The global AI boom has entered a more consequential phase, where the question is no longer how quickly companies can build computing capacity, but whether that capacity can generate enough economic value to sustain itself. SK Group Chairman Chey Tae-won has placed that question at the center of the industry’s next chapter, arguing that the extraordinary resources flowing into artificial intelligence ultimately need to produce their own returns. His warning cuts through the familiar excitement surrounding models, chips and data centers and moves the discussion toward a harder test: whether AI can mature from an investment story into a durable business.

Speaking at the 2026 Ulsan Forum at the Ulsan Exhibition & Convention Center, Chey argued that companies cannot indefinitely rely on enormous capital commitments without demonstrating how those investments will produce sustainable commercial outcomes. His position carries particular weight because SK Group sits across semiconductors, energy and infrastructure, giving the chairman a view of AI’s expansion that extends beyond software and into the physical systems required to support it. The emerging challenge, in his framing, is therefore not simply building more AI, but creating an ecosystem capable of financing, powering and commercially absorbing that expansion.

AI investment now faces a harder commercial test

Chey’s central concern is straightforward, but its implications reach across the entire AI infrastructure economy. “In addition to the existing three speed, scale and safety, we also need to consider business models and sustainability,” Chey said at the 2026 Ulsan Forum on September 11 at the Ulsan Exhibition & Convention Center. Chey warned that companies had poured enormous investment into AI and that the industry could fizzle out like a bubble if those investments failed to generate returns. The distinction matters because AI infrastructure requires commitments that often precede visible revenue, leaving companies to make large decisions today against uncertain commercial outcomes tomorrow. That creates a structural tension between the urgency to secure compute and the need to prove that each additional layer of investment can eventually support itself.

The scale of the spending illustrates why the sustainability question has become harder to dismiss as a theoretical concern. Capital spending by the world’s five largest technology companies could reach $791 billion this year and $1 trillion next year, according to JPMorgan. Those projections have intensified debate among investors and industry executives about whether the current AI expansion represents a durable technology cycle or an investment boom vulnerable to disappointment. The issue does not rest solely on the size of the checks companies write, because the composition of financing also reveals how technology companies approach the infrastructure race. The Bank of Korea said the five companies issued $169.15 billion in corporate bonds during the first half of this year, equal to 51.3% of their capital spending.The AI economy needs a return path

The AI economy needs a return path

Chey described the investment challenge in unusually blunt terms, emphasizing that AI spending has moved beyond the scale associated with many conventional technology programs. “I am worried if we make huge investments, but there is not much return,” Chey said. “If all of humanity has put in this much energy, we need to create better business models and commercialize them,” Chey said. The statement shifts the debate away from whether AI investment should continue and toward what mechanisms can turn infrastructure spending into recurring economic value. For infrastructure providers, that means the commercial proposition cannot depend indefinitely on expectations that demand will eventually catch up with capacity. For technology companies, it means increasingly expensive compute commitments will need stronger connections to products, services, enterprise adoption and other sources of revenue.

The question becomes especially important as AI infrastructure expands across several interdependent layers. Semiconductor capacity, data center construction, networking, cooling, electricity generation and grid connections all compete for capital while depending on one another for successful deployment. A new data center cannot create meaningful AI capacity without sufficient power, while power infrastructure alone does not generate value unless customers can use the resulting compute resources. Likewise, larger computing clusters create new requirements for thermal management and energy delivery, making infrastructure planning increasingly inseparable from commercial strategy. Chey’s argument effectively places a financial discipline across this chain, asking whether each investment ultimately contributes to a system that can sustain further investment. That perspective is particularly relevant for companies planning long-lived infrastructure around technologies that continue to evolve rapidly.

Manufacturing AI starts with data infrastructure

Chey’s comments extended beyond the economics of hyperscale AI to the industrial systems needed to bring artificial intelligence into manufacturing. His message was that manufacturers should establish the underlying data environment before expecting AI systems to transform production. “Manufacturing AI is not easy without a data platform of scale,” he said. “Gathering data and expanding the volume of data that is needed is a national challenge for South Korea, and doing that first raises the odds of success for manufacturing AI.” The observation highlights a frequently overlooked distinction between deploying an AI model and building an industrial AI system. Manufacturing environments generate information across machines, processes, supply chains and facilities, but that information only becomes useful when organizations can collect, structure and make it available for computation. Without that foundation, the promise of manufacturing AI can remain disconnected from the physical operations it is supposed to improve.

