$638 Billion in Remaining Obligations Increases AI Credit Burden

Investment in AI infrastructure is shifting from expectations of growth in tech stocks to a phase where long-term contracts and credit burdens are being considered together. The surge in Oracle's Remaining Performance Obligations (RPO), OpenAI's long-term computing agreements, and data center power contracts all point in the same direction.

Oracle reported on June 10 that its RPO for fiscal year 2026 increased by 363% year-on-year to $638 billion (approximately 883.63 trillion won). The same report indicated that the free cash flow for fiscal year 2026 was a deficit of $23.7 billion (approximately 32.82 trillion won), with debt financing at $43 billion (approximately 59.55 trillion won) and equity financing at $5 billion (approximately 6.93 trillion won).

RPO refers to the remaining performance obligations that have already been contracted but not yet recognized as revenue. In the cloud and data center business, it serves as an indicator of the visibility of future revenue, but as the scale of long-term contracts increases, the costs of fulfillment and the counterparty's payment ability must also be considered.

The issue is not the demand for AI itself. It is the fact that securing data centers, graphics processing units (GPUs), and power contracts first has become a larger structure that recoups costs through usage fees in the future. Groundbreaker argued in a column on July 2 that the AI boom should be viewed as closer to a credit and real estate infrastructure cycle than a technology cycle.

This column interpreted that risks emerge when the growth rate slows down before the acceleration of growth diminishes. However, this is an analysis by the original author and not a confirmed fact through disclosures. It is difficult to definitively state the total amount of agreements in the entire AI industry or the repayment burden for 2027-2028 based solely on public figures.

Official documents also confirm long-term agreements. OpenAI explained that customers can choose 1-3 year agreements for its Guaranteed Capacity product. Additionally, on July 22, it announced a power contract with Georgia Power for 3.2 gigawatts at the Effingham County data center project, stating it would gradually supply power from 2028 to 2032.

OpenAI also stated on January 14 that it would gradually introduce low-latency AI computing with a scale of 750 megawatts in collaboration with Cerebras by 2028. The method of securing power and computing capacity over several years shows that AI infrastructure investment is shifting from short-term server expansions to long-term facility contracts.

The Bank for International Settlements (BIS) evaluated on July 14 in a working paper that AI investment is one of the largest technology-led investment booms in U.S. history. The document analyzed that winner-takes-most competition, debt, and circular equity structures could exacerbate over-investment and financial vulnerability. This aligns with reports that AI agreements have accumulated as burdens off the books of big tech companies.

There are counterarguments as well. OpenAI stated in a company post on July 31 that the value of AI infrastructure lies not in its scale but in enabling more affordable and widely used intelligence. Oracle also mentioned that $75 billion (approximately 103.87 trillion won) of its large AI contracts are for customer prepayments or customer-supplied hardware, explaining that this structure reduces the capital financing burden for data center construction.

Reuters reported on July 22 that capital expenditures by hyperscalers such as Microsoft, Alphabet, Amazon, Meta, and Oracle could exceed their free cash flow in 2027. The same article analyzed that operating cash flow in 2027 would increase by about $340 billion (approximately 470.9 trillion won) compared to 2025, while capital expenditures would increase by about $534 billion (approximately 739.59 trillion won).

Hyperscalers are companies that operate large-scale data centers and cloud infrastructure. They must secure GPUs, network equipment, cooling facilities, and power contracts simultaneously to respond to the demand for AI model training and inference. Even if revenue follows, the initial investment and long-term contract burdens are reflected in the books and cash flow first.

Community opposition has also emerged as a variable in infrastructure expansion. Reuters reported on July 18 that there were 142 protests against data centers in 42 U.S. states as of that date. As the burden of expanding the power grid, water usage, and transmission facilities increases, data center investments could become a political issue not only in financial markets but also in local politics.

For domestic investors, this trend leads to a consideration of U.S. tech stocks and risk asset evaluations together. If AI infrastructure investment affects corporate bond issuance and long-term interest rate trends, the discount rates for tech stocks and the valuations of growth stocks could also come under pressure.

The direct connection with the crypto market is still limited. However, if the AI investment cycle influences the sentiment towards tech stocks and risk assets, the liquidity environment for Bitcoin (BTC) and Ethereum (ETH) could also be affected. There has been no evidence presented that could explain specific token prices solely based on AI infrastructure agreements.

The AI boom is simultaneously boosting semiconductor and cloud revenues while testing the power grid, borrowing, and cash flow. OpenAI's power supply in Effingham County, Georgia, is expected to be gradually implemented from 2028 to 2032.