On November 2, TrendForce reported, citing Taiwan’s Economic Daily News and MoneyDJ, that United Microelectronics Corporation (UMC) has secured major photonic integrated circuit (PIC) orders from Zhongji Innolight and Tencent-backed SILITH, with initial orders potentially totaling up to 10,000 wafers, produced at UMC’s Singapore facility. The products reportedly command prices around seven times those of conventional CMOS processes. UMC declined to comment on specific customers or the reported orders.
Read that paragraph again, because the verbs in it are the story. “Reportedly secured.” “Potentially totaling.” “Declined to comment.” This is not a company announcement, and this article does not treat it as one. It is a market report, attributed to unnamed industry sources through two Taiwanese publications, about an order the foundry in question will neither confirm nor deny. What makes it worth covering anyway is the context around it: a confirmed July production delivery to one of the two named customers, a foundry that told investors six days earlier that silicon photonics demand has risen sharply enough to convert existing capacity, and a supply chain in which the two established merchant foundries are reportedly fully booked. The report is unverified; the context that makes it plausible is on the record.
What was actually reported
Precision first. The TrendForce item, published November 2, 2026, cites Economic Daily News and MoneyDJ and rests on industry sources. Its claims, in order: UMC has reportedly secured PIC orders from Zhongji Innolight, a leading global supplier of data center optical transceiver modules serving major North American cloud service providers, and from SILITH, a silicon photonics PIC design company backed by Tencent’s investment arm. Innolight is expanding its 800G and 1.6T optical module businesses, according to the report; SILITH has cumulative shipments of 100G and 200G per-channel PICs exceeding 8 million units.
The volume: initial orders potentially total up to 10,000 wafers, produced at UMC’s Singapore facility, with additional orders possible. Note the phrasing — “up to” and “potentially.” The figure is a ceiling on an initial combined volume, not a confirmed shipment schedule, and the report gives no timeframe for delivery, no split between the two customers, and no indication of whether the wafers are already in production.
The price: the PIC products reportedly command prices around seven times those of conventional CMOS processes, which the report says could boost UMC’s profitability. No absolute price is given — no dollars per wafer, no revenue figure, no margin figure — and none can be derived from the multiple alone. Any coverage that converts “seven times” into a dollar revenue estimate is inventing the base it multiplies.
The report also says UMC is the first Taiwanese foundry capable of offering both volume production and cost advantages for key optical communication chips — a market characterization from its sources, not a company claim — and that UMC recently reached a milestone in the validation of its 1.6T optical transceiver platform by multiple hyperscalers, with a roadmap targeting 300G and eventually 400G per lane using hybrid bonding and chip-to-wafer integration. No dates are attached to those targets, and no hyperscaler is named.
Finally, the capacity context. Only a limited number of foundries worldwide have mass-production capability for PICs, the report says, naming Tower Semiconductor and GlobalFoundries, whose capacity is reportedly fully booked — the report’s sources speaking, not either foundry. UMC has announced plans to expand silicon photonics production at its P4 facility in Singapore and develop new P7 and P8 facilities at Southern Taiwan Science Park, targeting long-term PIC capacity of 18,000 wafers per month, with equipment installation expected to begin in late 2028 and full capacity targeted for 2030.
That is the full content of the report. Everything else here is confirmed context from UMC’s own disclosures, or analysis.
Why seven times CMOS pricing is the number that matters
The headline figure in most coverage will be 10,000 wafers. The more consequential figure is seven times. A foundry’s economics turn on what a wafer sells for relative to what it costs to make, and a seven-times multiple — if it holds — would make silicon photonics one of the most attractive wafer products in a mature foundry’s portfolio. But the multiple needs to be understood, because it is easy to misread in both directions.
A PIC wafer is not a CMOS wafer with a different label. Silicon photonics is built on silicon-on-insulator (SOI) substrates, which cost substantially more than the bulk silicon used for conventional CMOS logic. The process adds steps a logic flow does not have: deep silicon etches for waveguides and germanium growth for photodetectors. Volumes are far smaller than a mainstream logic process, so fixed development and qualification costs spread across fewer wafers. And the yield arithmetic is harsher: a photonic device fails if any of its optical channels fails, so a wafer acceptable for a forgiving logic product can be uneconomic for a multi-channel PIC.
All of that explains why a PIC wafer should command a premium. It does not explain the size of the premium, and the report does not try to. Seven times is a relative figure from unnamed sources, with no baseline specified — seven times what CMOS process, at what node, at what volume? Without that baseline, the multiple is a directional signal, not a price. It says these wafers are expensive relative to UMC’s mainstream business. It does not say what UMC will earn from them.
There is also a timing question. Premium pricing in a capacity-constrained market is a condition of scarcity, and scarcity pricing erodes as capacity comes online — UMC’s own plan puts large-scale capacity in 2030. The premium, if real, is a near-term phenomenon, not a permanent property of PIC wafers.
