|1 2023 06 14()SAC S1110517070005 SAC S1110521070003 1 MR 2023-05-31 2 23Q1 2023-05-21 3 AI 2023-04-09 AI AI AI Nvidia AI 2018 AI+AI=6 A100 640GB 3423/1 4000 Nvidia AI 800G AI ODM/AI Statista 2021 831.7 6.97%AI 156.3 39.1%AI AI 2026 347.1 5 CAGR 17.3%ODM/CSP 2022 ODM 50%ODM CR5 94.4%/43%17%14%12.8%7.6%AI AI AI GPU ASIC FPGA AI PCB DRAM PCB DRAM Chiplet AI 1)AI 2)PCB 3)4)5)6)AI 7)AI to B/8)Chiplet AI-17%-12%-7%-2%3%8%13%2022-06 2022-10 2023-02 2023-06 300|2 1.AI.4 1.1.AI.4 1.2.AI.4 1.3.AI.5 2.AI.6 2.1.AI.6 2.2.AI.8 2.3.Nvidia AI GPU.11 3.AI.19 3.1.19 3.2.Chiplet.22 3.3.23 3.3.1.AI PCB.23 3.3.2.24 3.3.3.24 3.4.AI.25 4.27 5.28 1 AI CPU+.4 2.7 3.8 4.8 5 AI.8 6 AI.8 7 ODM.9 8 2022 ODM.9 9 2021 H1 AI.9 10 2021 H1 AI.9 11 2021.10 12 2021.10 13 GPT GPU DGX A100 POD.11 14 NVIDIA DGX Systems.12 15 NVIDIA DGX A100.12 16 NVIDIA DGX H100.13 17 NVIDIA DGX H100 H100 NVLink NVSwitch.13 XVBYwPsPpMrQpRtQtOoRmN6MdN6MoMoOnPnOeRrRrPfQrQnQ8OqQuNwMmNnNMYoNoM|3 18 DGX A100 DGX H100 32 256 GPU NVIDIA SuperPOD.14 19 PCIe NVLink.14 20 NVIDIA NVLink.14 21 P100 NVLink.15 22 NVIDIA NVSwitch.15 23 NVSwitch 16 GPU.16 24 NVSwitch.16 25 NVIDIA Mellanox 400G InfiniBand.16 26 NVIDIA ConnectX-7(HCA).16 27 NVIDIA Quantum-2 QM9700 Series.17 28.17 29.17 30 800G Nexus.18 31.19 32 2018 IDC.19 33 AI.20 34 2019/2020.26 35 2021.26 36 AI 2019-2022 Capex.27 1.4 2 AI.5 3 AI.5 4 AI.5 5 2018-2020.6 6.6 7 ChatGPT.7 8 AI.10 9.18 10 AI.20 11 AI GPU.21 12 A100 A800GPU.21 13 GPU.22 14 BR100 370 NVIDIA A100.23 15 Chiplet.23 16 PCB.23 17.24 18.25 19 SSD.25 20.27|4 1.AI 1.1.AI AI AI AI AI CPU+GPU/TPU/ASIC/FPGA 1 AI CPU+AI AI CPU CPU CPU AI ChatGPT AI 1 IP CPU/CPU IT 70%ARM/GPU/NPU AR/VR AI/GPU/FPGA/ASIC CPU/1.2.AI|5 AI AI CPU+CPU CPU ASIC FPGA GPU CPU+AI 2 AI AI GPU CPU/CPU P2P GPU SXM3 P2P 300GB/s SXM2 P2P 50GB/s PCI3.0 P2P 32GB/s GPU-Purley NVlink TensorCore AI-1.3.AI AI AI CPU AI CPU+HGX GPU HGX-1 HGX-2 HGX A100 Tesla P100 GPU Tesla V100 A100 AI AI CPU+GPU AI CPU+GPU/ASIC/FPGA+AI 3 AI 2010-2014 2010 CPU+GPU 2015-2020 2015 Google ASIC TPU 2017 FPGA Project Brainwave 2018 AWS Inferentia 2021-+Kyligence 4 AI|6 2022-2023 IDC 2.AI 2.1.AI AI AlexNet ResNet Transformer BERT AI AI AI AI 5 2018-2020 2018 Open AI GPT-1 1.5 2018 Google Bert-Large 3.4 2019 MT-DNN 3.3 2019 Open AI GPT-2 15.42 2019 PERSEUS-BERT 1.1 2019 NVIDIA Pmojert