The AI Boom's Hidden Bottleneck Isn't Chips — It's Money

The Money Problem Nobody Is Talking About

Jensen Huang didn’t mince words at the DBS Asia Future Forward Conference in Singapore. While the world fixates on the next GPU launch or the latest benchmark race, the biggest threat to the AI era may be something far more mundane: there simply isn’t enough money chasing it.

Huang, CEO of Broadcom, has watched this cycle before. “Over the past two years we’ve had several bottlenecks, and we’ve overcome each one,” he said. The memory shortage right now? It may resolve in a year. Wafer capacity is expected to ease by 2028 or 2029. New power plants are coming online. But the single greatest constraint, he warned, is capital.

Lipu Tan, CEO of Intel, agreed — and went further. He flagged the rise in capital costs, driven by persistent global high-interest rates, as an urgent problem staring the industry in the face. Two CEOs. One shared conclusion: the AI boom is running into a wall it can’t out-compute its way through.

Why This Matters for Korea

The Korea connection isn’t incidental. It’s existential. Samsung Electronics — the world’s largest memory chipmaker and a cornerstone of the global AI supply chain — sits at the intersection of every problem Huang and Tan described. Memory demand is surging. Supply is constrained. Capital expenditure required to close the gap is enormous. And the cost of borrowing to fund that expansion is at multi-year highs.

Broadcom’s Huang framed OpenAI’s economics bluntly: the company is spending fortunes on model training and remains deeply unprofitable, with revenue potential from cutting-edge models unproven. That’s not unique to one company. It describes a sector-wide pattern where the gap between expenditure and return is widening, not narrowing.

For Korean policymakers and investors, the implication is sobering. Korea’s semiconductor sector contributes roughly 15 percent of GDP. Samsung alone accounts for around 20 percent of the KOSPI index. When AI capex demands accelerate and borrowing costs remain elevated, the pressure on Korean financial institutions to fund an industry whose returns are still uncertain becomes acute.

The Arcade Token Problem

Huang offered a particularly vivid analogy for where corporate AI adoption stands today. He compared it to a child given ten dollars worth of arcade tokens and spending all of it in ten minutes without learning anything.

“That’s what’s happening in companies today,” he said.

The picture is of organizations burning through AI infrastructure budgets — training models, deploying agents, building data pipelines — with little visible return. The infrastructure buildout is proceeding at hyper speed. The business value is nowhere near as clear.

Tan echoed the concern from the demand side. Both CEOs stressed that the long-term trajectory of AI growth depends on companies demonstrating that AI investments generate measurable profits. If next-generation models underwhelm or performance improvements stall, computing infrastructure demand could buckle.

The danger is asymmetric. On the upside, AI could unlock trillions in value. On the downside, billions in committed capex could face write-downs — and the companies most exposed are those that borrowed heavily to build the capacity.

The Timeline Nobody Is Discussing

Huang’s timeline suggests the current imbalance will persist through 2029 or 2030. Memory chip shortages may ease first; that segment could be the earliest to lose momentum if demand falters. Wafer capacity and power supply improvements follow later. The capex cycle stretches longer than most investors expect.

Tan cautioned that the next two to three years should see continued strong demand for computing power, driven by AI workload expansion. But he flagged memory as the first sector vulnerable to a slowdown. The sequence matters: if memory demand softens while the rest of the supply chain remains committed to expansion, the resulting oversupply hit would land hardest on Korean producers.

Samsung’s foundry business is already in a difficult position, trailing TSMC in leading-edge process nodes and struggling with yield rates. An AI demand correction would compound those structural challenges at precisely the moment the company needs to invest its way out of them.

The Energy Angle

Power supply was mentioned by both CEOs as a secondary but real constraint. New AI data centers require megawatts, not just chips. Huang noted that additional generation capacity could help ease the bottleneck. But building power infrastructure takes years, and in many regions it faces regulatory and environmental headwinds.

Korea’s own energy dynamics are relevant here. The country imports over 95 percent of its energy. Any shift in global energy demand driven by AI data center expansion carries direct implications for Korean energy security and costs. If AI-driven power demand pushes global electricity prices higher, Korean manufacturers — already among the most energy-intensive in the world — face a double squeeze: higher input costs and stiffer competition for capital.

What Happens Next

The message from Singapore wasn’t anti-AI. Both CEOs reaffirmed that near-term demand for AI computing infrastructure is overwhelming supply. The concern is financial sustainability. Capital is the binding constraint, not technology.

For the global market, the takeaway is straightforward: AI infrastructure buildout will continue until money runs out, not until demand is satisfied. The question is what happens when the two converge. For Korean markets specifically, the convergence risk is higher — the concentration of semiconductor exposure, the reliance on debt financing, and the pace of required investment all amplify the vulnerability.

The pink-hued forecasts from Silicon Valley are not wrong. They are incomplete. The bottleneck isn’t a lack of ingenuity. It’s a lack of capital — and in a world of elevated rates, that bottleneck has a different shape than anyone expected.

The Bottom Line

The AI boom is real. The funding gap is the risk. Companies, investors, and governments who treat capital as an infinite resource will pay for that assumption. Those who plan for scarcity will be better positioned when the music stops — and it will stop eventually.

Korea’s exposure makes this not just a business story. It’s a macro story with implications for export competitiveness, financial stability, and industrial policy. The question for Seoul isn’t whether AI will transform the economy. It’s whether the funding model behind the transformation is sustainable.