Si-Cap TAM Expansion: SEMCO’s W1.5 Trillion Contract Signals Broader HPC Opportunity – Taiwan DRAM Suppliers Stand to Benefit

Core Thesis

SEMCO’s recent announcement of a ∼W1.5 trillion (∼US$1bn) silicon capacitor supply contract, covering 2027–2028 for next-generation TPU EMIB-T, marks an inflection point. This is not an isolated AI-server procurement; it signals a secular expansion of the Si-Cap total addressable market (TAM) into autonomous driving, mobile, and other high-performance computing fields. Given the supply oligopoly, customers are increasingly locking in capacity via long-term agreements, which raises revenue visibility for suppliers. Legacy DRAM process technology is the ideal fabrication platform for Si-Cap, creating a structural opportunity for Taiwan’s DRAM manufacturers—most immediately Winbond (2344.TW)—to capture low-single-digit revenue mix contributions from potential partnerships with SEMCO.

What the Market Is Underappreciating

Three dimensions of the Si-Cap opportunity are not yet priced in:

1. TAM expansion velocity – The market focuses on AI server demand, but SEMCO explicitly plans to diversify into autonomous driving and mobile. AP Memory’s earnings call confirmed that mobile customers may follow. The TAM for Si-Cap could double or triple over the next two to three years as HPC applications proliferate.

2. Long-term contract model – The two-year duration of SEMCO’s contract is the first concrete evidence that customers pre-commit capacity in an oligopolistic supply environment. This structural shift reduces order volatility and enables suppliers to plan capex more efficiently, compressing the risk premium embedded in current valuations for exposed names.

3. Taiwan DRAM suppliers’ inverse leverage – PSMC (6770.TW) is capacity-bound; its fab is full, leaving no room for Si-Cap. Winbond, with available legacy DRAM capacity and process compatibility, is the most probable SEMCO partner. Even a low-single-digit revenue contribution in 2027e (∼1-3% of total revenue) would be additive to growth and could trigger valuation re-rating as investors assign a higher multiple to contracted revenue streams.

Evidence Chain

• SEMCO contract details: Announced May 2026, value ∼W1.5 trillion (∼US$1bn), term Jan 2027–Dec 2028. Primary application is EMIB-T for next-gen TPUs, but the company explicitly stated plans to expand into autonomous driving and mobile devices.
• Si-Cap supply oligopoly: Only a few players (SEMCO, AP Memory, others) have credible manufacturing capability. The contract’s two-year term confirms customers’ preference for locking supply, a pattern AP Memory has echoed in its calls.
• DRAM legacy process fit: The report directly states “the legacy DRAM process is ideal for Si-Cap fabrication.” This is because Si-Cap requires high-density trench capacitors built on mature DRAM nodes. Taiwan DRAM houses (Winbond, Nanya, PSMC) possess this exact capability.
• Capacity constraint at PSMC: PSMC is running at full utilization; no spare fab capacity for Si-Cap. This naturally funnels potential SEMCO partnership to Winbond, which has more flexibility and is already exploring such opportunities.
• Winbond revenue impact: If Winbond becomes SEMCO’s partner, the 2027e revenue contribution is estimated at low single-digit percent of total revenue. While not transformative on its own, the contract nature (multi-year, fixed) provides a base for margin stability and capex efficiency.

Key Risks

• Contract execution risk: SEMCO’s contract may be delayed, downsized, or subject to technology migration (e.g., to a different capacitor architecture) that reduces actual supply volume below expectations.
• Competitive pressure: New entrants (e.g., pure-play Si-Cap foundries or alternative technologies like deep-trench capacitors from other players) could erode pricing and margins in the oligopoly, diminishing the value of existing contracts.
• Demand shortfall: AI capex cycles are inherently volatile. A slowdown in AI server deployment, or weaker-than-expected adoption of Si-Cap in mobile/automotive, would compress TAM expansion and reduce the number of follow-on contracts.
• Technology readiness: Winbond (or another Taiwan DRAM supplier) must prove that its legacy DRAM process can meet SEMCO’s quality and cost requirements at scale. Yield ramps are rarely smooth in new capacitor integration, posing a risk of delayed revenue recognition.

Valuation & Trading Implications

• Winbond (2344.TW): The primary beneficiary among Taiwan DRAM suppliers. Current Equal-weight rating and price of NT$115.5 do not embed any Si-Cap contribution. If a partnership is confirmed, even a ∼2% revenue contribution could lift EPS by ∼3-5% given high incremental margins from contracted fab capacity. The stock would likely see multiple expansion as recurring revenue is valued higher. We see 15-25% upside scenario under a confirmed contract.
• AP Memory (6531.TW): Already rated Overweight (NT$973). As a key Si-Cap player, AP Memory benefits from overall TAM growth and industry hearing of long-term contracts. Its earnings call commentary on mobile adoption reinforces the theme. AP Memory serves as a direct proxy for Si-Cap expansion.
• PSMC (6770.TW): Overweight rated (NT$57.8), but limited near-term Si-Cap upside due to full capacity. The company remains a beneficiary only if it adds capacity or if demand shifts, but that is not the base case. We see no material increment from this specific contract.

Bottom line: The SEMCO contract is a catalyst that should force investors to re-evaluate the Si-Cap opportunity beyond AI servers. Focus on Winbond as the most asymmetric risk/reward play, while AP Memory offers pure-play exposure. Monitor SEMCO’s production ramp and any announcements of additional customer contracts for signal validation.

Appendix: Key Data Points