Megatron 83 2019 Facebook RoBERTa 3.35 2019 Facebook XILM 6.65 2019 NVIDIA Megaltron-Scaled Version of OpenAI GPT-2 83 2020 NLG 172 2020 Open AI GPT-3 1750 ittbank AI AI OpenAI 2020 6=6 GPT-3 1750 45 TB 3000 tokens=6 1.75 1011 3 1011=3.15 1023 FLOPS=3.15 108 PFLOPS OpenAI GPT-3 V100 GPU 21.3%GPT-3 1.48 109 PFLOPS(17117 PFLOPS-day)A100 640GB AI 5 PFLOPS 6.5 kw=AI=2.96 108 1 3423 1 6|7 1750 45TB 3.15*108 PFLOPS 21.3%1.48*109 PFLOPS A100 5 PFLOPS 1 3423 IDC 2021 57.6%42.4%2026 AI 62.2%ChatGPT Finetune ChatGPT GPT1-3 1.5 1750 3640PFlop/s-day SimilarWeb 2023 1 ChatGPT 1 31 2800 3.4%1570 CNBC 1 4000 1570 628 5714.8 PFlop/s-day 7 ChatGPT ChatGPT PFlop/s-day GPT-3 GPT-1/2/3 3640 Finetune-5714.8 2022-2023 IDC Similarweb ittbank AI+AI ChatGPT ChatGPT 3 16 AI 3 30 AI 4 7 4 8 IDC 2021 155.2 EFLOPS AI 2026 1271.4 EFLOPS 5 CAGR 52.3%IDC 5 65%AI 2|8 2022-2023 IDC 2.2.AI AI Statista 2021 831.7 6.97%AI 156.3 39.1%18.79%4.34pct AI AI AI IDC AI 2026 347.1 5 CAGR 17.3%2021 250.9 15.9%AI 59.2 68.2%23.6%IDC 80%2026 AI 123.4 5 CAGR 15.82%3 4 Statista 5 AI 6 AI 31.775155.2268427640.7922.81271.402004006008001000120014002019 2020 2021 2022E 2023E 2024E 2025E 2026E中国智能算力规模（EFLOPS）|9 2022-2023 IDC 2022-2023 IDC ODM ODM ODM ODM+Facebook 2011 OCP ODM 2014Q3 6.6%2019Q3 26.4%2022 848.7 4557 2022 ODM 50%CSP Digitimes Research ODM CR5 94.4%43%17%14%12.8%7.6%+Capex 7 ODM 8 2022 ODM AI AI 2021 TOP5 31.1%20%AI CR5 80%AI AI/AI AI+AI AI 9 2021 H1 AI 10 2021 H1 AI|10 IDC 11 2021 12 2021 8 AI IT 2022 CSP HPC ChatGPT AIGC AI AI AI 2022 AI HPC 2022 20%IT ICT Gartner|11-5G HP E IT X86 JDM IT 1975(EBG)1998 EBG IT 10 PaaS SaaS 20 PKS-CPU PKS PC NO.1 X86 TOP5 TOP9 AI NO.3 ifind wind 2.3.Nvidia AI GPU DGX POD NVIDIA AI HCA 1 NVIDIA DGX Systems DGX A100 DGX H100 Systems 2 HCA NVIDIA DGX H100 NVIDIA ConnectX-7 HCA NVIDIA DGX A100 ConnectX-7 ConnectX-6 HCA 3 DGX BasePOD NVIDIA GPT 10 1 1736 3264 GPU 4 8 DGX A100 POD 1450 1536 GPU GPU 10 20 SUs DGX A100 SuperPOD SU 20 DGX A100 13 GPT GPU DGX A100 POD|12 NVIDIA DGX Systems GPU 2016 Nvidia GPU NVIDIA DGX-1 21 2017 NVIDIA NVSwitch NVIDIA V100 Tensor Core GPU NVLink 2020 NVIDIA A100 Tensor Core GPU NVLink NVSwitch CPU 2022 NVLink NVSwitch NVIDIA H100 Tensor Core GPU 3.6 TB/s 450 GB/s 1.5 3 NVLink NVSwitch NVIDIA NVLink NVLink GPU 14 NVIDIA DGX Systems NVIDIA DGX A100 8 NVIDIA