Only Investing, Not Earning? Will the AI Rebound Continue?

Director Park Moon-hwan emphasized that for stock price increases to transcend a mere technical rebound and truly alter market trends, clear conditions are essential. This hints that recent stock gains might not have led to fundamental problem-solving.

The situation rapidly changed as Chinese AI startups gained attention in the global market. Open-weight models like Moonshot AI's Kimi K3 quickly matched the performance of U.S. big tech companies, altering the criteria for market competition. Beyond simple performance comparison, price, efficiency, and the entire development ecosystem have become competitive factors. This signifies the arrival of an era where the highest-performing chips do not necessarily guarantee the highest profits.

OpenAI drastically cut the price of its latest GPT model by 80% just three weeks after its release. This was a decision made to survive in a competitive market. However, such price reductions raise a fundamental question about how profitability will be maintained amid rising infrastructure costs.

Director Park Moon-hwan pointed out that the AI industry is caught in a fundamental dilemma. Infrastructure costs like electricity and semiconductors continue to rise, but model prices must be lowered to win in competition. This creates a situation where revenue decreases while investment scale must be maintained. Ultimately, questions arise about whether profitability can be sustained while maintaining the current pace of investment.

According to a report published by the Bank of Canada on June 18, the estimated annual revenue from AI services worldwide, excluding China, is approximately $217.5 billion.

In an estimate released by Exponential View on June 25, the combined revenue from major companies' apps and frontier model AI cloud amounted to $175 billion.

Ramp, a company specializing in automated corporate spending management, estimated AI service expenditures at $174.5 billion based on actual payment data from over 50,000 U.S. companies. The convergence of these three estimates from different institutions, all within the $170-220 billion range, increases the reliability of this figure.

A critical imbalance emerges here. The AI industry needs an annual revenue of $2.56 trillion to sustain its current level of investment. However, actual revenue is only $220 billion, about one-tenth of what is required. Director Park Moon-hwan emphasized that this is not just a problem of growth metrics, but a sign that sustained investment itself may become impossible.

News broke that OpenAI's active users surpassed 1 billion, but Director Park Moon-hwan viewed this as signifying the end of growth. Excluding children, the elderly, and those in remote areas from the globally available population, it means that most people who can realistically use the service are already doing so. In this state, increasing revenue by more than tenfold is practically impossible. Ultimately, the growth strategy reliant on user increase has hit a wall.

Director Park Moon-hwan meticulously analyzed the structure in which the AI industry is caught in a vicious cycle. As the model refresh cycle shortens, training costs continuously increase. Simultaneously, Chinese startups enter the market with low-price competition, necessitating a reduction in token prices. As a result, prices decrease, but the burden of computing capacity rental costs does not diminish at all. On top of this, rising interest rates increase capital procurement costs. Director Park's analysis concludes that there seems to be no way out of this vicious cycle where revenue decreases while investment costs continuously rise.