High-Bandwidth Memory (HBM) and Advanced Packaging Materials for Artificial Intelligence (AI) Semiconductors: 2027–2037 - Advanced and Emerging Technology Market Research

cover

Published: September 2026
Pages: 343
Tables: 89
Figures: 41

High-bandwidth memory (HBM) and advanced packaging materials form the supply base for artificial intelligence semiconductors. An AI accelerator package combines stacked DRAM, a logic base die, a silicon or glass interposer, a multilayer organic substrate and an integrated thermal solution. Memory and packaging together account for a larger share of package cost than the logic die.

The supply chain comprises five layers. HBM devices represent the largest share of value. Advanced packaging services, including TSMC's CoWoS and SoIC platforms, Intel's EMIB and Foveros, Samsung's I-Cube and X-Cube, and OSAT equivalents, form the second layer. Substrates and interposers form the third. Packaging and memory materials, including build-up dielectric film, copper-clad laminate, glass cloth, photoresist, plating chemistry, underfill and mould compound, form the fourth. Thermal materials and in-package cooling hardware form the fifth.

Demand is driven by accelerator shipments, by the number of HBM stacks per package, and by package body area. All three are increasing. Stack heights are progressing from 8-Hi and 12-Hi toward 16-Hi and beyond, build-up layer counts are rising, and package body areas are growing as reticle multiples increase.

Several technology transitions are underway. Hybrid copper-to-copper bonding is expected to displace micro-bump interconnect in HBM stacks, removing in-stack underfill and solder while introducing bonding dielectrics and associated consumables. Glass core substrates and glass interposers are entering qualification as alternatives to organic cores and silicon interposers. Panel-format packaging is being developed as a successor to wafer-format assembly. Package thermal design power is rising, moving thermal management from conventional lids and thermal interface materials toward micro-channel lids and direct-to-silicon microfluidic cooling.

The materials layer is characterised by high supplier concentration. Build-up dielectric film, low-coefficient glass cloth and several assembly consumables have limited qualified supply bases. Qualification cycles for these materials are long relative to product generations, which constrains the pace at which alternative sources can be introduced.

Custom HBM, in which accelerator vendors specify the logic base die, is transferring value from memory manufacturers to logic foundries. Production is concentrated in Korea, Taiwan and Japan. Korea and Taiwan account for the majority of value across the chain. The United States is increasing share through domestic memory and packaging investment. China is developing a domestic supply chain under export-control conditions.

High-Bandwidth Memory (HBM) and Advanced Packaging Materials for Artificial Intelligence (AI) Semiconductors: 2027–2037 provides market analysis and eleven-year forecasts for high-bandwidth memory and the advanced packaging materials used in AI semiconductor manufacture. It covers five value layers: HBM devices, advanced packaging services, substrates and interposers, packaging and memory materials, and thermal materials and in-package cooling.

Forecasts are presented annually from 2027 to 2037 in constant 2026 US dollars, segmented by layer, material class, application and region, with base, bear and bull scenarios. Material demand is modelled from physical drivers, including stack heights, build-up layer counts, bonding interface area and package body area, and is presented in both value and volume terms.

The report includes a per-stack materials consumption model for HBM covering through-silicon via and plating chemistry, CMP consumables, bonding dielectrics, underfill and non-conductive film, micro-bump solder, and thinning and handling materials. Supplier concentration is assessed for each material class using CR1, CR3 and Herfindahl-Hirschman Index measures calculated at AI-relevant grade.

The study is intended for materials suppliers, substrate and packaging companies, memory manufacturers, semiconductor equipment suppliers, accelerator vendors, and investors requiring segment-level forecasts and supply-chain risk assessment.

