Jiangsu Pacific Quartz, a high-purity quartz (HPQ) producer in China, has achieved a significant milestone by having its quartz products pass qualification at a leading domestic DRAM manufacturer, reportedly CXMT, for use in 300-mm wafer production. This development is notable because it establishes a largely domestic supply chain for critical chipmaking components, starting from Pacific Quartz’s purified high-purity sand and processed into semiconductor furnace-tube material destined for a DRAM fab. While these achievements demonstrate China’s growing capability to secure domestic supplies for various semiconductor process equipment consumables, the most critical component - the high-purity fused quartz required for wafer ingots - remains subject to international monopolies. For now, the supply of quartz in significant volumes for wafer ingots is limited to Sibelco and The Quartz Corp., operating in Spruce Pine, North Carolina, alongside Russian Quartz LLC, according to Tom’s Hardware.
What happened
The qualification of process consumables marks a maturing local ecosystem for equipment components. Pacific Quartz has already supplied ultra-high-purity quartz consumables, such as LPCVD diffusion tubes and wafer boats, to major equipment manufacturers like Lam Research and Tokyo Electron since 2019-2020. However, the structural bottleneck remains at the foundational material level. Ingots for semiconductor wafers require high-purity fused quartz, typically exceeding 99.999% SiO₂ (5N+), with extremely tight limits on individual contaminants. This distinction highlights that while regional self-sufficiency is improving for process tools, the ultimate capacity for wafer production is still tethered to the global supply of ultra-high-purity raw materials.
Supply backdrop
The broader semiconductor landscape is currently defined by the intense global demand for advanced memory, where the focus has shifted from raw compute power to data flow. High Bandwidth Memory (HBM) remains critical for AI accelerators, driving massive investment and geopolitical attention. This trend is evident in SK hynix’s plans to establish a U.S. production hub in Indiana, aiming to supply next-generation HBM that has undergone on-site packaging and testing starting in the second half of 2029. Furthermore, the market is seeing architectural shifts, with technologies like High Bandwidth Flash (HBF) emerging. HBF stacks NAND flash memory dies, offering a path to extremely large capacity - potentially 10 to 100 times larger than DRAM - while maintaining high read throughput, making it a compelling alternative for large-scale AI inference systems (Kioxia).
Why it matters
The race for advanced memory is also creating significant supply and pricing pressures. Following US export controls, Chinese AI chip manufacturers have increasingly relied on the gray market to obtain advanced HBM, leading to dramatic price surges. For instance, the quoted price for certain advanced accelerators has seen increases ranging between 20% and 50% in just two months. This cost pressure is directly reflected in the product selling price, underscoring how critical the global supply chain remains, even as regional manufacturing capabilities advance. Meanwhile, major players like OpenAI are deepening cooperation with Samsung, extending beyond simple memory supply into joint R&D and advanced packaging for next-generation chips.
How to read it
Beyond component supply, the industry is grappling with a fundamental architectural shift: the bottleneck is moving from compute to data. As AI workloads evolve toward inference and agentic AI, the core constraint is no longer the GPU’s theoretical compute capability. Instead, performance is determined by where data resides and how quickly it moves through the system. This realization has prompted memory manufacturers to focus on advanced packaging and memory solutions that optimize data retrieval and flow, making memory semiconductors a foundational technology for the AI era. The focus is on minimizing latency and maximizing data throughput to keep compute devices fully utilized.