South Korea’s industrial position makes the issue particularly significant because manufacturing remains closely tied to the country’s economic and technological identity. AI adoption in factories requires more than access to advanced models, since organizations must create systems capable of connecting operational information with decision-making processes. That places data infrastructure alongside computing capacity as a strategic prerequisite for industrial AI. It also means that national AI strategies increasingly intersect with industrial policy, infrastructure planning and digital capabilities. Chey’s emphasis on data platforms suggests that the competition for AI leadership will not be determined solely by access to the most advanced models. Countries and companies that can organize the underlying information environment may have a stronger foundation for turning AI capabilities into industrial productivity.

Ulsan becomes a test of AI infrastructure ambition

Chey’s remarks also offered a glimpse into how SK Group is translating its AI strategy into physical infrastructure. SK has partnered with Amazon Web Services to develop an AI data center in Ulsan’s Mipo National Industrial Complex, with the facility originally planned at 103 megawatts. Chey said discussions with major technology companies had made “considerable progress,” while the project has added 900 megawatts to the original plan. That expansion brings the project to a total of 1 gigawatt, according to Chey’s remarks, marking a substantial change in its intended scale. The development illustrates how quickly infrastructure planning can evolve as expectations for AI compute demand change. It also demonstrates why power availability has become inseparable from the strategic conversation surrounding large AI data center projects.

A 1-gigawatt ambition carries implications far beyond the walls of a single facility. Such a project requires coordination among data center developers, technology companies, utilities, power providers and public authorities, because computing capacity cannot expand independently of electricity infrastructure. The location of Ulsan also connects the project to South Korea’s broader industrial ecosystem, where manufacturing and energy infrastructure already play major roles. The strategic proposition is therefore broader than simply adding another large data center to the country’s infrastructure portfolio. It involves creating an environment where computing, industrial demand and energy systems can develop together. For SK Group, the project provides a practical setting in which Chey’s argument about sustainable AI investment can be tested through real infrastructure commitments.

Energy infrastructure becomes part of the AI business model

Chey’s comments about SK Innovation point toward another part of the strategy: linking AI growth with the energy systems required to support it. “using AI to create energy solutions and building the electrification solutions needed for AI data centers.” He added that it “has many global partners and investment plans, and will expand them further.” The remarks show that SK sees AI infrastructure as extending beyond computing into energy solutions and the electrification systems required by AI data centers. That approach reflects a broader shift in the economics of AI infrastructure, where access to reliable electricity increasingly influences the viability of computing projects. As data center developers seek greater capacity, energy infrastructure becomes a strategic input rather than a background utility.

The relationship creates a potentially important feedback loop for diversified infrastructure groups. AI requires electricity, but AI can also be used to improve energy systems, creating opportunities for technology and infrastructure to reinforce each other. That does not eliminate the economic risks associated with large capital programs, but it can broaden the sources of value available to companies operating across the infrastructure chain. SK’s positioning across semiconductors, energy and AI infrastructure gives it an opportunity to approach those connections as a portfolio rather than as isolated investments. However, the long-term test remains the same one Chey identified at the forum: the investments must ultimately produce sustainable business activity. Infrastructure can enable AI growth, but only durable demand can justify the continued expansion of the infrastructure base.

The next AI race may be about economic durability

Chey’s remarks put greater emphasis on whether the industry’s enormous infrastructure investment can develop into business models that generate returns and support continued investment. Massive capital commitments can accelerate technological development, but they also increase the consequences of weak demand, underused infrastructure or business models that fail to capture sufficient value. Chey’s warning does not amount to a rejection of AI investment, nor does it suggest that the current expansion will inevitably collapse. Instead, it identifies the point at which enthusiasm must meet economics. AI companies and infrastructure providers will need to demonstrate that the capital entering the sector can generate enough value to support the next round of investment. The emphasis on monetization places business-model development alongside compute capacity, data centers, energy infrastructure and industrial AI in the sustainability discussion.