Contents include:

Executive summary, headline findings and market summary tables
Scope, methodology, model architecture and scenario definitions
AI demand base: accelerator shipments, HBM stacks per package, package bill of materials, package area roadmap
HBM technology: HBM2 to HBM6 specifications, custom HBM, base-die value migration, bonding roadmap, in-stack thermal management, 3D DRAM
HBM market forecasts: shipments, average selling prices, revenue, generation mix, supplier shares, capacity and demand by customer
HBM stack materials: per-stack consumption model and forecasts by material class
Advanced packaging platforms: CoWoS variants, SoIC, EMIB and Foveros, OSAT platforms, panel-level packaging, co-packaged optics, capacity analysis
Advanced packaging and substrate materials: build-up dielectric film, core materials, copper foil and plating, photoresist, interposers, glass core transition
Thermal materials and in-package cooling: thermal interface materials by chemistry, micro-channel lids, microfluidic cooling
Supply chain analysis: concentration, chokepoints, export controls, regional production and consumption
Consolidated forecasts by layer, application, region and material class, with price indices and scenario analysis
142 company profiles, including ownership, financial data, funding history and technology positions. Companies profiled include 3M, Advanced Micro Devices (AMD), AGC (Asahi Glass), Ajinomoto Fine-Techno, Akhan Semiconductor, Alibaba T-Head (PingTouGe), Alpha Assembly Solutions (MacDermid Alpha), Alphabet (Google), Amazon Web Services (AWS), Amkor Technology, Apple, Applied Materials, ASE Technology Holding, Asetek, Asia Vital Components (AVC), ASMPT, Asperitas, Astera Labs, AT&S, Auras Technology, Avalanche Technology, BE Semiconductor Industries (BESI), Biren Technology, Broadcom, Cambricon Technologies, Carbice, Cerebras Systems, Chemours, ChipMOS Technologies, Cooler Master, CoolIT Systems, CoreWeave, Corintis, Corning, Crossbar, CXMT (ChangXin Memory Technologies), Delta Electronics, d-Matrix, Dow, Element Six, Eliyan, Engineered Fluids, Etched, EV Group, Everspin Technologies, Fabric8Labs, Ferroelectric Memory Company (FMC), Foxconn Industrial Internet (FII), Frore Systems, Fujipoly, Furiosa AI, G42, Green Revolution Cooling (GRC), Groq, Henkel, Heraeus, Hesheng Silicon Industry, Hon Hai (Foxconn), Honeywell International, Hua Tian Technology, Huawei Technologies (HiSilicon), Ibiden, Iceotope Technologies, Iluvatar CoreX, Indium Corporation, Intel, Intel Foundry, JCET Group, JetCool Technologies, Kaneka, Kinsus Interconnect Technology, Kioxia Holdings, Kulicke & Soffa, Kyocera, Lam Research, Lambda, LG Innotek, Liquid Wire, LiquidStack, Macronix International, Marvell Technology, MatX, MediaTek, Meta Platforms, M&I Materials, Micron Technology, Microsoft, Montage Technology, Moore Threads Technology, Multibeam, Murata Manufacturing, Nan Ya PCB, Nanya Technology, Neo Semiconductor and more.....
Market outlook, technology roadmap and strategic conclusions

1. EXECUTIVE SUMMARY 17

1.1 Scope and headline definitions 17
1.2 Key findings 17
1.3 Market size, growth and CAGR summary 19
1.4 Headline forecasts by material class 20
1.5 Headline forecasts by region 22
1.6 Supply concentration at a glance 23
1.7 The asymmetry between device growth and material growth 25
1.8 Strategic implications for suppliers 26
1.9 Strategic implications for buyers 27
1.10 Major market players 27

2. INTRODUCTION, SCOPE AND METHODOLOGY 29

2.1 Report objectives 29
2.2 Product scope: the five value layers 29
2.3 Geographic scope 30
2.4 Units, conventions and abbreviations 30
2.5 Forecast methodology 30
- 2.5.1 Device shipments to stack demand 31
- 2.5.2 Stack demand to material demand 31
- 2.5.3 Price modelling and ASP erosion 31
2.6 Demand scenario definitions 31

