data is stored and processed.

The harder challenge is competing internationally. United States customers already have access to hyperscale infrastructure from Amazon Web Services, Microsoft Azure, Google Cloud and other specialised AI cloud providers. NAVER will need to compete through price, performance, regional access, customised models or services that those larger platforms cannot provide as effectively.

Why has sovereign AI become a commercially important market rather than a political slogan?

Sovereign AI refers broadly to the ability of a country or organisation to develop, operate and govern artificial intelligence using infrastructure, data and models that remain under an acceptable level of national or institutional control.

The demand is being driven by concerns about data residency, national security, language, cultural representation and dependence on foreign technology platforms. Governments and regulated industries may not want sensitive information processed through infrastructure over which they have limited operational or legal control.

NAVER is well positioned within South Korea because it already operates domestic search, cloud, payments, mapping and communication services. Its HyperCLOVA X models have also been developed around Korean-language and regional data, giving the company a different position from providers whose models were trained primarily on English-language material.

The GAK Sejong expansion gives NAVER the infrastructure layer needed to support that proposition. Sovereign AI requires more than a domestically branded model. It requires access to processors, networking, storage, power, security, data-management systems and engineers capable of operating the entire computing environment.

NAVER’s partnership with NVIDIA creates an interesting compromise. The Korean company can operate the infrastructure and develop local models, but the underlying accelerated-computing stack remains heavily dependent on a United States semiconductor supplier.

This means the project increases operational sovereignty without creating complete technological independence. NAVER would still be exposed to NVIDIA’s product roadmap, pricing, supply availability and applicable United States export controls.

That dependency does not invalidate the model. Few countries or companies can reproduce NVIDIA’s complete hardware and software ecosystem economically. The commercial test is whether NAVER can provide customers with enough control over data, model customisation and deployment to satisfy sovereign requirements despite relying on imported processors.

How will NVIDIA Blackwell, Vera Rubin and DSX infrastructure support NAVER’s AI ambitions?

NAVER plans to build the facility using the NVIDIA DSX platform, with computing systems based on Blackwell and the newer Vera Rubin architecture. DSX is intended to provide an integrated AI factory design incorporating computing, networking, storage and software rather than requiring customers to assemble every component independently.

The use of different NVIDIA generations may allow NAVER to bring capacity online in stages. Blackwell systems can support near-term model training and inference, while Vera Rubin is expected to address future workloads requiring greater performance, memory bandwidth and energy efficiency.

The infrastructure will support more than NAVER’s internal models. The companies intend to create a pool of computing capacity for emerging AI businesses in South Korea and the United States. This could make NAVER an alternative infrastructure partner for startups that lack the capital to purchase large processor clusters independently.

The strategic value of the NVIDIA relationship also extends into software and model development. NAVER is adapting HyperCLOVA X using NVIDIA Nemotron 3 Ultra open models alongside its proprietary data and training capabilities. It has joined the NVIDIA Nemotron Coalition, through which companies collaborate on model pretraining, post-training and reinforcement learning.

NAVER plans to launch an AI agent platform in South Korea during the second half of 2026 using NVIDIA Agent Toolkit software, including NemoClaw blueprints. It is also developing a Seoul World Model using NAVER’s urban street-view and spatial-modelling data together with NVIDIA Cosmos world foundation models.

These projects provide potential internal demand for the AI factory. HyperCLOVA X, agent platforms, robotics, digital twins and world models require substantial computing resources. The infrastructure could therefore support NAVER’s product development even before external utilisation reaches scale.

However, internal demand should not be confused with external revenue. Using the facility for NAVER’s own products may strengthen its technology position, but returns will ultimately depend on whether those products generate incremental advertising, commerce, cloud or subscription income.

Can NAVER convert AI factory capacity into revenue across cloud, agents and physical AI?

NAVER’s strongest opportunity may come from combining infrastructure with applications rather than competing as a commodity provider of processor time.

The company operates an ecosystem spanning search, advertising, shopping, payments, maps, cloud software, workplace applications, content and consumer-to-consumer marketplaces. AI agents developed across those services could perform commercial tasks such as finding products, comparing options, booking services, arranging payments and managing customer interactions.

NAVER has described its strategy as “action-oriented AI,” meaning systems that do more than provide information. An agent connected to search, commerce and NAVER Pay could potentially complete an entire transaction rather than directing the user through several separate services.

The company’s first-quarter results suggest that AI is already contributing to its existing businesses. NAVER reported that advertising revenue increased 9.3% year over year and that AI-related improvements accounted for more than half of that growth. Its enterprise business, which includes NAVER Cloud Platform, LINE WORKS and NAVER LABS, grew 18.8% as AI and digital-twin projects expanded.

