Hyundai Motor Securities projected that growth in U.S. frontier artificial intelligence (AI) models will be confirmed again starting in the third quarter this year. As demand that had shifted to China's low-priced AI models returns to U.S. models—centered on high-performance coding, reasoning, and agentic AI—the concern over AI revenue monetization is also expected to ease.
Kim Jae-seung, an analyst at Hyundai Motor Securities, said in a report on the 21st that the slowdown in OpenAI's second-quarter revenue growth should not be taken as a weakening of its AI model competitiveness. OpenAI's second-quarter revenue rose only 18% from the previous quarter. Compared with Anthropic's revenue more than doubling in the same period, the growth appears to have slowed.
However, because second-quarter revenue is a lagging indicator, the recent trend should be examined together. GPT-5.6 Sol, which OpenAI unveiled in June, received favorable reviews for coding performance and price competitiveness, and revenue from Codex, a developer-focused AI service, is also rising quickly.
TickerTrends' estimated tracked annual recurring revenue (Tracked ARR) for Codex rose about 50%, from $5.88 billion on July 6 to $8.83 billion on Aug. 10. Over the same period, Anthropic's Claude Code rose only 6%, from $14.26 billion to $15.12 billion. Kim interpreted this as a signal that OpenAI's growth axis is shifting to the developer-centered Codex and enterprise AI.
Changes are also appearing in global AI usage. By country, the token demand index, which shows paid AI inference token usage, rebounded this month for U.S. AI models after growth stalled last month. The analysis is that some demand that had shifted to China's low-priced open AI models since May is returning to the United States' closed frontier models.
In particular, corporate demand for U.S. AI models is expanding in areas that require high performance, such as coding, reasoning, and agentic AI. Since the 16th, the token expenditure index, which shows AI token usage expense, has also rebounded. As the token expense index for closed AI rose while that for open models fell, corporations are beginning to choose high-performance models that deliver better work outcomes rather than simply cheaper models.
The low-price offensive by Chinese AI firms is also unlikely to continue for long. Considering the expense of improving model performance and running services, there are limits to continuously lowering prices. As the criteria for selecting AI models shift from simple per-token expense to the expense of actually completing tasks, U.S. frontier models could gain competitiveness.
Kim also cited the fact that Blackwell graphics processing unit (GPU) computing rental prices in the United States remain high as evidence that demand for high-performance AI is continuing. The analysis is that demand for coding and agentic AI, which require high-performance inference, is connecting to demand for high-performance computing.
Hyundai Motor Securities projected that, starting in the third quarter, as demand shifts to high-performance U.S. frontier AI models, the concern that has been raised over AI revenue monetization will gradually diminish.