Core Insights - The global AI arms race is hitting a physical wall due to TSMC's production capacity constraints, leading to a significant supply-demand gap in the semiconductor industry [1] - Major tech companies like NVIDIA and Google are struggling to secure sufficient chip supply from TSMC, which is currently unable to meet the surging demand [2] - TSMC's production lines are under pressure from both AI chip demand and traditional client orders, complicating capacity allocation [3] Group 1: Demand Surge and Allocation Challenges - TSMC is facing a difficult balancing act between maintaining stability for existing clients and addressing the unpredictable demand from the AI sector [3] - The demand for chips is driven by multiple factors, including OpenAI's plans for super data centers and Google's aggressive procurement of NVIDIA GPUs [3] - TSMC adheres to strict annual schedules for capacity and pricing negotiations, limiting flexibility for clients to adjust orders based on market conditions [3] Group 2: Expansion Plans and Limitations - TSMC is adjusting its global footprint to address capacity shortages, including shifting a new factory in Japan to produce advanced 2nm chips, expected to be completed by 2027 [4] - The company is accelerating the construction of a second factory in Arizona, aiming to start 3nm chip production a year earlier than planned in 2027 [4] - Current expansion efforts will not resolve immediate capacity issues, as TSMC is primarily redesigning existing factory space to accommodate new production lines [4] Group 3: Investment Caution Amid Cyclical Nature - Despite the booming AI demand, TSMC is cautious about committing to new factory constructions due to the cyclical nature of the semiconductor industry [6] - Building a cutting-edge fab costs billions and takes years, while demand can fluctuate rapidly, as seen during the pandemic [6] - TSMC's pure foundry model limits its investment flexibility, as it relies entirely on customer orders and faces risks of idle capacity if clients cancel orders [6] Group 4: Packaging Bottlenecks - Advanced packaging has emerged as another critical bottleneck, essential for high-end AI chips [7] - TSMC has reallocated some older chip production capacity to advanced packaging, but the complexity of the process remains a challenge [7] - NVIDIA has previously faced packaging capacity shortages, leading to difficulties for other clients like Google when trying to increase their orders [7]