Physical AI creates another possible market. NAVER LABS has invested in robotics, mapping, digital twins and spatial intelligence. The Seoul World Model could provide a simulation environment for robotics, autonomous systems, logistics and urban planning.

These applications may differentiate NAVER from data-centre operators that provide infrastructure without proprietary models or real-world datasets. The company can potentially sell customers an integrated offering covering computing, software, models and deployment support.

The risk is organisational complexity. Search, advertising, commerce, cloud, robotics and infrastructure have different sales cycles and financial characteristics. NAVER must ensure that the AI factory supports a coherent commercial strategy rather than becoming an expensive resource distributed across loosely connected experiments.

Does NAVER’s financial performance provide enough room for an infrastructure investment of this scale?

NAVER generated full-year 2025 revenue of KRW 12.035 trillion, an increase of 12.1%, while operating profit rose 11.6% to KRW 2.2081 trillion. Fourth-quarter revenue reached KRW 3.1951 trillion and operating profit was KRW 610.6 billion, giving the quarter an operating margin of 19.1%.

Growth accelerated during the first quarter of 2026. Revenue increased 16.3% year over year to KRW 3.2411 trillion, while operating profit rose 7.2% to KRW 541.8 billion. The slower increase in profit compared with revenue partly reflected infrastructure spending intended to secure NAVER’s AI competitiveness.

That gap illustrates the capital-allocation tension. NAVER’s core advertising, commerce and payments businesses continue to expand, giving the company financial capacity to invest. Yet AI infrastructure could absorb increasing amounts of capital before producing proportional revenue.

NAVER Platform generated first-quarter revenue of KRW 1.8398 trillion, Financial Platform contributed KRW 459.7 billion and Global Initiatives produced KRW 941.6 billion. The company therefore has a diversified operating base rather than relying solely on an emerging cloud business.

The Brookfield and NVIDIA arrangements may reduce the amount NAVER must fund directly, which could protect its balance sheet and preserve spending capacity for models, software and customer acquisition.

The remaining economic question is who owns the infrastructure, who receives the operating cash flows and how commercial returns are divided among the partners. The headline financing announcement does not provide enough detail to calculate NAVER’s expected return on invested capital.

Investors will need clarity on construction commitments, lease or supply arrangements, depreciation, power costs and revenue-sharing terms. A project can achieve technical completion while still delivering disappointing shareholder returns if capital costs are high or computing prices decline.

Does NVIDIA’s 4.5% stake provide validation or create unnecessary shareholder dilution?

NVIDIA’s investment provides powerful strategic validation. The semiconductor company is selecting NAVER as a major regional partner and committing capital alongside its computing platform.

The relationship could improve NAVER’s access to future processor generations and give the Korean company greater credibility when approaching international AI customers. NVIDIA’s participation may also reduce concerns that GAK Sejong will become a technologically isolated domestic project.

Existing shareholders, however, are giving up part of their ownership. The issuance of more than 7.2 million shares will dilute voting rights and future earnings per share unless the capital generates returns exceeding the economic cost of the new equity.

The placement price of KRW 204,500 has already become an interesting market reference point. NAVER shares rallied 8.43% to KRW 225,000 on July 27 after the equity investment was disclosed. The stock then fell 6.67% to KRW 210,000 on July 28 and declined another 4.05% to KRW 201,500 on July 29, moving below NVIDIA’s subscription price.

The reversal suggests that investors initially welcomed NVIDIA’s backing but quickly reconsidered the scale, dilution and execution risks. It may also reflect profit-taking after NAVER’s sharp July rally rather than a definitive rejection of the strategy.

Across the five sessions from July 22 to July 29, NAVER shares still gained about 2.3%, rising from KRW 196,900 to KRW 201,500. The stock nevertheless remained around 34% below its 52-week high of KRW 304,000 and had lost approximately 14% over the preceding year.

Current sentiment is therefore mixed. NVIDIA’s investment has improved the strategic narrative, but the market is not assigning an immediate premium to the project without clearer evidence on economics and delivery.

How does NAVER’s sovereign AI factory compare with South Korea’s wider infrastructure race?

NAVER is not developing the project in isolation. South Korea is pursuing a broader effort to become a major centre for AI processors, high-bandwidth memory, data centres and physical AI.

SK Group has announced plans involving NVIDIA for a substantially larger two-gigawatt AI data-centre programme, while SK Hynix is expanding its role in high-bandwidth memory used by advanced AI accelerators. Samsung Electronics is also pursuing major AI semiconductor, foundry, memory and advanced-packaging opportunities.

NAVER’s advantage is that it combines infrastructure with a major domestic internet platform. It has users, commercial services, payments, cloud operations, proprietary data and models that can create demand for computing capacity.

