Simmtech Invests $295 Million to Build AI Server Substrate Plant as Nvidia Vera Rubin Drives SOCAMM Demand

Reviewed by AtlasPCB Engineering Team

Simmtech Commits 400 Billion Won to AI Server Substrate Expansion in Cheongju

Simmtech, one of South Korea’s leading publicly traded PCB substrate manufacturers, signed a memorandum of understanding with North Chungcheong Province and Cheongju City on September 2, 2026, committing 400 billion won (approximately $295 million) to expand AI server substrate production capacity. The phased investment will fund both a new manufacturing site and upgrades to the company’s existing Cheongju plant, positioning Simmtech to meet surging demand for advanced substrates driven by next-generation AI accelerator platforms.

The new facility will occupy a 16,529-square-meter site in Oksan-myeon, Heungdeok-gu, Cheongju, with 11,349 square meters of floor area for production and offices. The expansion targets three product categories at the intersection of advanced packaging and high-density PCB manufacturing: SOCAMM (Small Outline Compression Attached Memory Module) substrates, MCP (multichip package) substrates for LPDDR5X memory, and high-layer-count mSAP (modified semi-additive process) substrates for system IC applications.

What Is SOCAMM and Why Does It Matter?

SOCAMM represents a new memory module form factor designed for the extreme bandwidth requirements of AI training and inference. Unlike conventional SO-DIMM modules that use pin-based connectors, SOCAMM employs a compression-attached interface that reduces signal path length and enables significantly higher data rates. Each SOCAMM module integrates four LPDDR5X memory packages into a compact footprint, delivering the bandwidth next-generation AI accelerators demand while occupying less space on the server board.

The immediate driver behind Simmtech’s investment is Nvidia’s Vera Rubin AI accelerator platform, which pairs the Vera CPU with next-generation GPU silicon. Vera Rubin is expected to adopt SOCAMM as its primary memory module interface, creating a new substrate demand category that existing production lines cannot serve. The substrates inside SOCAMM modules must support high layer counts, fine line and space geometries, and tight impedance control across densely packed signal channels — requirements that push beyond what conventional PCB fabrication can deliver reliably at volume.

From a PCB manufacturing perspective, SOCAMM substrates are technically demanding. The compression-attached interface requires precise pad geometries and surface planarity beyond standard connector-based modules. Routing four LPDDR5X packages within the module’s compact outline demands high-density interconnect techniques with microvia stacking and sequential lamination. The combination of high layer count, fine features, and tight electrical specifications places SOCAMM substrates firmly in the advanced substrate category.

Investment Breakdown: Where the Money Goes

Simmtech’s board of directors approved 274.2 billion won in capital expenditure on August 25, with deployment through December 2027. The approved amount breaks down into three categories that reveal the company’s strategic priorities.

The largest allocation, 145.9 billion won, targets SOCAMM module PCB production capacity. This investment will increase Simmtech’s annual SOCAMM substrate output from 300 billion won to 500 billion won in revenue capacity, a 67 percent increase that reflects the company’s confidence in sustained demand from the Nvidia Vera Rubin supply chain and potentially other AI platform adopters.

The second major allocation of 105.1 billion won funds high-layer-count mSAP substrate production. The modified semi-additive process has been gaining traction beyond its traditional smartphone application, and Simmtech’s investment confirms that mSAP is becoming essential for AI server substrates. Unlike conventional subtractive etching, which is limited to approximately 50-micrometer line and space due to copper undercut, mSAP builds traces by plating copper onto an ultra-thin seed layer through patterned photoresist. The thin seed layer — typically two to three micrometers — produces minimal lateral undercut when removed, enabling line and space below 30 micrometers. For AI server substrates routing thousands of high-speed signals across many layers, mSAP’s density advantage is not optional but necessary.

The remaining 23.2 billion won covers semiconductor substrates for system ICs, a category that includes the package substrates used beneath processor and controller dies in advanced electronic systems.

Together, these investments will add approximately 250 billion won per year in MCP and semiconductor substrate capacity on top of the SOCAMM expansion, bringing the total annual capacity increase to roughly 450 billion won. New production lines begin operations sequentially from 2028, and the expansion will create approximately 100 new jobs.

The mSAP Imperative for AI Substrates

Simmtech’s heavy investment in mSAP capacity signals a broader industry shift. For years, mSAP production was concentrated almost exclusively in smartphone mainboard fabrication, where Apple, Samsung, and a handful of other OEMs justified the process’s cost premium through extreme density requirements. The expansion of mSAP into AI server substrates represents a fundamental change in the addressable market.

The technical rationale is straightforward. AI accelerator packages generate enormous pin counts that must be routed through substrates and module PCBs to reach memory, power, and I/O interfaces. As die sizes grow and interconnect density increases, conventional subtractive processes run out of routing capacity. High-layer-count mSAP substrates provide two to three times the wiring density per layer compared to subtractive methods, enabling more routing capacity within the same layer count or fewer layers for a given routing requirement.

For PCB manufacturers and designers working on AI hardware, the implications are clear. The substrate requirements for platforms like Nvidia Vera Rubin are pulling mSAP out of its smartphone niche and into mainstream advanced packaging applications. Companies with mSAP process capability — or access to it through qualified manufacturing partners — will be positioned to serve this growing segment. Those relying solely on conventional subtractive processes will find themselves unable to meet the line and space specifications that next-generation AI substrates demand.

What This Signals for the Advanced Substrate Market

Simmtech CEO Kim Young-koo emphasized the strategic significance of the Cheongju expansion, framing it as a response to structural demand growth in the AI accelerator ecosystem. The investment timeline — board approval in August 2026, spending through December 2027, production from 2028 — aligns with the expected ramp of Nvidia’s Vera Rubin platform and suggests Simmtech has secured supply agreements or strong demand signals from downstream customers.

The broader context reinforces this reading. Multiple substrate and PCB manufacturers across Asia have announced capacity expansions targeting AI server applications throughout 2025 and 2026. AI hardware’s appetite for advanced substrates — high layer counts, fine features, controlled dielectric properties — is growing faster than existing capacity can serve. Simmtech’s commitment adds to a wave of investment pointing toward a structural shortage of advanced substrate capacity in the 2027 to 2029 timeframe if demand continues on its current trajectory.

For hardware engineers and procurement teams designing AI server or high-performance computing systems, the practical takeaway is that advanced substrate lead times will likely remain extended through the end of this decade. Engaging PCB manufacturing partners early in the design cycle, locking in capacity before production ramp, and designing with manufacturing constraints in mind will be critical to avoiding bottlenecks as the industry transitions to next-generation AI architectures.

Source: The Elec, September 2, 2026