A100 GPU 640 GB GPU GPU 12 NVLink GPU GPU 600 GB 6 NVSwitch 4.8 TB 2 10 NVIDIA ConnectX-7 200GB 500 GB 15 NVIDIA DGX A100|13 NVIDIA DGX H100 8 NVIDIA H100 GPU GPU 640GB GPU 18 NVIDIA NVLink 900GB GPU GPU 4 NVIDIA NVSwitch 7.2TB GPU 1.5 8 NVIDIA ConnectX-7 2 400Gb NVIDIA BLUEFIELD DPU 16 NVIDIA DGX H100 gdep DGX H100 8 GPU+NVLink+NVSwitch DGX H100 H100 GPU NV H100 GPU NVLink NVLink NVSwitch H100 Tensor Core GPU 17 NVIDIA DGX H100 H100 NVLink NVSwitch,DGX H100 256 GPU 57.6TB NVLINK Network NVSwitch GPU GPU NVLink 2:1 NVLink Switch NVSwitch|14 DGX H100 SuperPod 256 GPU 57.6 TB NVSwitch NVLink Switch NVLink Switch System 18 DGX A100 DGX H100 32 256 GPU NVIDIA SuperPOD NVLink PCIe Volta NVLink 300GB/s PCIe 3.0 16GB/s NVLink 100 Gbps PCIe Gen 5 32 Gbps NVLink NVLink GPU 1.5 NVIDIA H100 Tensor GPU 18 NVLink 900 GB(GB/)5 PCIe 7 19 PCIe NVLink IBM Redbooks NVLink 0.5 GPU PCIe NVLink NVIDIA A100 Tensor GPU NVLink 12 NVLink 600 GB H100 GPU NVLink H100 18 NVLink 900 GB H100 GPU A100 1.5 20 NVIDIA NVLink|15 NVLink PHY DL TL 1 PL PHY 2 DL 3 TL 21 P100 NVLink NVSwitch NVLink GPU NVSwitch GPU NVSwitch 64 NVLink GPU NVSwitch NVLink NVLink GPU 22 NVIDIA NVSwitch NVSwitch NVSwitch GPU|16 GPU GPU NVLink InfiniBand 4.5 14 TB InfiniBand H100 NVLink H100 NVSwitch Nvidia TSMC 4N 251 NVIDIA V100 Tensor Core GPU 294 50 mm x 50 mm 2645 23 NVSwitch 16 GPU 24 NVSwitch NVIDIA Quantum InfiniBand InfiniBand InfiniBand InfiniBand(HCA)NVIDIA 2020 Nvidia SC20 NVIDIA Mellanox 400G InfiniBand 400Gb/s(AI)25 NVIDIA Mellanox 400G InfiniBand NVIDIA ConnectX InfiniBand NVIDIA ConnectX I/O ConnectX-7(HCA)NVIDIA Quantum-2 InfiniBand 400GB ConnectX-6 HDR(HCA)NVIDIA Quantum InfiniBand 200GB 26 NVIDIA ConnectX-7(HCA)|17 NVIDIA Quantum InfiniBand QM8700 InfiniBand 16Tb 40 200Gb QM9700 InfiniBand 64 400Gbps 128 200Gbps 400Gbps 2,048 200Gbps 4,096 27 NVIDIA Quantum-2 QM9700 Series Nvidia OSI LD LED 28 Huawei 18%73%4%5%18%29|18 1 VCSEL FP DFB EML 2 PIN APD 9 VCSEL 800-900nm 500 FP 1310-1550nm DFE 1270-1610nm EML 1270-1610nm PIN 830-860/1100-1600nm APD 1270-1610nm 800G 2022 10 OCP 800G 2023 1 Juniper Virgin Media O2 Juniper IP Virgin Media O2 Juniper PTX10008 PTX10008 400G 800G 2023 4 S12500G-EF 400G,800G,30 800G Nexus 73%4%5%18%光器件 外壳 印刷电路板 光芯片|19 SDN LAB 3.AI AI AI AI PCB/AI GPU Chiplet 31 Wind 3.1.IDC 2018 32%80%32 2018 