3. THE AI DEMAND BASE 33

3.1 The compute stack and where memory sits in it 33
3.2 AI accelerator shipments and package demand 33
3.3 Stacks per package and the HBM demand bridge 35
3.4 Why memory and packaging dominate accelerator cost 36
3.5 The memory wall in serving 37
3.6 Package area growth and the reticle constraint 38
3.7 Hyperscaler capital expenditure and the demand envelope 39
3.8 Sovereign and Chinese demand 40

4. HIGH-BANDWIDTH MEMORY: Technology and Roadmap 42

4.1 HBM architecture and TSV stacking fundamentals 42
4.2 Generation roadmap 43
- 4.2.1 HBM3E: the generation that funded the transition 45
- 4.2.2 HBM4 and HBM4E: interface doubling and the logic base die 45
- 4.2.3 HBM5, HBM5E and HBM6 46
4.3 Custom HBM and the base-die transfer of value 47
4.4 The bonding integration roadmap 49
4.5 Thermal architecture inside the stack 52
4.6 Processing-in-memory and near-memory compute 52
4.7 The memory tier behind HBM 53
4.8 3D DRAM and the post-2032 path 54

5. HBM MARKET FORECASTS 2027–2037 56

5.1 Unit, ASP and revenue forecast 56
5.2 Revenue by generation 57
5.3 Supplier share and competitive structure 58
5.4 Capacity outlook 60
5.5 HBM demand by customer 61
5.6 Regional HBM manufacture 62
5.7 HBM pricing dynamics 63

6. HBM STACK MATERIALS AND CONSUMABLES 65

6.1 The HBM materials bill 65
6.2 Material consumption per stack by generation 66
6.3 TSV formation: plating, barrier and seed 68
6.4 CMP: the class that hybrid bonding creates 68
6.5 Micro-bump metallurgy and its decline 69
6.6 Underfill: capillary, non-conductive film and molded 69
6.7 Hybrid bonding dielectric and consumables 71
6.8 Thinning, carrier bonding and handling 72
6.9 HBM materials market forecast 72
6.10 Concentration and qualification in the HBM material base 74

7. ADVANCED PACKAGING PLATFORMS FOR AI SEMICONDUCTORS 75

7.1 The 2.5D and 3D architecture continuum 75
7.2 CoWoS and the capacity constraint 77
7.2.1 The CoWoS sub-variants 77
7.3 SoIC and 3D hybrid-bonded logic 80
7.4 Intel and Samsung platforms 80
7.5 The OSAT layer 81
7.6 Panel-level packaging 83
7.7 The photonics interface 83
7.8 Advanced packaging services market forecast 84

8. ADVANCED PACKAGING AND SUBSTRATE MATERIALS 86

8.1 The package materials stack 86
8.2 Substrate architecture and body size 87
8.3 Build-up dielectric film 88
8.4 Copper-clad laminate, prepreg, glass cloth and fillers 89
8.5 Copper foil, plating chemistry and photoresist 90
8.6 Interposer materials 91
8.7 The glass core transition and its material consequences 93
8.8 Assembly materials: underfill, mold compound and solder 94
8.9 Package materials market forecast 95
8.10 Regional supply of package materials 96

9. THERMAL MATERIALS AND IN-PACKAGE COOLING 98

9.1 The thermal problem at the package 98
9.2 The thermal interface stack 99
9.3 Liquid metal, phase change and the confinement problem 100
9.4 In-package and direct-to-silicon cooling 100
9.5 Diamond and engineered carbon 101
9.6 Thermal materials market forecast 102
9.7 Supplier consolidation in thermal 104

10. SUPPLY CHAIN, CONCENTRATION AND GEOPOLITICS 105

10.1 The chain from material to accelerator 105
10.2 Export controls and the parallel stack 105
10.3 Qualification duration as the binding constraint 106
10.4 Capacity investment asymmetry 107
10.5 Risk register 108

11. CONSOLIDATED MARKET FORECASTS 2027–2037 110

11.1 Total value pool 110
11.2 Forecast by application segment 111
11.3 Forecast by region 112
11.4 Consolidated materials forecast 113
11.5 Price and ASP trajectories 115
11.6 Scenario analysis 116
- 11.6.1 Sensitivity to the stacks-per-package coefficient 118
- 11.6.2 Sensitivity to hybrid bonding timing 118
- 11.6.3 Sensitivity to glass core adoption 118

