NVIDIA CEO Jensen Huang's recent call for "More HBM" is a clear signal of the company's aggressive strategy to secure memory for its next-generation AI supercomputer.
This isn't just a casual request; it's a direct reflection of a critical supply chain reality. The primary driver is NVIDIA's new 'Vera Rubin' platform, which recently entered full production. As NVIDIA prepares to ship these powerful systems in the second half of 2026, the demand for High-Bandwidth Memory (HBM) has become a confirmed, urgent need rather than a forecast. This makes HBM the single most significant bottleneck in the entire AI hardware ecosystem.
So, why is this happening now? The causal chain is clear. First, NVIDIA's official announcement of the Rubin platform's production ramp-up in early June turned future demand into immediate orders, intensifying the race to secure components. Second, NVIDIA's stellar Q1 earnings confirmed that demand for its AI chips continues to outstrip supply, giving the company immense leverage to demand priority allocation from suppliers. This leads directly to SK hynix, which has solidified its role as the dominant HBM provider.
SK hynix itself has stated that customer requests for the next three years already exceed its production capacity. This scarcity gives NVIDIA a strong incentive to publicly call for more supply, reinforcing its partnership with SK hynix and ensuring it remains at the front of the line. Adding another layer of complexity are geopolitical factors. The U.S. government's export controls, under an 'America First' policy, prioritize semiconductor supply to the U.S. and its allies. This policy effectively limits allocations to China and concentrates demand within the U.S.-Korea-Taiwan supply chain, further benefiting established players like SK hynix and TSMC.
In essence, Huang's comment isn't just about needing more chips. It's the culmination of a production ramp-up, a persistent supply bottleneck, and a geopolitical strategy that all point to one thing: the critical importance of HBM in the ongoing AI revolution.
- HBM (High-Bandwidth Memory): A type of high-performance computer memory used in conjunction with high-performance GPUs and accelerators. It stacks memory chips vertically to achieve faster data transfer speeds and lower power consumption, making it essential for AI applications.
- Rubin Platform: NVIDIA's next-generation AI platform, succeeding the Blackwell architecture. It features the Vera Rubin GPU and Vera CPU, designed for massive-scale AI training and inference.
- CoWoS (Chip-on-Wafer-on-Substrate): An advanced semiconductor packaging technology developed by TSMC. It allows multiple chips, such as a GPU and HBM, to be integrated onto a single interposer, enabling high-speed communication between them.