IDC|20 IDC AI CPU AI AI GPU FPGA ASIC CPU+AI 2021 AI 297 2025 1385 2021-2025 CAGR 47%33 AI GPU IDC 2 GPU 3 FPGA 3 ASIC 18 4 GPU 7 FPGA 5 ASIC GPU 90%GPU ASIC FPGA NPU GPU 10%GPU AI 10 AI AI GPU 126197297441647946138556.35%50.76%48.48%46.71%46.21%46.41%0.00%10.00%20.00%30.00%40.00%50.00%60.00%0500100015002019 2020 2021E 2022E 2023E 2024E 2025E中国AI 芯片市场规模（亿元）yoy|21 AMD Intel FPGA FPGA FPGA 40mm 55mm ASIC GPU+AI GPU GPU GPU AI Nvidia AMD GPU Nvidia 2021Q1 83%Nvidia GPU AI 100%GPT-3.5 2 GPU 3 AI Nvidia AI GPU AI/I/O 600GB/s TOPS 4800 A100 AI AI IT 2022 AI 179.5 GPU AI 11 AI GPU GPU NF5688M6 8 NVIDIA NVSwitch 500W Ampere GPU X660 G45 LP 8 NVIDIA A800 AI 8 Atlas300I PR4910W 8 NVIDIA A800/10 NVIDIA A40/30 12 A100 A800GPU A100 A800 600GB/s 400GB/s 2TB/s 2TB/s 80GB 80GB FP64 9.7TFLOPS 9.7TFLOPS|22 FP32 19.5TFLOPS 19.5TFLOPS Tensor Float 32 TF32 156TFLOPS/312 TFLOPS 156TFLOPS/312 TFLOPS 51CTO GPU GPU GPU 13 GPU 300474.SZ GPU JM9 2021 11 16 2022 6 28 CAD 688047.SH 2022 16 3C5000 32 3D5000 688041.SH DCU 688521.SH Vivante3DGPGPUIP AI 2021 3 GPU 100 2022 12 GPU 100 2022 9 GPU BR100 Chiplet Wind IC 3.2.Chiplet Chiplet AI ChatGPT+GPT ChatGPT Chiplet HPC AMD 2021 6 3D Chiplet 2019 Chiplet 7nm 920 20%Chiplet Chiplet AI A100 Chiplet Chiplet Chiplet|23 Chiplet ittbank 2021 14 BR100 370 NVIDIA A100 BR100 2022 370-X8 2021 NVIDIA A100 2020 770 390 542 1074mm2-828 mm2 7nm 7nm 7nm 64GB HBM2E 48GB 80GB HBM2E 2048TOPS INT8 256TOPS INT8 624TOPS INT8 TDP 550W 250W 400W Chiplet Chiplet 2022 12 Chiplet 15 Chiplet Chiplet XDFOI TM Chiplet 2.5/3D Chiplet Chiplet 3D FO SiP HFCBGA Chiplet 5G 3.3.3.3.1.AI PCB AI PCB PCB Riser PCB PCB PCB PCB 8-10 M6 3400 AI PCB 18-20 M8 10350 PCB 14-16 M6 7140 AI PCB PCB 2-3 PCB Prismark 2021 PCB 78.04 2026 124.94 5 CAGR 9.9%PCB+/PC PCB PCB 16 PCB PCB PCB AI 2020-2022 PCB 60%Intel Purley Whitley Eagle Stream AMD PCB ARM|24 Eagle Stream AI HDI 4 HDI 6 HDI 2021 PCB 3.3.2.NVIDIA DGX A100 AI 6.5KW 30-50%1000-3500 15-25 AI 17 PCM EMI/IDC IDC Intel CPU Wind 3.3.3.DRAM NAND Flash SSD DRAM DRAM JEDEC DRAM DDR DDR DDR DDR AI DDR GPU HBM ChatGPT HBM AI 2023 ChatGPT SK HBM AI A100 V100 HBM2 DRAM HBM DRAM|25 DRAM SK HBM DRAM DDR GDDR 2020 DRAM CR3 94.5%2020 DRAM 34.9%18 DDR/DDR2/DDR3/DDR4 DDR5 DDR4/LPDDR4/LPDDR4X/DDR5/LPDDR5 GDDR6 GDDR6 DDR3/DDR4/LPDDR4 Flash DRAM DDR3/LPDDR2 