Chinese display and cover-glass companies are pushing into AI semiconductor packaging substrates. On July 24, 2026, Intel and Lens Technology announced a strategic collaboration on advanced packaging using glass substrates. In the first half of the same year, BOE also ran fully automated equipment on a panel-level glass core substrate pilot line designed for 1,000 panels per month.
However, neither company has begun mass production. Lens's disclosure is a one-year memorandum that leaves process verification, qualification, and mass production to separate agreements. BOE's figure of 1,000 panels per month is likewise a pilot-line design value, and the company explicitly states that mass-production yield and mass-production revenue have not yet been achieved. TSMC's CoPoS, under development in Taiwan, is also a panel-level packaging approach that may use glass carriers for process purposes—not necessarily the same thing as a glass core substrate that remains in the final product.
If all of these moves are lumped together as a single "battle for glass substrates," both the products and the timelines get misread. What has actually begun is not so much a competition over glass materials themselves, but a competition over manufacturing processes: drilling fine holes, filling them with copper, forming multilayer wiring, and suppressing warpage and misalignment across large panel areas.
What Lens's Memorandum Actually Defines Is the Scope for Exploring TGV Process Collaboration
Intel's announcement states that it combines Lens's precision glass processing, high-precision lasers, and large-scale production capacity with Intel's semiconductor design and advanced packaging technology. The target applications are AI, data centers, and application-specific computing, with the goal of developing glass substrate packaging that increases connection density and improves power efficiency.
Reading further into Lens's own corporate disclosure reveals a finer level of detail regarding the collaboration. The focus area is TGV (Through-Glass Via), and the two companies will jointly examine and evaluate drilling holes in glass substrates, high-precision laser processing, forming metallized through-holes, and multilayer interconnects. Intel will provide explanatory architecture information, Design for Manufacturability (DFM) guidelines, benchmarks, and verification methods.
This is not a statement that Intel has placed a mass-production order with Lens. Process verification, product qualification, and mass-production activities require separate written contracts, and the current memorandum—aside from matters such as confidentiality and intellectual property—does not commit either party to any transaction or commercial outcome. Its term is also just one year from signing.
Lens already operates TGV-related pilot lines and has shipped samples to some prospective customers. It states that some of these have passed initial proof-of-concept stages and moved on to technical trials. Even so, the TGV business currently has no material impact on the company's financial results. The timing, capacity, and customers for mass production have not been disclosed.
Expectations on the technical side are high. In 2023, Intel indicated that its own glass substrates could reduce pattern distortion by 50% compared with organic substrates and increase connection density by up to 10 times. Following more than a decade of research, the company has set a goal of bringing this to market in the late 2020s, with an ambition to fit one trillion transistors into a single package by 2030. That said, these figures represent Intel's own technology roadmap, not performance measured on joint samples with Lens.
BOE's 1,000 Panels per Month Is a Pilot-Line Design Value, Not "Mass-Production Capacity"
BOE's progress is easier to trace through its history of capital investment compared with Lens. The company began technical research on glass core substrates in 2020, and in 2022 invested RMB 390 million to build a wafer-level experimental platform compatible with both glass and silicon. In 2024, it invested RMB 993 million to scale up to a panel-level glass core substrate pilot line. Major equipment installation and adjustment were completed in 2025, and fully automated equipment was brought online in the first half of 2026.
The design capacity of this pilot line is 1,000 panels per month. BOE has run the entire process—from forming fine TGV holes and copper filling through build-up and multilayer wiring—and in 2025 developed large-format samples with a 9-2-9 configuration totaling 20 layers, which it shipped to customers. The target application is Glass Core Substrates for large-scale computing chips, and some domestic Chinese customers have completed proof-of-concept trials and moved on to technical trials.
There remains, however, a considerable gap between bringing equipment online and achieving mass production. According to the explanation BOE disclosed alongside its collaboration memorandum with Corning, the pilot line's yield has not yet reached mass-production levels, and neither continuous production nor mass-production revenue has begun. The memorandum with Corning is also a three-year statement of intent to collaborate, with investment for individual projects, commercial terms, and formal contracts to be negotiated separately.
Therefore, the figure of 1,000 panels per month cannot be treated as current shippable good-unit capacity. This number merely indicates the input scale the equipment is designed to handle. To gauge actual supply capability, one would need utilization rates, crack rates after TGV formation, yield rates for copper filling and wiring, warpage figures, reliability test results, and the number of panels that have passed customer qualification—none of which has been disclosed.
