China is building a domestic TGV glass substrate supply chain for AI/HPC packaging, with reported 2027 deployment targets. The effort spans specialty glass, laser/etch via formation, copper filling, redistribution layers and planarization—because yield depends on mastering the full process, not merely drilling a hol...
Published byEdited with GPT-5.6 TerraImages generated with GPT Image 2
Research answer

Create a landscape editorial hero image for this Studio Global article: How is China’s rapidly scaling Through-Glass Via (TGV) glass-substrate industry addressing the advanced-packaging needs of AI accelerators a. Article summary: China’s TGV push is best understood as a strategic advanced-packaging supply-chain buildout, not yet a proven wholesale replacement for silicon interposers or organic substrates. It aims to make large, low-loss glass int. Topic tags: general, general web, education. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake nu
Glass is becoming a serious candidate for the large, densely connected packages used by AI accelerators and high-performance computing (HPC) systems. China’s response is not simply to manufacture a new substrate: it is to assemble a domestic chain for through-glass vias (TGVs), glass interposers, glass-core substrates and the specialized equipment behind them.
The important qualification is timing. Multiple Chinese firms have moved into sampling, pilot lines and process development, but broad commercial adoption in the most demanding AI packages still depends on yield, reliability testing and customer qualification. 1
8
12
A TGV is a vertical microvia formed through a glass sheet and then metallized—commonly with copper—to connect circuitry on its two sides. These vertical links allow glass to serve either as an interposer between dies or as the structural core of a package substrate. 2
15
For AI packages, the attraction is physical as much as electrical. Large accelerator packages must connect logic dies, memory stacks and package routing with very fine geometries while keeping the assembly flat through thermal cycles. Glass offers high dimensional stability, low electrical loss and a coefficient of thermal expansion (CTE) that can be tailored closer to silicon than conventional organic substrates. Those properties can reduce warpage risk and support finer redistribution-layer (RDL) patterning and high-speed signal paths. 29
31
40
41
That matters when a package must carry many high-bandwidth connections, including links between compute dies and HBM. But better CTE matching is not the same thing as better heat removal: package architects still need dedicated thermal paths for high-power AI devices.
The industry is pursuing two related products:
Against organic substrates, glass has the strongest case in flatness, dimensional stability, lower CTE mismatch and high-frequency behavior. Against silicon interposers, glass may offer larger rectangular panel utilization and a potentially lower-cost material path, while silicon remains the more established platform for very high-density interposer production. 24
29
The strategic value lies in process integration. TrendForce’s supply-chain breakdown includes specialty raw glass and tools such as ultrafast lasers, wet-etch systems, PVD and electroplating equipment; it then extends through thinning, TGV formation, metallization, copper filling, CMP, RDL fabrication and final package integration. 1
Each stage can constrain output:
This is why a reported capacity announcement alone says little about readiness for AI accelerators. Yield and customer certification are the practical gatekeepers.
China’s ecosystem now includes glass and substrate companies, display-panel manufacturers, PCB companies, packaging specialists and equipment suppliers. TrendForce identified 18 publicly traded Chinese companies in the glass-core substrate supply chain in August 2026. 1
BOE has said that glass-based advanced-packaging substrates were sampled with domestic customers, with some concept verification completed and technical testing underway. It also reported full-process integration spanning TGV drilling, deep copper filling and buildup-layer deposition. 8 Separately, WG Tech was reported to have delivered samples of glass substrates for 1.6T optical modules, a useful indication that high-speed networking can be an early proving ground.
6
11
These developments support the view that China is progressing from technology development toward pilots and qualification. They do not establish that domestic glass substrates have already displaced imported materials or silicon interposers at scale.
AI/HPC and HBM are the highest-value targets, but the same capabilities can serve 1.6T optical modules, RF applications, MEMS and advanced fan-out or substrate packaging. 1
6
9
That broader demand matters commercially. Networking and RF products can create production volume and process learning for glass handling, via formation, plating and inspection before the toughest accelerator packages are ready. It also makes the investment less dependent on a single AI-chip customer or packaging architecture.
Reports have linked a 2027 mass-production roadmap to Huawei’s domestic supply-chain plans for glass substrates used in advanced chips, including Ascend AI accelerators. 19 Industry reports also generally frame 2027–2028 as a period when early production could emerge.
27
Those are forward-looking roadmaps, not proof of an achieved market outcome. China’s TGV push is best viewed as a supply-chain and packaging-capability buildout: promising for large, low-loss, high-density packages, but still dependent on difficult execution in specialty glass, via yield, advanced deposition and plating, lithography, metrology, reliability and customer qualification. 1
13
For AI-chip buyers and packaging teams, the near-term signal is clear: glass is moving from a materials concept into a competitive manufacturing race. The winner will be the supplier that can repeatedly deliver flat, reliable, high-yield substrates—not simply the first to announce a TGV line.
Studio Global AI
This page includes a source-backed answer you can continue inside Studio Global.
China is building a domestic TGV glass substrate supply chain for AI/HPC packaging, with reported 2027 deployment targets.
China is building a domestic TGV glass substrate supply chain for AI/HPC packaging, with reported 2027 deployment targets. The effort spans specialty glass, laser/etch via formation, copper filling, redistribution layers and planarization—because yield depends on mastering the full process, not merely drilling a hole through glass.
Early deployments in 1.6T optical modules, RF and other high speed products may provide the process learning needed before the most demanding AI accelerator and HBM packages scale.
China is building a domestic TGV glass substrate supply chain for AI/HPC packaging, with reported 2027 deployment targets. The effort spans specialty glass, laser/etch via formation, copper filling, redistribution layers and planarization—because yield depends on mastering the full process, not merely drilling a hol...
