The Bank of Korea forecast that the unprecedented semiconductor boom, powered by investment demand for artificial intelligence (AI), will continue through the first half of next year. It assessed that whether the expansion will keep going is “highly fluid,” saying it depends on when US big tech companies that are pouring money into AI can begin to earn profits and whether they can secure funding stably until then.

On the 12th, in a report titled ‘Assessment of the Sustainability of the Global Semiconductor Upcycle,’ the Bank of Korea projected that “given constraints on semiconductor supply, the expansion phase of the semiconductor cycle is expected to continue at least through the first half of next year.”

Current demand for memory semiconductors relies entirely on AI investment. As the United States leads a large-scale buildout of infrastructure such as AI data centers, demand has risen in tandem for high-bandwidth memory (HBM) installed in AI accelerator chips and for general-purpose DRAM. Buoyed by this demand, Samsung Electronics recorded its largest-ever operating profit of more than 57 trillion won in the first quarter of this year, and SK hynix is also expected to post more than 40 trillion won in profit.

While demand is large, supply is showing a ‘bottleneck.’ HBM has high process complexity, so mass production takes considerable time. Although domestic memory manufacturers have recently been rushing to expand facilities, it is not enough to meet the growing demand. The report assessed that “in this upturn, supply-demand imbalances are becoming larger and more prolonged than in the past due to supply constraints.”

The key question is whether the semiconductor boom will continue beyond the first half of next year. The report analyzed that a range of variables must be considered to judge this.

First, it cited ‘the profitability of AI investment.’ Depending on when companies that have invested in AI infrastructure begin to earn meaningful revenue, the durability of semiconductor demand could be determined.

The report predicted that “from next year, when market attention is expected to shift gradually toward actual monetization potential, it will be difficult to expand AI infrastructure investment at the same pace as last year and this year.” Although AI firms such as OpenAI and Anthropic are currently investing competitively to secure leadership, if monetization proceeds more slowly than expected, semiconductor demand could slacken.

Another key factor is how long big tech companies can maintain stable cash flows. The report noted that “as the scale of AI infrastructure investment has recently expanded, it has become difficult to cover it with internal funds alone,” adding that “as cash has been depleted, since the second half of last year big tech firms have reduced share buybacks and sharply increased corporate bond issuance.”

In addition, if technologies are developed that deliver the same computational performance while using far less memory, this could negatively affect demand for memory semiconductors. Google's ‘Turbo Quant’ technology is a representative example. The report expressed concern that “memory-saving technologies could also worsen the profitability of memory already invested.”

The report also pointed to variables such as the pace of capacity expansion by Samsung Electronics and SK hynix, increased supply and technological catch-up by Chinese memory companies, and others as factors that will determine the continuation of the semiconductor cycle’s expansion.

The report also stated, regarding the impact of a war between the United States and Iran on the semiconductor economy, that it would be “limited at the current stage,” adding that “if the war situation and financial and economic repercussions become more severe, the timing for verifying the profitability of AI investment could be brought forward or big tech could face some difficulties in securing funds, leading to reduced semiconductor demand.”