The strength BOE brings is its accumulated expertise in large-area glass processing, thin-film formation, and integrated manufacturing from the display industry. The company itself positions the characteristics of glass-based packaging as "large format, low warpage, high integration." While display panel equipment cannot simply be repurposed as-is, the experience of handling large glass sheets and forming uniform films and fine patterns provides a starting point for panel-level packaging.
Glass Cores, Glass Carriers, and CoPoS Play Different Roles
Glass plays at least two distinct roles in advanced packaging. One is as the glass core that remains in the final product—replacing the organic core material of a packaging substrate with glass and routing electrical signals and power vertically through TGVs. BOE explicitly identifies its target product as a Glass Core Substrate. Lens's disclosure, by contrast, specifies only the glass substrate process using TGV, and does not state that this remains as a glass core in the final product.
The other role is as a glass carrier, which supports thin wafers or reconstituted panels during manufacturing. Corning describes this as a support structure used for wafer thinning, fan-out, and temporary bonding in 2.5D/3D packaging, stating it can accommodate square panel formats such as 515×510mm and 600×600mm. It notes that transparency enables laser debonding and alignment, and claims it can reduce in-process warpage by up to 40%. However, this is Corning's own product description, and carriers are typically removed after processing.
TSMC's CoPoS is a panel-level packaging architecture that integrates chips on rectangular panels rather than circular wafers. Even if glass carriers are used in the process, that alone does not mean the final substrate becomes a glass core.
TSMC's own official statements also need to be read with these two timelines kept separate. In April 2026, CEO C.C. Wei explained that a CoPoS pilot line was under construction and that production was still several years away, with large CoWoS remaining the primary supply at that point. At the July earnings call, a questioner lumped together glass core, glass substrate, and glass carrier when asking about progress. Wei responded that the majority of current supply is still CoWoS, that the company is developing alternative technologies to reduce costs along with processes involving substrate vendors, and that maturation would take roughly another year.
From this response, one cannot conclude that a specific glass core substrate will enter mass production in one year. This is because the question itself bundled together three different terms, and the answer did not specify any particular structure or product. What can be confirmed is only that TSMC is developing its CoPoS pilot line in parallel with a range of substrate and process options that include glass.
What Will Decide the Race Is Integration Yield at Large Panel Sizes
TrendForce projects that TSMC's CoPoS will standardize on 310×310mm panels, with 2026 as a period for equipment and material verification, 2027 for trial production, and mass production coming in late 2028. It further indicates that commercial-scale production of glass core substrates will not arrive until 2030 or later. This is not a forecast that CoPoS and glass core substrates will ramp up at the same time.
Taiwanese display companies have already stepped into adjacent processes. In its first-half 2025 investor materials, Innolux presented a glass-based FOPLP solution for AI/HPC applications that includes TGV, describing supply formats of 95×240mm strips or 297×240mm sub-panels. This demonstrates capability in handling large-area glass, RDL-first, and chip-first approaches together with TGV within a single manufacturing system. It is not, however, a disclosure indicating qualification for TSMC's 310×310mm CoPoS.
The factors that complicate mass production only amplify as dimensions grow larger. TrendForce cites challenges including variation in TGV diameter caused by fluctuations in laser energy, micro-cracks generated during drilling, metallization of sub-10µm vias, and dynamic alignment within mass-production equipment. On the materials side, maintaining nanometer-level flatness across panels larger than 500×500mm and suppressing warpage arising from differences in thermal expansion coefficients between dissimilar materials remain challenges. Warpage disrupts exposure and alignment, ultimately degrading yield.
Chinese players bring precision glass and laser processing from Lens and large panels with thin-film process expertise from BOE. Taiwanese players can connect panel companies and foundries in close proximity to equipment and materials makers. But neither side's strengths simply add up to mass production. Only by stabilizing TGV, copper filling, RDL, warpage control, and inspection on a single panel—and passing customers' reliability requirements—does this become actual supply capability.
The near-term checkpoints are not the forecast dates of 2027 or 2028 themselves. They are whether Lens can translate Intel's DFM and evaluation methods into actual processes and advance to individual contracts and qualification; whether BOE can reach mass-production yield on its 1,000-panel-per-month pilot line; and how far TSMC discloses its CoPoS panel specifications and substrate materials, and whether Taiwan's supply chain achieves qualification. The race for glass substrates is shifting from a stage measured in announcements to one measured in good units and mass-production revenue.