12. COMPANY PROFILES 119

12.1 Memory manufacturers 121 (21 company profiles)
12.2 AI accelerator, hyperscaler and system vendors 156 (35 company profiles)
12.3 Foundries, IDMs and OSATs 219 (18 company profiles)
12.4 Substrate, interposer and glass suppliers 245 (15 company profiles)
12.5 Packaging and electronic materials suppliers 265 (15 company profiles)
12.6 Thermal management and in-package cooling 282 (26 company profiles)
12.7 Equipment and process technology 314 (12 company profiles)

13. MARKET OUTLOOK 328

13.1 Market outlook 2027–2037 328
13.2 Technology outlook 329
13.3 Ten strategic conclusions 330
13.4 What would change this forecast 331
13.5 Closing observation 331

APPENDICES 331

Appendix A — Methodology notes 331

REFERENCES 333

14.1 Retrieved sources 333
14.2 Primary corporate, standards and institutional sources 339

List of Tables

Table 1. Headline findings summary (Base case). 18
Table 2. The AI memory and advanced packaging value pool, 2027–2037 (US$ billion). 19
Table 3. Materials and consumables summary by stream, 2027–2037 (US$ million). 21
Table 4. Value pool by region of manufacture, 2027–2037 (US$ billion). 22
Table 5. Concentration summary, HBM and advanced packaging material classes (concentration ratios in per cent; qualification in months). 23
Table 6. Indexed growth, 2027 = 100. 25
Table 7. Model architecture and principal coefficients (coefficients and their units as stated). 30
Table 8. Scenario summary, total value pool 2027–2037 (US$ billion). 31
Table 9. Training versus inference: memory and packaging requirements (qualitative comparison). 33
Table 10. AI accelerator package shipments by class, 2026–2037 (thousand units). 34
Table 11. HBM stacks per package by accelerator class, 2026–2037 (HBM stacks per package). 35
Table 12. HBM stack demand bridge, 2026–2037 (million stacks). 35
Table 13. Indicative flagship accelerator package cost structure, 2024–2037 (per cent of factory cost). 36
Table 14. Compute-to-bandwidth ratio by accelerator generation (relative throughput indexed to Volta = 1.0; bandwidth in TB/s). 38
Table 15. Package area roadmap for flagship accelerators, 2024–2037 (package body area in mm²; layers per side). 38
Table 16. AI infrastructure capital expenditure and derived component demand, 2026–2037 (capital expenditure in US$ billion; derived demand as stated). 40
Table 17. Non-hyperscaler AI accelerator demand, 2027–2037 (thousand packages). 40
Table 18. HBM stack construction elements and their function (descriptive). 43
Table 19. HBM generation technical specifications, HBM2 to HBM6 (specifications per stack). 43
Table 20. HBM4 supplier status and positioning, mid-2026 (status as at mid-2026). 45
Table 21. Custom HBM programmes, suppliers and base-die sources (programmes and first volume year). 47
Table 22. Standard versus custom HBM revenue split, 2026–2037 (US$ billion). 48
Table 23. HBM bonding integration roadmap (bond pitch in µm; interconnect density in I/O per mm²). 49
Table 24. Hybrid bonding adoption in DRAM stacks, 2026–2037 (per cent of stacks shipped). 51
Table 25. In-stack thermal architecture adoption and content per stack (adoption in per cent of stacks; content in US$ per stack). 52
Table 26. Near-memory and in-memory computing landscape (qualitative landscape). 53
Table 27. Memory tier characteristics for AI serving (bandwidth in GB/s; latency in ns; cost in US$ per GB). 53
Table 28. 3D DRAM readiness assessment, 2026 (readiness assessment, 2026). 54
Table 29. HBM unit shipments, ASP and revenue forecast, 2026–2037 (shipments in millions; ASP in US$ per stack; revenue in US$ billion). 56
Table 30. HBM revenue by generation, 2026–2037 (per cent of HBM revenue). 57
Table 31. HBM capacity per stack and bit shipment forecast, 2026–2037 (capacity in GB per stack; bit shipments in exabytes). 58
Table 32. HBM revenue share by supplier, 2022–2037F (per cent). 58
Table 33. HBM wafer capacity and share of total DRAM wafer starts, 2026–2037 (wafer starts in thousand wafers per month; share in per cent). 60
Table 34. Announced HBM-related capacity investments, 2025–2030 (announced investment in US$ billion). 60
