The company is making expensive efforts to increase the capacity for 3nm AI chips. Taiwan Semiconductor Manufacturing (TSM -3. 45%) adjusted its semiconductor industry outlook for the remainder of the year due to ongoing macroeconomic and geopolitical uncertainty. TSMC, the leader in the foundry market, now anticipates a more gradual and moderate recovery for this year. Despite the mixed overall demand, TSMC projects a significant rise in AI accelerator production compared to 2023. The company expects AI accelerator revenue to more than double in 2024, making up a low-teens percentage of total sales. Over the next five years, TSMC forecasts a 50% annual growth rate for AI accelerator revenue. Overcoming challenges to increase capacity TSMC has encountered challenges with its 3-nanometer process node. While Apple employs this advanced process for its custom chips, AI accelerators have been limited to older 5nm technology. Although speculation suggested that Nvidia's new Blackwell data center GPUs would utilize TSMC's 3nm node, they will actually employ a modified version of the same process as their predecessors. To enhance its capacity for 3nm AI chips, TSMC is taking extraordinary measures. The company plans to convert some of its 5nm tools to support the 3nm process, consequently reducing its 5nm capacity to accommodate additional 3nm capacity. However, this move will impact profitability. TSMC expects a 1 to 2 percentage point reduction in gross margin in the second half of this year. Although the company has maintained its long-term gross margin forecast, it will work toward cost-cutting measures to counteract the impact on gross margin. Unfortunately, there are limitations to how much additional capacity for newer nodes TSMC can create using this approach.

When asked by an analyst during the earnings call if it would be possible to bolster 5nm capacity by reallocating resources from the currently underutilized 7nm node, TSMC CEO C. C. Wei explained that the physical proximity of the equipment for 5nm and 3nm made this move viable, but it would not be feasible for other nodes. TSMC anticipates a smoother process for its upcoming 2nm node. Wei believes that the 2nm node will have more tape-outs within the first two years compared to the 3nm or 5nm nodes, largely due to the demand related to AI. Volume production for the 2nm node is expected to commence in 2025. The wildcard of Intel While the demand for AI chips continues to rise, TSMC is poised to face fierce competition from Intel (INTC -2. 40%) in the coming years for semiconductor foundry dominance. Intel aims to become the world's second-largest foundry by 2030, relying on new process nodes that could surpass TSMC technologically. TSMC remains optimistic that its 2nm process will prevail; however, Intel is equally confident in its Intel 18A process, set to enter production early next year. Intel holds a critical advantage with its utilization of backside power delivery technology in Intel 18A, which can significantly improve performance and efficiency. TSMC will not implement backside power delivery until the second generation of its 2nm nodes. While the demand for AI chips will contribute to TSMC's growth in the future, the market will not be exclusive to the company. The entry of capacity from Intel and Samsung that rivals TSMC's top process nodes could exert pressure on its market share and profitability.