Published byEdited with GPT-5.6 TerraImages generated with GPT Image 2
Research answer

Create a landscape editorial hero image for this Studio Global article: How is China’s rapidly scaling Through-Glass Via (TGV) glass-substrate industry addressing the advanced-packaging needs of AI accelerators a. Article summary: China’s TGV push is best understood as a strategic advanced-packaging supply-chain buildout, not yet a proven wholesale replacement for silicon interposers or organic substrates. It aims to make large, low-loss glass int. Topic tags: general, general web, education. Style: premium digital editorial illustration, source-backed research mood, clean composition, high detail, modern web publication hero. Use reference image context only for broad subject, composition, and topical grounding; do not copy the exact image. Avoid: logos, brand marks, copyrighted characters, real person likenesses, fake screenshots, UI text, readable text, watermarks, charts with fake nu
Glass is becoming a serious candidate for the large, densely connected packages used by AI accelerators and high-performance computing (HPC) systems. China’s response is not simply to manufacture a new substrate: it is to assemble a domestic chain for through-glass vias (TGVs), glass interposers, glass-core substrates and the specialized equipment behind them.
The important qualification is timing. Multiple Chinese firms have moved into sampling, pilot lines and process development, but broad commercial adoption in the most demanding AI packages still depends on yield, reliability testing and customer qualification. 1
8
12
A TGV is a vertical microvia formed through a glass sheet and then metallized—commonly with copper—to connect circuitry on its two sides. These vertical links allow glass to serve either as an interposer between dies or as the structural core of a package substrate. 2
15
For AI packages, the attraction is physical as much as electrical. Large accelerator packages must connect logic dies, memory stacks and package routing with very fine geometries while keeping the assembly flat through thermal cycles. Glass offers high dimensional stability, low electrical loss and a coefficient of thermal expansion (CTE) that can be tailored closer to silicon than conventional organic substrates. Those properties can reduce warpage risk and support finer redistribution-layer (RDL) patterning and high-speed signal paths. 29
31
40
41
That matters when a package must carry many high-bandwidth connections, including links between compute dies and HBM. But better CTE matching is not the same thing as better heat removal: package architects still need dedicated thermal paths for high-power AI devices.
The industry is pursuing two related products:
Against organic substrates, glass has the strongest case in flatness, dimensional stability, lower CTE mismatch and high-frequency behavior. Against silicon interposers, glass may offer larger rectangular panel utilization and a potentially lower-cost material path, while silicon remains the more established platform for very high-density interposer production. 24
29
The strategic value lies in process integration. TrendForce’s supply-chain breakdown includes specialty raw glass and tools such as ultrafast lasers, wet-etch systems, PVD and electroplating equipment; it then extends through thinning, TGV formation, metallization, copper filling, CMP, RDL fabrication and final package integration. 1
Each stage can constrain output:
This is why a reported capacity announcement alone says little about readiness for AI accelerators. Yield and customer certification are the practical gatekeepers.
China’s ecosystem now includes glass and substrate companies, display-panel manufacturers, PCB companies, packaging specialists and equipment suppliers. TrendForce identified 18 publicly traded Chinese companies in the glass-core substrate supply chain in August 2026. 1
BOE has said that glass-based advanced-packaging substrates were sampled with domestic customers, with some concept verification completed and technical testing underway. It also reported full-process integration spanning TGV drilling, deep copper filling and buildup-layer deposition. 8 Separately, WG Tech was reported to have delivered samples of glass substrates for 1.6T optical modules, a useful indication that high-speed networking can be an early proving ground.
6
11
These developments support the view that China is progressing from technology development toward pilots and qualification. They do not establish that domestic glass substrates have already displaced imported materials or silicon interposers at scale.
AI/HPC and HBM are the highest-value targets, but the same capabilities can serve 1.6T optical modules, RF applications, MEMS and advanced fan-out or substrate packaging. 1
6
9
That broader demand matters commercially. Networking and RF products can create production volume and process learning for glass handling, via formation, plating and inspection before the toughest accelerator packages are ready. It also makes the investment less dependent on a single AI-chip customer or packaging architecture.
Reports have linked a 2027 mass-production roadmap to Huawei’s domestic supply-chain plans for glass substrates used in advanced chips, including Ascend AI accelerators. 19 Industry reports also generally frame 2027–2028 as a period when early production could emerge.
27
Those are forward-looking roadmaps, not proof of an achieved market outcome. China’s TGV push is best viewed as a supply-chain and packaging-capability buildout: promising for large, low-loss, high-density packages, but still dependent on difficult execution in specialty glass, via yield, advanced deposition and plating, lithography, metrology, reliability and customer qualification. 1
13
For AI-chip buyers and packaging teams, the near-term signal is clear: glass is moving from a materials concept into a competitive manufacturing race. The winner will be the supplier that can repeatedly deliver flat, reliable, high-yield substrates—not simply the first to announce a TGV line.
Studio Global AI
This page includes a source-backed answer you can continue inside Studio Global.
China is building a domestic TGV glass substrate supply chain for AI/HPC packaging, with reported 2027 deployment targets.
China is building a domestic TGV glass substrate supply chain for AI/HPC packaging, with reported 2027 deployment targets. The effort spans specialty glass, laser/etch via formation, copper filling, redistribution layers and planarization—because yield depends on mastering the full process, not merely drilling a hole through glass.
Early deployments in 1.6T optical modules, RF and other high speed products may provide the process learning needed before the most demanding AI accelerator and HBM packages scale.