DDR3/DDR4/DDR4X DDR5/LPDDR4X Wind SSD SSD I/O SSD X86 PCIe 4.0 SSD SSD SSD AI SSD SSD SSD SSD&SSD AI SSD 19 SSD PM7 SSD 20%30.72TB Ultrastar DC SN840 SSD 1 3 DW/D TCG HPC 5G DC1500M Gen 3.0 x4 PCIe NVMe 3D NAND NVM SSD SSD SSD 3.4.AI HDC Netflix Zoom|26 CSP AWS Azure 34 2019/2020 TrendForce 40%BAT IDC 2022 2022 60 200 2023 capex 2022 AI AI 35 2021 IDC AI Trendforce 2022 AI AWS Meta 66%6%2.3%1.5%1.5%AI Capex capex AI ChatGPT AI AI 企业资料中心 超大型资料中心电信资料中心 其他20192020|27 36 AI 2019-2022 Capex Wind 4.AI 1)AI 2)PCB 3)4)5)6)AI 7)AI to B/:8)Chiplet 20 EPS/PE 2021A 2022A 2023E 2021A 2022A 2023E 2021A 2022A 2023E 4395.57 5118.50 5745.00 1.01 1.01 1.20 11.83 9.08 19.03 670.58 695.25 826.86 1.38 1.42 1.80 26.01 15.14 29.18 22.30 22.37 30.92 0.07-0.81 0.10 124.76-8.17 194.58 177.90 140.27 189.02 0.20 0.04 0.17 69.63 273.55 84.88 328.35 305.65 347.14 1.13 0.77 1.03 20.82 14.76 19.84 GPU 10.93 11.54 17.21 0.97 0.64 0.90 156.62 85.84 114.11 12.01 7.39 16.59 0.66 0.13 0.52-662.11 276.31 23.10 51.25 74.46 0.16 0.35 0.56-116.05 145.46 21.39 26.79 33.88 0.03 0.15 0.30 2881.83 297.18 294.27 FPGA 25.77 35.39 43.27 0.63 1.32 1.73 79.70 52.94 32.09 53.42 71.20 92.27 3.22 3.10 4.05 69.88 42.55 21.34 ASIC 7.21 7.29 10.75-2.06-3.13-1.95-46.12-17.40-115.19 305.02 337.62 337.61 1.66 1.82 1.52 18.66 12.70 21.35 158.12 214.29 246.29 0.72 0.33 0.50 26.99 49.68 51.00 120.97 119.06 133.80 0.44 0.24 0.25 28.77 35.23 38.05 DRAM 676.38 740.58 850.81 0.75 0.75 0.94 30.43 25.86 36.45 85.10 81.30 71.53 3.50 3.08 2.00 50.23 33.30 56.00 11.34 11.46 13.45 0.59 0.42 0.55 76.32 62.41 67.04 97.49 83.30 101.16 2.73 0.18 0.71-334.50 159.21 PCB 333.15 362.11 409.00 1.43 2.16 2.23 29.69 12.71 11.13 20.76 24.12-0.27 0.74-139.43 139.92 157.82 3.03 3.20 3.60 40.25 22.57 22.00|28 36.47 35.35 38.97 0.32 0.38 0.44 43.94 24.75 28.66 74.32 78.85 93.33 0.78 0.92 1.09 38.97 14.13 24.49 12.48 15.92 20.12 0.47 0.69 0.92 46.81 21.50 27.26 30.58 41.25 54.70 0.06 0.19 0.46 422.50 75.99 37.04 5.8 11.50 22.81-0.38 0.04 0.21-21.79 190.65 34.54 8.29 10.08 18.12 2.31 1.37 1.40 55.39 67.28 35.78 16.79 19.04 12.85 0.23 0.93 0.37 68.65 10.68 59.75 2023 6 14 Wind 5.AI GPU AI AI|29/6 300 6 300 20%10%-20%-10%-10%-10%5%-5%-5%-5%36 100031 3 A 23 2301 570102(0898)-65365390 6 4 200086(8621)-65055515(8621)-61069806 5033 71 518000(86755)-23915663(86755)-82571995