Its disadvantage is scale. Global hyperscalers and major Korean industrial groups can commit far larger amounts of capital. NAVER must therefore concentrate on areas where local data, Korean-language capability and integrated services matter more than the absolute size of the facility.

The company may also need to collaborate with competitors. Sovereign AI ecosystems require chip manufacturers, power providers, telecommunications companies, cloud operators, universities, startups and government agencies. No single company can build the entire value chain independently.

NAVER’s long-term ambition is to move beyond 200 megawatts toward gigawatt-scale infrastructure serving customers in South Korea and overseas. That vision creates substantial upside if the company becomes a recognised regional AI cloud operator. It also creates a risk of escalating capital commitments before the first phase demonstrates acceptable utilisation and returns.

What are the main execution risks facing the NAVER AI factory before 2028?

The first risk is construction delivery. Expanding from 55 megawatts to 200 megawatts requires electrical infrastructure, cooling, networking, servers and buildings to be installed on time and within budget.

The second risk is power availability. AI facilities require large and reliable electricity supplies. NAVER will need access to adequate generation and grid capacity while managing customer expectations around cost and sustainability.

The third risk is hardware supply. The project depends heavily on NVIDIA processors and networking technology. Production delays, supply constraints or export-policy changes could affect deployment schedules.

The fourth risk is utilisation. The facility needs committed customers capable of consuming large quantities of computing capacity. Speculative demand forecasts will not generate cash flow if startups, governments and enterprises delay projects.

The fifth risk is pricing. AI computing prices could decline as more capacity enters the market and processor efficiency improves. NAVER must recover infrastructure costs even if customers expect lower prices over time.

The sixth risk is concentration. A deep relationship with NVIDIA offers technical advantages but may reduce NAVER’s flexibility to adopt competing accelerators or negotiate future supply terms.

The seventh risk is margin pressure. Infrastructure depreciation, electricity, staffing and financing costs could cause profit growth to lag revenue for several years.

The eighth risk is strategic distraction. NAVER must continue defending and expanding its profitable search, advertising, commerce and payments businesses while managing a complex international infrastructure project.

What evidence will show whether NAVER’s $10 billion AI factory strategy is succeeding?

The first measurable milestone will be completion of NVIDIA’s share subscription, scheduled for October 30, 2026. Any change to the payment date, subscription terms or ownership outcome would affect the financing thesis.

The second milestone will be NAVER’s planned launch of its AI agent platform during the second half of 2026. Adoption across search, commerce, payments and enterprise customers would demonstrate that the company can monetise its models rather than merely finance more computing capacity.

The third milestone will be disclosed construction progress at GAK Sejong. Investors should look for clear delivery stages between the initial 55-megawatt deployment and the planned 200-megawatt capacity in 2028.

The fourth will be customer commitments. Long-term contracts with Korean enterprises, government organisations, startups or United States AI companies would reduce utilisation risk.

The fifth will be financial disclosure. Growth in NAVER’s enterprise and cloud businesses, accompanied by stable group margins, would suggest that infrastructure investment is producing commercial returns rather than simply increasing expenses.

The strategic logic behind the project is credible. South Korea needs access to large-scale AI infrastructure, NVIDIA wants regional cloud partners and NAVER has the data, products and operating experience needed to build services above the hardware layer.

What has improved is NAVER’s access to capital, advanced processors and global partners. What remains unresolved is whether the project economics can support a listed company whose shareholders are accustomed to profitable search, advertising and commerce operations.

The thesis would strengthen through signed customers, rising cloud revenue, successful agent-platform adoption and disciplined infrastructure spending. It would weaken if construction costs rise, capacity remains underused or operating-profit growth continues to trail revenue without a visible path to stronger returns.

What are the key takeaways from NAVER’s NVIDIA and Brookfield AI factory expansion?

• NAVER plans to expand its GAK Sejong AI factory from 55 megawatts to 200 megawatts by 2028.
• The enlarged facility could include approximately 100,000 NVIDIA graphics processing units.
• Brookfield has agreed to provide up to $9 billion as the exclusive capital partner.
• NVIDIA will invest approximately $1 billion through newly issued NAVER shares.
• NVIDIA is expected to receive a 4.5% stake after subscribing for 7,241,564 shares.
• The facility will use NVIDIA Blackwell, Vera Rubin and DSX infrastructure.
• NAVER plans to support external AI companies as well as HyperCLOVA X, agentic AI and physical AI projects.
• First-quarter 2026 revenue rose 16.3%, but operating profit increased only 7.2% as infrastructure investment weighed on growth.
• NAVER shares initially rallied after the announcement but closed below NVIDIA’s subscription price on July 29.
• Customer contracts, utilisation, construction progress and cloud margins will determine whether the project creates lasting shareholder value.