--- title: "Everbright Securities: NVIDIA plans to introduce the LPU solution, and AI inference is expected to extend to the PCB equipment field" type: "News" locale: "en" url: "https://longbridge.com/en/news/277592843.md" description: "Everbright Securities released a research report stating that NVIDIA plans to introduce the LPU solution in its new chip architecture, marking a shift in the AI computing market from \"general computing\" to \"dedicated inference.\" The low latency and high bandwidth characteristics of LPU complement GPUs and are expected to significantly increase PCB demand, driving incremental demand for PCB micro-drilling and processing equipment. Global demand for AI computing continues to grow, and the implementation of the LPU solution will accelerate the extension of industry prosperity into the PCB equipment sector" datetime: "2026-03-03T09:14:06.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/277592843.md) - [en](https://longbridge.com/en/news/277592843.md) - [zh-HK](https://longbridge.com/zh-HK/news/277592843.md) --- > Supported Languages: [简体中文](https://longbridge.com/zh-CN/news/277592843.md) | [繁體中文](https://longbridge.com/zh-HK/news/277592843.md) # Everbright Securities: NVIDIA plans to introduce the LPU solution, and AI inference is expected to extend to the PCB equipment field According to the Zhitong Finance APP, Everbright Securities released a research report stating that NVIDIA (NVDA.US) plans to introduce the LPU solution in its new chip architecture, marking a shift in the AI computing power market from "general computing" to "dedicated inference." If NVIDIA's LPU solution is implemented, the demand area for PCBs will significantly increase, and the processing difficulty of PCB materials will further rise, which is expected to bring substantial incremental demand for PCB micro-drills and PCB processing equipment. The global demand for AI computing power continues to grow rapidly, while the demand for low latency in AI inference is constantly increasing. The heterogeneous architecture of GPU + LPU is expected to accelerate implementation, and the industry's prosperity is likely to extend to the PCB equipment sector, with PCB drill needles potentially experiencing a situation of supply not meeting demand and product price increases. ## The main points of Everbright Securities are as follows: **LPU has low latency and high bandwidth characteristics, complementing GPUs in AI workflows** LPU (Language Processing Unit) is a dedicated processor designed specifically for AI inference, especially for low-latency real-time interactions (such as conversations). Its core is to achieve deterministic execution through "compiler-driven" static scheduling and relies on high-speed on-chip SRAM (with bandwidth up to 80TB/s) to eliminate memory bottlenecks, thereby reducing initial latency to about a hundred milliseconds. In inference on mainstream large models (taking the Llama2-70B model as an example), it is approximately 10 times faster than the H100 GPU, with overall energy efficiency improved by about 10 times. In contrast, GPUs (such as the H100) are general-purpose high-throughput architectures that rely on large-capacity HBM memory and excel in large-scale parallel computing, serving as the mainstay for large model training and high-throughput tasks. However, in scenarios requiring single-sequence, real-time generation, they are limited by memory bandwidth and runtime scheduling, making it difficult to break through the low-latency bottleneck. **LPU and GPU are forming a complementary relationship in AI workflows:** GPUs, with their powerful parallel capabilities and large memory, remain central to model training and processing large contexts (Prefill stage); whereas LPUs have significant advantages in text generation that requires immediate response (Decode stage), making them suitable for high-concurrency online inference services. NVIDIA's plan to introduce the LPU solution in its new chip architecture marks a shift in the AI computing power market from "general computing" to "dedicated inference." **The incremental impact of LPU on PCB equipment and drill needles is primarily reflected in the increase in PCB value and advanced packaging in two aspects** **1) Increased PCB usage area and upgraded PCB material requirements:** Due to the capacity bottleneck of a single LPU's 230MB SRAM, running large-scale models requires hundreds of LPU units in