The July delivery is the part that is on the record
The reason this report deserves to be taken seriously, despite being unconfirmed, is that half of it is anchored in a confirmed event from four months ago. On July 14, 2026, UMC and SILITH announced the first mass-production wafer delivery of photonic ICs from UMC’s Singapore 12-inch fab, supporting SILITH’s 1.6T silicon photonics platform. The announcement is a joint company release, not a market report. The two companies said the platform moved from development to production readiness in 18 months, has demonstrated production-level performance with high yield and reliability, and has since been qualified by a leading cloud infrastructure customer for volume deployment. The customer is not named. UMC also said it would make its own 12-inch silicon photonics platform available for customer product development in 2027.
That July announcement does three things for the November report. First, UMC’s Singapore fab was already producing PIC wafers at volume for SILITH — one of the two named customers — before the reported orders existed, so the report claims an expansion of an existing relationship, not a new one. Second, the 1.6T platform is a real product with a cloud qualification behind it, the same platform the report describes as validated by multiple hyperscalers. Third, it gives UMC a credible basis for the “volume production” half of the report’s “first Taiwanese foundry” characterization: a 12-inch line that has already delivered mass-production wafers.
But the July announcement also sets a limit. It confirms a production relationship with SILITH. It says nothing about Innolight as a UMC customer, nothing about a 10,000-wafer order, and nothing about pricing. The Innolight half of the November report has no confirmed anchor of this kind. Innolight is a confirmed Tower customer — its NPO device entered early production on Tower’s PH18M process on November 3 — but a Tower customer is not automatically a UMC customer, and no company statement connects Innolight to UMC’s Singapore fab. Treat the two halves of the report differently: the SILITH half extends a confirmed relationship, and the Innolight half is a new claim resting entirely on unnamed sources. One naming detail: the company is SILITH Technology, a Singapore-headquartered silicon photonics fabless company, per the July release; some coverage renders it “Silith,” but the TrendForce report and the UMC release both use SILITH, and this article follows them.
The capacity plan, and what UMC itself has said
The most solid part of the story is what UMC has said in its own voice, to investors, six days before the report appeared. On October 28, 2026, UMC reported third-quarter results: consolidated revenue of NT$74.24 billion, up 8.0% sequentially and 25.6% year over year; gross margin of 34.0%; capacity utilization of 92%, up from 85% in the second quarter, with several 12-inch fabs operating at full capacity; and wafer shipments of more than 1.2 million 12-inch-equivalent wafers, a record high, up 7.3% sequentially. On the same day, the board approved approximately US$3.2 billion in additional capital expenditure to accelerate expansion in Singapore and Tainan, covering silicon photonics and advanced packaging.
On the earnings call, management said demand for silicon photonics and advanced packaging has increased significantly and that the company is converting some existing capacity to address near-term unfulfilled demand. Silicon photonics is expected to ramp gradually in 2027, and advanced packaging to widen its revenue contribution in 2028. The long-term PIC capacity target of about 18,000 wafers per month, mainly in Singapore, matches the figure in the November report — but note the sourcing: the target comes from UMC’s own earnings disclosures, while the equipment-installation and 2030 dates come from the market report. Both are consistent, but they are not the same kind of evidence.
The Singapore footprint is worth a sentence of confirmed context. UMC’s new US$5 billion Singapore fab started commercial production earlier in 2026, and the expansion plan is expected to take the site’s annual output to about 1.3 million wafers in the next few years. The July PIC delivery came from that 12-inch Singapore operation. None of this confirms the order. What it confirms is the demand environment: a foundry at 92% utilization, converting capacity for silicon photonics, spending US$3.2 billion to expand, and already producing PIC wafers for one of the two named customers. A fabricated order would be a strange thing to invent against that background; a real order would be the least surprising thing in it. That is the honest calibration: plausible, unconfirmed, and consistent with the confirmed record.
The second foundry name, and the Innolight thread
Step back from UMC, because the report’s significance is structural, not just commercial. Until this month, the silicon photonics foundry story had two merchant names that mattered for AI optics: Tower Semiconductor, with its PH18 platform, and GlobalFoundries. Both are, according to the report’s sources, fully booked. A third foundry with volume PIC capability adds a source for the most constrained input in the AI optical chain.
That is why the Tower parallel matters. On November 3, the day after the UMC report, Tower and Innolight announced that Innolight’s first near-package optics device had entered early production on Tower’s PH18M process in Tower’s California factory. Innolight is now, on the public record, a Tower manufacturing customer across four Tower facilities, and, on the reported record, a UMC customer in Singapore. If both are true, Innolight is deliberately dual-sourcing its photonics manufacturing across two foundries on two continents. But the UMC half of that story is unconfirmed, and the dual-sourcing inference collapses if it is wrong. Do not present Innolight as dual-sourced until UMC or Innolight says so.
The two announcements also establish different things. Tower–Innolight is a company announcement: named process, named factory, named production status. UMC–Innolight is a market report: named customers, named volume, unnamed sources, company declined to comment. The Tower story’s uncertainty is about scale. The UMC story’s uncertainty is about existence. Those are different kinds of uncertainty, and they should not be averaged into a single impression of “the week’s foundry news.” The same day also brought the proposed Marvell–Lumentum combination, which would consolidate the laser supply chain while the foundry stories expand the silicon photonics supply chain.