Table 35. HBM consumption by customer, 2026–2037 (million stacks). 61
Table 36. HBM manufacture by region, 2027–2037 (US$ billion). 62
Table 37. HBM pricing by generation and configuration, 2026–2037 (US$ per stack). 63
Table 38. HBM stack material classes, function and supply base (descriptive; supply base as stated). 65
Table 39. Material consumption per HBM stack by generation (consumption per stack, units as stated per row). 66
Table 40. TSV formation material demand, 2027–2037 (US$ million). 68
Table 41. CMP and cleaning consumables demand, 2027–2037 (US$ million). 69
Table 42. Micro-bump and solder material demand, 2027–2037 (US$ million). 69
Table 43. Underfill and encapsulation demand, 2027–2037 (US$ million). 70
Table 44. Hybrid bonding consumables demand, 2027–2037 (US$ million). 71
Table 45. Thinning, carrier and handling consumables demand, 2027–2037 (US$ million). 72
Table 46. HBM stack materials and consumables market, 2027–2037 (US$ million). 72
Table 47. HBM material class concentration and entry conditions. 74
Table 48. Advanced packaging platform comparison for AI applications. 75
Table 49. CoWoS-class capacity forecast by variant, 2026–2037 (thousand wafer equivalents per month). 78
Table 50. CoWoS-class supply and demand balance, 2026–2033. 79
Table 51. SoIC variants and AI adoption. 80
Table 52. Comparative platform roadmap, TSMC / Intel / Samsung. 81
Table 53. OSAT capacity, AI exposure and advanced packaging platforms. 81
Table 54. Panel-level packaging adoption forecast, 2027–2037. 83
Table 55. Co-packaged optics packaging demand, 2027–2037. 84
Table 56. Advanced packaging services revenue for AI semiconductors, 2027–2037 (US$ billion). 84
Table 57. Package material classes mapped to process step and multiplier. 87
Table 58. Substrate specification roadmap for AI accelerators, 2027–2037. 88
Table 59. Build-up dielectric film demand, 2027–2037. 88
Table 60. Core material demand, 2027–2037. 90
Table 61. Plating, foil and resist demand, 2027–2037. 91
Table 62. Interposer material comparison. 91
Table 63. Interposer material mix and value, 2027–2037. 92
Table 64. Glass core adoption and material redistribution, 2027–2037. 94
Table 65. Package assembly materials demand, 2027–2037 (US$ million). 94
Table 66. Advanced package materials and consumables market, 2027–2037 (US$ million). 95
Table 67. Package material supply by producing region, 2037. 96
Table 68. Thermal design power and flux by package generation. 98
Table 69. Thermal interface material families for AI packages. 99
Table 70. In-package cooling architecture adoption, 2027–2037 (per cent of flagship AI packages). 100
Table 71. Advanced thermal materials, readiness and application. 101
Table 72. Thermal materials and in-package cooling market for AI semiconductors, 2027–2037 (US$ million). 102
Table 73. TIM demand by chemistry family, AI packages, 2027–2037 (US$ million). 102
Table 74. Supply chain tier concentration and geographic exposure. 105
Table 75. Chinese domestic HBM and packaging stack, 2027–2037. 105
Table 76. Qualification cycle by class and implied earliest volume entry. 106
Table 77. Announced capacity investment by tier, 2025–2030 commitments (US$ billion, indicative). 107
Table 78. Key risk register, probability and impact. 108
Table 79. Total AI memory and advanced packaging value pool, 2027–2037 (US$ billion). 110
Table 80. Value pool by end application, 2027–2037 (US$ billion). 111
Table 81. Value pool by region of manufacture, 2027–2037 (US$ billion). 112
Table 82. Total materials and consumables market, 2027–2037 (US$ million). 113
Table 83. Top fifteen material classes by 2037 value. 114
Table 84. Indexed real price trajectories by class, 2027 = 100. 115
Table 85. Scenario detail by layer, 2037 (US$ billion). 116
Table 86. Scenario trajectories, total value pool 2027–2037 (US$ billion). 117
Table 87. Structural corporate events in the AI hardware and thermal supply base, 2024–2026. 119
Table 88. Technology roadmap summary, 2027–2037. 329
Table 89. Strategic conclusions mapped to stakeholder and action window. 330