series. Therefore, if LPU is widely applied, the required PCB substrate area will increase several times compared to pure GPU architecture solutions. Additionally, considering signal transmission efficiency, the material requirements for PCBs needed for LPU will be higher, and it is expected to use a reinforced solution of 52-layer M9-grade copper-clad laminates + Q fabric, which will lead to a huge consumption of drill needles, with the consumption of PCB drill needles expected to increase significantly. In summary, the implementation of NVIDIA's LPU solution will significantly increase the demand area for PCBs and further elevate the processing difficulty of PCB materials, which is expected to bring substantial incremental demand for PCB micro-drills and PCB processing equipment **2) PD Separation and 3D Stacking Solutions or Enhanced Advanced Packaging Requirements:** NVIDIA proposed the Prefill-Decode Disaggregation (PD separation) technology at the 2025 GTC conference, which splits LLM inference into two stages: **Compute-intensive Prefill and Memory-intensive Decode** In response to issues such as insufficient server layout space and increased wiring density caused by large-scale LPU cluster stacking, NVIDIA is expected to achieve complementary coexistence of GPU and LPU through PD separation technology, reducing the deployment scale of a single LPU. At the same time, it is anticipated that 3D stacking technology will allow Groq's LPU units to be directly stacked on top of the GPU main chip, compensating for insufficient SRAM capacity through multi-chip collaboration while maintaining low latency advantages, achieving deep integration of general computing and dedicated inference at the physical layer. The heterogeneous architecture of GPU + LPU has high requirements for packaging technology and precision, and it is expected that the demand for high-precision assembly equipment in the PCB electronic assembly process will further increase. **Target Aspects** It is recommended to pay attention to equipment manufacturers in the core manufacturing segment of PCB: 1) For high-precision drilling and exposure, focus on Dazhu CNC (301200.SZ, 03200), Inno Laser (301021.SZ), and Dier Laser (300776.SZ); 2) For high-precision assembly equipment for PCB, focus on Kaige Precision Machinery (301338.SZ) and Jintuo Co., Ltd. (300400.SZ); 3) For high-end PCB drilling needles, focus on Dingtai Gaoke (301377.SZ), World (688028.SH), and Sifangda (300179.SZ); 4) For advanced plating processes, focus on Dongwei Technology (688700.SH), etc. **Risk Warning** Risks include intensified industry competition, technological iteration, industrial transfer, and terminal demand falling short of expectations ### Related Stocks - [NVIDIA Corporation (NVDA.US)](https://longbridge.com/en/quote/NVDA.US.md) - [YieldMax NVDA Option Income Strategy ETF (NVDY.US)](https://longbridge.com/en/quote/NVDY.US.md) - [VanEck Semiconductor ETF (SMH.US)](https://longbridge.com/en/quote/SMH.US.md) - [iShares Semiconductor ETF (SOXX.US)](https://longbridge.com/en/quote/SOXX.US.md) - [GraniteShares 2x Long NVDA Daily ETF (NVDL.US)](https://longbridge.com/en/quote/NVDL.US.md) - [Direxion Daily Semicondct Bull 3X ETF (SOXL.US)](https://longbridge.com/en/quote/SOXL.US.md) - [Direxion Daily NVDA Bull 2X Shares (NVDU.US)](https://longbridge.com/en/quote/NVDU.US.md) - [T-REX 2X Long NVIDIA Daily Target ETF (NVDX.US)](https://longbridge.com/en/quote/NVDX.US.md) - [GraniteShares 2x Short NVDA Daily ETF (NVD.US)](https://longbridge.com/en/quote/NVD.US.md) - [XL2CSOPNVDA (07788.HK)](https://longbridge.com/en/quote/07788.HK.md) - [XI2CSOPNVDA (07388.HK)](https://longbridge.com/en/quote/07388.HK.md) - [Direxion Daily NVDA Bear 1X ETF (NVDD.US)](https://longbridge.com/en/quote/NVDD.US.md) - [T-REX 2X Inverse NVIDIA Daily Target ETF (NVDQ.US)](https://longbridge.com/en/quote/NVDQ.US.md) ## Related News & Research - [LPU: Nvidia popularizes a new AI acronym](https://longbridge.com/en/news/279662346.md) - [Nvidia Asked by US Senators to Provide Details on Groq Deal](https://longbridge.com/en/news/279931631.md) - [Nvidia GTC Day 3 Highlights Open Models, Inferencing Driving Revenues, UBS Says](https://longbridge.com/en/news/279813617.md) - [Nvidia In Crosshairs As Senators Question China Export Controls](https://longbridge.com/en/news/280338466.md) - [How Autonomous AI Agents Become Secure by Design With NVIDIA OpenShell](https://longbridge.com/en/news/280185602.md)