The Wuxi conference’s clearest message was that AI could drive more than $4 trillion of new global data center infrastructure investment by 2028, including $2.8 trillion in semiconductor spending—but HBM supply, advan... SEMI China projected a 2026 HBM market of $54.6 billion, up about 58%, while estimating a 50%–60...
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: What key forecasts, bottlenecks, technological developments, and Chinese industry initiatives were highlighted at the August 31, 2026 Integr. Article summary: The conference’s central message was that AI is pulling semiconductor demand forward sharply, while memory supply, advanced packaging, power delivery, cooling, and equipment self-sufficiency have become the main constrai. Topic tags: general, general web. 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 numbers, clic
AI infrastructure demand was the dominant theme at the 2026 Integrated Circuit Innovation and Development Conference, which opened in Wuxi on August 31. The event’s forecasts were notably bullish, but the more consequential takeaway was where constraints are moving: from leading-edge chip fabrication toward high-bandwidth memory (HBM), heterogeneous integration, power delivery, thermal management and supporting equipment. 21
37
SEMI China president Feng Li said global investment in newly built data-center infrastructure could exceed $4 trillion by 2028, with semiconductor spending reaching about $2.8 trillion. She also said the global semiconductor industry could pass the $1 trillion mark by the end of 2026, earlier than prior expectations that placed that milestone around 2030. These are forecasts presented at the conference, not realized market figures. 21
20
The implication is straightforward: AI demand is not limited to accelerators. It is pulling spending across memory, networking, packaging, manufacturing equipment and materials. That broadening demand helps explain why supply-chain attention is increasingly focused on the technologies required to turn individual dies into high-performance computing systems. 19
37
HBM emerged as the most visible bottleneck. The conference cited a forecast that the 2026 HBM market would grow about 58% to $54.6 billion, approaching 40% of the DRAM market. Yet available capacity was still expected to fall 50%–60% short of demand, even with Samsung, SK hynix and Micron allocating 70% of new or flexible capacity toward HBM. 21
26
That imbalance matters because AI systems require memory bandwidth alongside compute. In practice, it also increases pressure on the advanced-packaging ecosystem that must connect memory stacks, logic dies and other components at high density.
Cao Liqiang of the Chinese Academy of Sciences described the next phase of advanced packaging through four directions: denser interconnects, larger and more functionally integrated packages, more vertical power delivery, and improved heat removal. 34
Cao said conventional packaging provides interconnect densities of roughly tens of I/O connections per square millimeter, while hybrid bonding can exceed one million per square millimeter. The potential benefit is greater bandwidth between components, which could help address memory and interconnect constraints in highly integrated AI hardware. 34
The conference also pointed to a possible transition from planar, three-level power architectures to vertical, two-level approaches for AI systems. The intended result is lower power loss and better efficiency—an increasingly important goal as compute density rises. 34
Higher chip and AI data-center server power raises the importance of cooling as a design constraint rather than a facility afterthought. Cao cited Nvidia’s use of two-phase immersion cooling as an example of the direction of travel. 34
Equipment discussions running alongside the event reflected the same packaging focus. One forum highlighted domestic work in thinning, cutting and hybrid-bonding equipment, including reported hybrid-bond alignment options of 80 nm, 50 nm and 30 nm. 35
The conference was also a vehicle for building regional semiconductor capacity. Jiangsu announced the construction of three provincial key laboratories with total planned investment of more than 500 million yuan, and inaugurated the Jiangsu Integrated Circuit Industry Alliance. 3
41
Among the announced efforts were:
The alliance is intended to link companies, universities, research bodies and public technology platforms across the province. Its relevance is less about a single product announcement than about organizing capabilities across the design, equipment, packaging, materials and power-device layers needed for a more self-sufficient regional ecosystem. 41
42
Fudan Microelectronics chairman and general manager Zhang Wei argued that FPGA development could be reshaped by collaboration with AI agents, creating a tighter development-and-verification loop. He said this could shorten product iteration from months to weeks or even days, and reported that Fudan Microelectronics planned to open trial access to its fmfpga agent development platform in September. 34
The Wuxi event framed AI as a semiconductor-system challenge rather than solely a race to manufacture more advanced logic chips. The critical watchpoints are whether HBM supply can improve, whether hybrid bonding and other advanced-packaging techniques can scale economically, and whether power and cooling architectures keep pace with rising compute density.
For China’s industry strategy, the key test will be execution: whether newly announced laboratories, alliance coordination and equipment development translate into durable capabilities across the parts of the supply chain now under the greatest AI-driven pressure. 33
35
Studio Global AI
This page includes a source-backed answer you can continue inside Studio Global.
The Wuxi conference’s clearest message was that AI could drive more than $4 trillion of new global data center infrastructure investment by 2028, including $2.8 trillion in semiconductor spending—but HBM supply, advan...
The Wuxi conference’s clearest message was that AI could drive more than $4 trillion of new global data center infrastructure investment by 2028, including $2.8 trillion in semiconductor spending—but HBM supply, advan... SEMI China projected a 2026 HBM market of $54.6 billion, up about 58%, while estimating a 50%–60% capacity shortfall despite major memory makers directing 70% of new or flexible capacity to HBM.
Jiangsu launched three provincial key laboratories and the Jiangsu Integrated Circuit Industry Alliance, with initiatives spanning optical electrical microsystems, advanced packaging and AI power chips.
