Artificial intelligence is emerging as one of the world’s largest new sources of demand for capital. As hundreds of billions of dollars flow through bond markets and private credit to finance data centers, GPUs and power grids, the AI investment race is moving beyond technology. The decisive question is increasingly who can secure long-term capital—and afford its cost.
The striking development is that Federal Reserve policy and market interest rates are moving in different directions. In a Reuters survey conducted from August 12 to 17, most economists expected the Fed to keep its benchmark rate at the current 3.50%–3.75% through the end of the year. Yet the yield on the 30-year U.S. Treasury climbed above 5.3% this week, reaching its highest level since 2007.
AI alone cannot explain the rise in long-term rates. The U.S. fiscal deficit, heavy Treasury issuance, inflation concerns and weakening foreign demand for U.S. government debt are all contributing. What is new is the additional pressure from record private-sector demand for capital to fund AI infrastructure.
Is AI Beginning to Move Global Capital?
Amazon, Alphabet, Meta and Oracle issued approximately $194 billion in corporate bonds through July 7 this year, 79% more than the roughly $108 billion issued during all of 2025. Goldman Sachs projects that bond issuance by five hyperscalers, including Microsoft, will reach about $250 billion this year and $400 billion in 2027.
The scale becomes even larger when AI-related borrowing is viewed more broadly. Reuters estimated that hyperscalers had recently raised about $220 billion for AI infrastructure this year, while the real yield on 30-year Treasury Inflation-Protected Securities approached 3%, near an 18-year high. Governments and AI companies are now competing for a limited pool of long-duration funding, intensifying what markets describe as “capital scarcity.”
The shift is also visible in corporate bond demand. Investor orders for hyperscaler bonds fell from about five times the amount offered in February to less than two times in July. As investors demand higher yields to absorb new supply, companies’ actual financing costs are rising.
This is where the distinction between policy rates and the cost of capital becomes critical. Even if the Fed does not raise its benchmark rate, higher Treasury yields and wider corporate bond spreads can increase the long-term financing costs borne by data center operators. For AI companies, the rate at which they can secure capital for several decades is becoming as important as whether the Fed raises rates.
Alphabet’s first Australian bond issuance on August 19 illustrates this trend. The company raised A$5.5 billion, or approximately $3.89 billion, as the AI investments of major technology companies are expected to exceed $730 billion this year. Even companies with enormous cash-generating capacity are expanding beyond the U.S. dollar bond market to tap capital worldwide.
What Nvidia’s $105 Billion Guarantee Reveals About AI Finance
Nvidia’s latest move shows how far the competition for AI capital has expanded. The company has agreed to provide guarantees of up to $105 billion to support a 20-year lease for an Ohio data center that will be used by OpenAI. Developed by SoftBank affiliate SB Energy, the project is intended to reach 8 gigawatts of capacity, with the first 800 megawatts scheduled to become operational by 2028. Nvidia is also investing $1.5 billion directly in SB Energy.
The transaction is notable because the world’s largest supplier of AI chips is assuming part of the financing risk for a customer facility that will use its GPUs. Nvidia is also working with Goldman Sachs, Apollo, Blackstone, BlackRock, Brookfield and KKR on a financial platform intended to channel more than $500 billion into AI infrastructure. Goldman Sachs is discussing ways to bring in capital from insurers and asset managers by combining private credit with bond financing.
The structure is straightforward. AI models generate greater demand for computing power, and data center operators secure GPUs and power infrastructure to meet that demand. As the capital required becomes too large for conventional financing alone, Nvidia is moving beyond supplying GPUs and joining financial institutions in supporting data center funding.
As this structure expands, however, circular financing warrants closer scrutiny. Nvidia supports the financing of data centers, those facilities purchase Nvidia GPUs, and the resulting sales flow back into Nvidia’s revenue. Reuters has noted that the $105 billion guarantee raises precisely this concern.
Circular structures are not inherently abnormal. Vendor financing, long-term purchase agreements and project finance are common in capital-intensive industries such as power generation, aviation and telecommunications. The more important question for AI is whether the computing capacity created through this financing can generate sufficient cash flow.
If it can, today’s borrowing will accelerate the construction of essential AI infrastructure. If data center investment outpaces the monetization of AI services, however, high interest expenses, GPU depreciation and power costs could converge into a significant financial burden.
In Korea, the AI Capital Boom Is Turning into Semiconductor Profits
The same AI capital boom is taking a different form in Korea. Backed by the AI semiconductor upcycle, the Korean government expects the economy to grow 3.0% this year, its strongest pace in five years. Moody’s raised its forecast for Korea’s 2026 growth from 1.8% in February to 2.5% in May and then lifted it by another percentage point to 3.5%.
In only six months, Moody’s nearly doubled its projection. Its latest forecast exceeds both the government’s 3.0% estimate and the 3.2% average projected by eight major investment banks. Moody’s cited the semiconductor boom and robust exports, emphasizing that AI memory demand is strengthening the profitability and cash-generating capacity of Korean chipmakers.
S&P Global also expects memory prices to remain strong because of AI demand and constrained supply growth. It assesses Korean companies as being among the best positioned to benefit from the current supercycle in both high-bandwidth memory and conventional memory.
Viewed together, developments in the United States and Korea reveal the two sides of the AI investment boom. In the United States, the immense demand for capital to build data centers and power infrastructure is reshaping bond markets. In Korea, the HBM and advanced memory installed in that infrastructure are driving exports and economic growth.
But if Korea remains solely a memory supplier, much of the higher-value financing and infrastructure returns generated by AI investment could accrue to U.S. technology companies, asset managers and private capital firms. Alongside capacity expansion by Samsung Electronics and SK hynix, Korea needs an industrial structure that connects power, data centers, advanced packaging and long-term finance.
In the AI era, having access to abundant money is not enough. Companies and countries must be able to raise trillions of dollars in capital over time, convert it into productive semiconductor, data center and power capacity, and then generate sufficient cash flow from those assets. With long-term rates rising even as the Fed stands still, the new cost of competing in AI is beginning to emerge not in the price of GPUs, but in the price of capital.