List of Figures

Figure 1. The AI memory and packaging value pool, layer map 2027–2037. 19
Figure 2. Value pool by layer, stacked area 2027–2037. 20
Figure 3. Materials market value by class, 2027–2037. 21
Figure 4. Value pool by region, 2027–2037. 23
Figure 5. Concentration heat map by material class. 24
Figure 6. Material demand growth versus device unit growth, indexed. 26
Figure 7. Bull, Base and Bear scenario comparison, 2027–2037. 32
Figure 8. AI accelerator package shipments by class, 2026–2037. 34
Figure 9. HBM stack demand versus modelled supply, 2026–2037. 36
Figure 10. Accelerator package cost structure evolution, 2024–2037. 37
Figure 11. Package area and build-up layer roadmap, 2024–2037. 39
Figure 12. Non-hyperscaler accelerator demand by pool, 2027–2037. 41
Figure 13. HBM stack architecture, cross-section. 42
Figure 14. HBM bandwidth and capacity evolution, HBM2 to HBM6. 44
Figure 15. HBM generation transition timeline and bond-type migration, 2024–2037. 46
Figure 16. Custom HBM share and base-die value migration, 2026–2037. 49
Figure 17. Micro-bump versus hybrid bond stack cross-sections and interconnect density. 51
Figure 18. DRAM cell architecture roadmap and its intersection with the HBM generation cadence. 55
Figure 19. HBM unit shipments and revenue, 2026–2037. 57
Figure 20. HBM revenue share by supplier, 2022–2037. 59
Figure 21. HBM consumption by customer, 2026–2037. 62
Figure 22. HBM price per gigabyte by generation, 2026–2037. 64
Figure 23. Material consumption per HBM stack by generation, indexed. 67
Figure 24. Underfill versus bonding dielectric demand crossover, 2027–2037. 71
Figure 25. HBM stack materials market by class, 2027–2037. 73
Figure 26. Advanced packaging platform positioning by reticle multiple and interconnect density. 77
Figure 27. CoWoS-class capacity by variant, 2026–2037. 79
Figure 28. Advanced packaging capacity by provider and geography, 2026–2037. 82
Figure 29. Co-packaged optics adoption in AI packages, 2027–2037. 84
Figure 30. The AI accelerator package materials stack, exploded cross-section. 86
Figure 31. Build-up dielectric film demand and constrained-grade share, 2027–2037. 89
Figure 32. Interposer material mix by area, 2027–2037. 93
Figure 33. Advanced package materials market by class, 2027–2037. 96
Figure 34. Material production versus consumption by region, 2037. 97
Figure 35. In-package cooling architecture adoption, 2027–2037. 101
Figure 36. TIM demand by chemistry family, 2027–2037. 103
Figure 37. Risk matrix, probability against impact. 109
Figure 38. Total value pool by layer, 2027–2037. 111
Figure 39. Value pool by region, 2027–2037. 113
Figure 40. Material classes by 2037 value and growth rate. 115
Figure 41. Scenario trajectories, 2027–2037. 117

High-Bandwidth Memory (HBM) and Advanced Packaging Materials for Artificial Intelligence (AI) Semiconductors: 2027–2037

PDF + Excel download.