United Microelectronics Corporation, Taiwan’s second-largest contract chipmaker, began mass production of silicon photonics wafers at its Fab 12i facility in Singapore on Tuesday, delivering the first commercially produced wafers of this type from the facility to SILITH Technology, a Singaporean fabless chip design company developing 1.6T optical interconnect platforms for AI and hyperscale data center networks. The transition from development to mass production readiness took eighteen months from initiation of the collaboration. UMC has separately announced that it will make its own 12-inch silicon photonics platform available for broader customer product development by 2027, creating a commercialization pathway that extends beyond the initial SILITH partnership. UMC shares rose approximately 2.3% in pre-market trading on the announcement. NEWSCENTRAL notes that the scale of this milestone extends well beyond UMC’s own commercial trajectory: silicon photonics is the enabling technology for the optical interconnects that allow AI chips to communicate with each other at the speeds that large-scale inference and training systems require, and mass production at a 12-inch wafer scale is the inflection point at which the economics become viable for broad deployment in AI infrastructure.
The technical distinction between 8-inch and 12-inch wafer production is commercially significant and explains why the Singapore facility milestone matters specifically. Larger wafer diameter means more chips produced per production run, which directly reduces per-unit cost and enables the volume commitments that major AI infrastructure buyers require before building optical interconnects into their standard server rack designs. UMC licensed imec’s iSiPP300 silicon photonics process platform in December 2025, giving it access to a proven technology stack rather than requiring the years of independent process development that building from scratch would have demanded. The 18-month path from that license to mass production is notably fast for a semiconductor manufacturing transition, reflecting both the quality of the licensed process and the urgency of the demand.
The demand environment that UMC is positioning to serve is one of exceptional and growing urgency within the AI infrastructure buildout. The transition from electrical to optical data transmission inside data centers has been a long-standing engineering objective, but it became commercially critical as AI workloads began generating data traffic volumes that copper interconnects cannot sustain at the speeds and power efficiencies that modern AI rack designs require. Silicon photonics-based transceivers operating at 1.6T – 1.6 terabits per second – are the current target specification for next-generation AI data center interconnects, and the supply chain for those components has been constrained by the limited number of chipmakers capable of producing silicon photonics wafers at commercial scale. Tower Semiconductor, which announced a $3 billion Japan expansion for silicon photonics production on the same day, is UMC’s primary competitor in this emerging segment. Lucas Grant, Semiconductor and Manufacturing Strategy Analyst at NEWSCENTRAL, observes that the simultaneous announcements by UMC and Tower Semiconductor on the same day reflect the intensity of the race to build silicon photonics production capacity ahead of what both companies are clearly projecting to be a sustained and large-scale demand cycle from AI infrastructure builders.
The improving outlook that market analysts have projected for UMC in the second half of 2026 is grounded in a commercial recovery that the silicon photonics milestone accelerates rather than creates. The company’s June revenue grew approximately 23% year-on-year to approximately NT$23.1 billion, with the year-to-date revenue up 11% compared with the same period in 2025. Analysts forecast 13% quarter-on-quarter revenue growth in Q3 2026, and UMC’s own capital expenditure guidance for 2026 has been set to support the expansion required to meet silicon photonics demand alongside its existing specialty node business. The July 29 second-quarter earnings report will be the next formal milestone for assessing whether the trajectory is tracking against those forecasts.
Singapore’s role as the geography for UMC’s silicon photonics production is itself commercially significant. The city-state has invested substantially in semiconductor manufacturing infrastructure through the Economic Development Board and maintained regulatory and business conditions that make it an attractive alternative to Taiwan for companies seeking to diversify their manufacturing footprint against geopolitical risk. For AI infrastructure buyers who are simultaneously seeking supply chain diversification away from Taiwan-concentrated manufacturing, UMC’s Singapore facility offers silicon photonics production capacity that is both geographically diversified and technically validated by a company with a long commercial relationship with major hyperscalers.
The competitive timing is also commercially relevant. UMC’s Singapore production and Tower Semiconductor’s Japan expansion were announced on the same day, reflecting a simultaneous recognition by two different companies that silicon photonics has crossed the commercialization threshold and that the production ramp window is now open. NEWSCENTRAL considers this coordinated-but-independent timing the strongest available market signal that the silicon photonics supply constraint will ease materially over the next 18 months, which has implications for the pricing power that current producers hold and for the cost trajectory of optical interconnects in AI infrastructure.
NEWS CENTRAL assesses the UMC Singapore milestone as a significant supply-side event in the AI infrastructure market rather than merely a corporate operational achievement. The number of companies capable of producing silicon photonics wafers at 12-inch scale and commercial volume has increased by one this week – and that increase matters because the constraint on optical interconnect deployment in AI data centers has been the ability to manufacture the underlying components at volume, not the demand for them. Every new production source that comes online expands the supply available to AI infrastructure builders and reduces the premium pricing power that current producers hold. That is good for hyperscalers building out compute capacity and good for the long-term economics of the AI infrastructure buildout.