The Wuxi conference’s clearest message was that AI could drive more than $4 trillion of new global data center infrastructure investment by 2028, including $2.8 trillion in semiconductor spending—but HBM supply, advan... SEMI China projected a 2026 HBM market of $54.6 billion, up about 58%, while estimating a 50%–60...
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: What key forecasts, bottlenecks, technological developments, and Chinese industry initiatives were highlighted at the August 31, 2026 Integr. Article summary: The conference’s central message was that AI is pulling semiconductor demand forward sharply, while memory supply, advanced packaging, power delivery, cooling, and equipment self-sufficiency have become the main constrai. Topic tags: general, general web. 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 numbers, clic
AI infrastructure demand was the dominant theme at the 2026 Integrated Circuit Innovation and Development Conference, which opened in Wuxi on August 31. The event’s forecasts were notably bullish, but the more consequential takeaway was where constraints are moving: from leading-edge chip fabrication toward high-bandwidth memory (HBM), heterogeneous integration, power delivery, thermal management and supporting equipment. 21
37
SEMI China president Feng Li said global investment in newly built data-center infrastructure could exceed $4 trillion by 2028, with semiconductor spending reaching about $2.8 trillion. She also said the global semiconductor industry could pass the $1 trillion mark by the end of 2026, earlier than prior expectations that placed that milestone around 2030. These are forecasts presented at the conference, not realized market figures. 21
20
The implication is straightforward: AI demand is not limited to accelerators. It is pulling spending across memory, networking, packaging, manufacturing equipment and materials. That broadening demand helps explain why supply-chain attention is increasingly focused on the technologies required to turn individual dies into high-performance computing systems. 19
37
HBM emerged as the most visible bottleneck. The conference cited a forecast that the 2026 HBM market would grow about 58% to $54.6 billion, approaching 40% of the DRAM market. Yet available capacity was still expected to fall 50%–60% short of demand, even with Samsung, SK hynix and Micron allocating 70% of new or flexible capacity toward HBM. 21
26
That imbalance matters because AI systems require memory bandwidth alongside compute. In practice, it also increases pressure on the advanced-packaging ecosystem that must connect memory stacks, logic dies and other components at high density.
Cao Liqiang of the Chinese Academy of Sciences described the next phase of advanced packaging through four directions: denser interconnects, larger and more functionally integrated packages, more vertical power delivery, and improved heat removal. 34
Cao said conventional packaging provides interconnect densities of roughly tens of I/O connections per square millimeter, while hybrid bonding can exceed one million per square millimeter. The potential benefit is greater bandwidth between components, which could help address memory and interconnect constraints in highly integrated AI hardware. 34
The conference also pointed to a possible transition from planar, three-level power architectures to vertical, two-level approaches for AI systems. The intended result is lower power loss and better efficiency—an increasingly important goal as compute density rises. 34
Higher chip and AI data-center server power raises the importance of cooling as a design constraint rather than a facility afterthought. Cao cited Nvidia’s use of two-phase immersion cooling as an example of the direction of travel. 34
Equipment discussions running alongside the event reflected the same packaging focus. One forum highlighted domestic work in thinning, cutting and hybrid-bonding equipment, including reported hybrid-bond alignment options of 80 nm, 50 nm and 30 nm. 35
The conference was also a vehicle for building regional semiconductor capacity. Jiangsu announced the construction of three provincial key laboratories with total planned investment of more than 500 million yuan, and inaugurated the Jiangsu Integrated Circuit Industry Alliance. 3
41
Among the announced efforts were:
The alliance is intended to link companies, universities, research bodies and public technology platforms across the province. Its relevance is less about a single product announcement than about organizing capabilities across the design, equipment, packaging, materials and power-device layers needed for a more self-sufficient regional ecosystem. 41
42
Fudan Microelectronics chairman and general manager Zhang Wei argued that FPGA development could be reshaped by collaboration with AI agents, creating a tighter development-and-verification loop. He said this could shorten product iteration from months to weeks or even days, and reported that Fudan Microelectronics planned to open trial access to its fmfpga agent development platform in September. 34
The Wuxi event framed AI as a semiconductor-system challenge rather than solely a race to manufacture more advanced logic chips. The critical watchpoints are whether HBM supply can improve, whether hybrid bonding and other advanced-packaging techniques can scale economically, and whether power and cooling architectures keep pace with rising compute density.
For China’s industry strategy, the key test will be execution: whether newly announced laboratories, alliance coordination and equipment development translate into durable capabilities across the parts of the supply chain now under the greatest AI-driven pressure. 33
35
Studio Global AI
This page includes a source-backed answer you can continue inside Studio Global.
The Wuxi conference’s clearest message was that AI could drive more than $4 trillion of new global data center infrastructure investment by 2028, including $2.8 trillion in semiconductor spending—but HBM supply, advan...
The Wuxi conference’s clearest message was that AI could drive more than $4 trillion of new global data center infrastructure investment by 2028, including $2.8 trillion in semiconductor spending—but HBM supply, advan... SEMI China projected a 2026 HBM market of $54.6 billion, up about 58%, while estimating a 50%–60% capacity shortfall despite major memory makers directing 70% of new or flexible capacity to HBM.
Jiangsu launched three provincial key laboratories and the Jiangsu Integrated Circuit Industry Alliance, with initiatives spanning optical electrical microsystems, advanced packaging and AI power chips.