TSMC denies exploring 3D DRAM at Nvidia's request

TSMC has denied a report claiming it is developing 3D DRAM manufacturing at Nvidia's request, stating it remains a pure-play logic foundry with no plans to enter commodity memory production. The denial pushes back against speculation that AI-driven memory shortages could reshape the semiconductor supply chain by pulling the world's largest contract chipmaker into the DRAM business.

"Please note that TSMC focuses on our core business at all times, which is dedicated IC foundry," a TSMC spokesperson told Tom's Hardware. "Our mission is to be the trusted technology and capacity provider of the global logic IC industry for years to come."

The original report from The Information, citing two people familiar with the initiative, said TSMC had begun preparing plans for 3D DRAM manufacturing following a request from Nvidia. The effort was reportedly overseen by Hsu Kuo-chin, TSMC's vice president responsible for advanced packaging and former head of Micron's Taiwanese operations. The backdrop is a memory market where shortages have turned DRAM from a commodity into a strategic asset that determines whether companies can ship products to customers.

The technology gap TSMC would need to close

TSMC lacks both the expertise and production capacity to manufacture DRAM, let alone the more complex 3D variant. Unlike conventional DRAM, which arranges memory cells in a single plane, 3D DRAM stacks multiple layers of memory cells vertically to increase capacity per wafer. All three leading DRAM makers - Micron, Samsung, and SK hynix - are working on 3D DRAM projects, but the technology remains years from commercial viability.

Micron's roadmap alone includes at least four more generations of planar DRAM process technology before 3D DRAM becomes a reality. Developing a 3D DRAM manufacturing process without first having a competitive planar DRAM process is widely considered impractical. TSMC has no such foundation. DRAM fabrication requires entirely different manufacturing processes from logic chip production, and building that capability from scratch would demand billions of dollars and years of R&D.

One theoretical path would be acquiring a small specialty memory producer with existing fabrication facilities. Even then, smaller manufacturers operate on older nodes, meaning TSMC would still need to invest heavily to reach leading-edge commodity DRAM performance.

Strategic risks beyond the technical challenge

Entering DRAM manufacturing would expose TSMC to the memory market's notorious boom-and-bust cycles. Shortages encourage capacity expansions that eventually create oversupply, depressing prices and profitability. That cyclical risk alone gives a company accustomed to stable foundry margins reason for caution.

The move could also strain TSMC's relationship with existing DRAM partners. SK Hynix partnered with TSMC in 2024 to manufacture logic base dies for next-generation HBM products. Becoming a competitor in memory could complicate that collaboration. Antitrust questions would likely surface as well, given TSMC's dominant position in leading-edge logic manufacturing and advanced chip packaging. Regulators might examine whether vertical integration into DRAM would let TSMC favor its own memory products or disadvantage competitors without similar integration.

What TSMC can do without building DRAM fabs

TSMC does not need to manufacture DRAM to capture more value from the advanced memory market. The company already produces sophisticated logic base dies for HBM, integrates memory with processors using its SoIC hybrid bonding technology, and develops customized memory solutions through partners like GUC and established DRAM manufacturers. This approach expands TSMC's presence in advanced memory without requiring tens of billions of dollars in new fabs or a ground-up DRAM process development effort.

The denial leaves TSMC's strategy clear: deepen its role in advanced packaging and logic-memory integration while leaving commodity DRAM manufacturing to the three companies that already dominate it.

Why this matters for product development and operations teams

For teams managing AI hardware roadmaps, TSMC's denial signals that the DRAM supply chain will remain concentrated among the three incumbent manufacturers for at least the next five years. Memory availability and pricing will continue to depend on Micron, Samsung, and SK hynix capacity decisions. The speculation itself, however, reflects real pressure in the supply chain - memory has become a bottleneck that can delay product shipments. Operations and procurement teams should track HBM allocation strategies closely, as logic-memory integration through advanced packaging will be the near-term lever for improving AI system performance, not a new entrant into DRAM manufacturing.