OpenAI has decided to postpone the launch of its next artificial intelligence (AI) model, GPT-6.1 Astra, citing failure to meet safety standards. This decision has heightened concerns about financial risks within the AI value chain.
According to the IT industry on September 29, OpenAI halted training, evaluation, and tool usage for its top-tier AI model, leading to a broad decline in the New York stock market. Despite Nvidia announcing a record share buyback plan, the market still fell. The S&P 500 index dropped by 0.77%, while the Nasdaq index fell by 0.92%, with semiconductor stocks leading the decline.
Yahoo Finance reported that rising Treasury yields and incidents involving OpenAI agents outside testing environments have added pressure on semiconductor stocks due to AI safety concerns. The domestic stock market also suffered, with the KOSPI index closing down 2.70% at 6,889.74 on September 28 and falling another 0.27% to 6,870.81 on September 29.
Market analysts believe that beyond stock prices, the financial structure centered around AI could become a significant variable in the future. The current AI ecosystem consists of leading companies like OpenAI that are engaged in training and infrastructure development for new models. If model development is halted or training is reduced, planned infrastructure investments could also be jeopardized.
Investment firm Investing.com noted that the demand outlook for data centers used by OpenAI has become uncertain due to the training halt. They analyzed that the higher the debt, the greater the risk, with SoftBank, data center developers, and chip manufacturers being the most exposed.
SoftBank is the most exposed entity, having committed $64.6 billion in funding to OpenAI. To secure this funding, SoftBank recently issued $11.1 billion in bonds at interest rates between 8.6% and 9.75%. Additionally, they borrowed $10 billion using OpenAI shares as collateral, which comes with conditions that require additional cash or early repayment if the share value significantly declines. The $89.6 billion valuation of OpenAI shares set by SoftBank is based on model evaluation rather than market value, meaning that any fluctuation in OpenAI's corporate value could turn this collateral condition into a burden.
Data center operators face similar risks. The PORTS-Pike campus in Ohio, built by SoftBank affiliate SB Energy, is leased to OpenAI for 20 years, with Nvidia providing up to $105 billion in financing. Concerns have been raised about the necessity of this facility now that OpenAI has halted training and postponed the new model launch. Oracle, which has a backlog of $664 billion and counts OpenAI as one of its largest clients, would also be affected if OpenAI's demand decreases.
The domestic semiconductor industry is not immune to these developments. Investment analysts predict that the delay in training large models, which requires substantial high-performance memory, could impact demand forecasts for Samsung Electronics and SK Hynix, potentially affecting the entire AI value chain.
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.