What is undisclosed, and why it matters
An honest accounting of what is not on the record is part of the story. No order has been confirmed by UMC, Innolight, or SILITH. No contract value, delivery schedule, or wafer split has been disclosed. No wafer price in dollars has been disclosed, and no revenue or margin impact can be calculated from the reported multiple. No product is named — no PIC part number, no module, no lane rate, no reach, no power figure. No qualification status is stated for the reported Innolight wafers: the July qualification belongs to SILITH’s 1.6T platform, and it cannot be transferred to an Innolight order by association.
Each omission has a plausible innocent explanation: foundries do not comment on specific customers as a matter of policy, and order volumes and prices are commercially sensitive. But the pattern of omissions also means the claims that matter most — is there an order, of what size, at what price, for delivery when — are entirely untested in public.
What to watch
The signals that will move this story from report to evidence are specific. First, a confirmation: any statement from UMC, Innolight, or SILITH acknowledging the order, the volume, or the relationship. Second, a customer disclosure: Innolight naming UMC as a foundry supplier, or UMC naming Innolight in an earnings call or investor presentation. Third, a capacity marker: disclosure of PIC wafer starts, ramp timing, or Singapore fab utilization attributable to the reported orders. Fourth, the 2027 platform opening: whether UMC’s own 12-inch silicon photonics platform attracts named customers beyond SILITH. Fifth, the capacity race: whether Tower or GlobalFoundries announces PIC capacity additions of its own.
The fair summary is a narrow one. On November 2, TrendForce reported, citing Economic Daily News and MoneyDJ, that UMC has won PIC orders from Innolight and SILITH totaling up to 10,000 wafers at roughly seven times conventional CMOS pricing, produced in Singapore. UMC declined to comment. What is confirmed is adjacent and substantial: UMC delivered its first mass-production PIC wafers to SILITH from Singapore in July, qualified by a leading cloud customer, and reported 92% utilization with US$3.2 billion in additional capital spending approved on October 28, with silicon photonics expected to ramp in 2027 and long-term PIC capacity targeted at 18,000 wafers per month by 2030. The reported order is plausible, consistent with the confirmed record, and unconfirmed. That is the whole story, and it is enough.
Sources and further reading
TrendForce, “UMC Reportedly Wins Photonic Integrated Circuit Orders; Initial Orders Up to 10K Wafers at 7× CMOS Pricing,” November 2, 2026 — citing Economic Daily News and MoneyDJ; reported PIC orders from Zhongji Innolight and Tencent-backed SILITH, initial orders potentially totaling up to 10,000 wafers, production at UMC’s Singapore facility, prices around seven times conventional CMOS processes, UMC declined to comment; SILITH cumulative shipments of 100G and 200G per-channel PICs exceeding 8 million units; 1.6T platform validated by multiple hyperscalers, roadmap targeting 300G and eventually 400G per lane with hybrid bonding and chip-to-wafer integration; Tower Semiconductor and GlobalFoundries capacity reportedly fully booked; P4 Singapore expansion and P7/P8 facilities at Southern Taiwan Science Park; long-term PIC capacity target of 18,000 wafers per month, equipment installation from late 2028, full capacity by 2030. All order, volume, customer, and pricing claims in this article drawn from this report are presented as reported, not confirmed.
UMC and SILITH Technology, “UMC Delivers First Mass-Produced Silicon Photonics Wafers from Singapore 12-Inch Fab,” July 14, 2026 — first mass-production wafer delivery of photonic ICs from UMC’s Singapore fab supporting SILITH’s 1.6T silicon photonics platform; platform brought from development to production readiness in 18 months; demonstrated production-level performance with high yield and reliability; qualified by a leading cloud infrastructure customer for volume deployment; UMC’s own 12-inch silicon photonics platform to be available for customer product development in 2027.
UMC, third-quarter 2026 results and earnings call, October 28, 2026 — revenue NT$74.24 billion, up 8.0% sequentially and 25.6% year over year; gross margin 34.0%; utilization 92%, up from 85% in the second quarter, with several 12-inch fabs at full capacity; wafer shipments of more than 1.2 million 12-inch-equivalent wafers, a record high; board approval of approximately US$3.2 billion in additional capital expenditure; demand for silicon photonics and advanced packaging increased significantly, with existing capacity being converted to address near-term unfulfilled demand; silicon photonics expected to ramp gradually in 2027; long-term PIC capacity target of about 18,000 wafers per month, mainly in Singapore.
Singapore Economic Development Board, September 2, 2026 — UMC’s new US$5 billion Singapore fab began commercial production earlier in 2026; expansion expected to take the site’s annual wafer output to about 1.3 million in the next few years. This publication’s account of the Tower–Innolight production milestone (November 3, 2026) and the Marvell–Lumentum agreement (November 2, 2026) is drawn from its own prior coverage. No order, volume, customer, or wafer price is asserted here beyond what the November 2 report states, because none is confirmed by UMC, Innolight, or SILITH. This article is analysis, not investment advice.