AI hardware builders are becoming AI users

Overview

This report discusses AI’s dual impact on the Asia tech hardware industry: on the one hand, tech hardware suppliers are key enablers of AI infrastructure expansion, spanning the PCB, ODM, power/thermal, connectivity, server assembly, and semiconductor supply chains; on the other hand, these “builders” are applying AI to their own R&D, manufacturing, and product innovation processes. The report clearly places the industry in the AI-impacted “Sooner” camp, arguing that AI is already changing the industry today rather than only having an effect in the distant future.

Core views

The core views include: first, AI training and inference demand is driving significant growth in profits and valuations across the hardware supply chain, with relevant subsegments delivering around 45%-75% EBIT CAGR in 2023-2026E and leading companies’ market caps expanding by around 5x-8x. Second, AI can reduce repetitive work, compress design cycles, and optimize design outcomes in hardware R&D, with typical use cases including RF/antenna, PCB, industrial design, and new materials screening. Third, AI manufacturing applications have evolved from point tools to “AI factories,” covering demand forecasting, robotic automation, digital twins, predictive maintenance, visual quality inspection, and AMR logistics. Fourth, AI will drive functional upgrades in VR/MR, AR/AI glasses, smartphones, and PCs, with Agentic AI in particular potentially creating new demand for edge hardware.

Analysis framework

The report adopts an analytical approach combining supply chain and application scenarios: it first validates the benefits to the hardware industry through the financial performance and market-cap changes of the AI server supply chain, then breaks down AI use cases in R&D, manufacturing, and end products, and finally discusses the potential boost from Agentic AI to upgrade cycles in smartphones, PCs, and edge devices.

Methodology notes

• thematic_frameworkSooner/Later AI framework

  The report uses the “Sooner” and “Later” framework to judge the pace of AI impact. Asia tech hardware is classified into the “Sooner” camp.

  This framework distinguishes whether AI is already changing an industry at the current stage. The report argues that tech hardware suppliers both build AI infrastructure and adopt AI in their own processes, and therefore belong to an industry that benefits and transforms relatively early.

• supply_chain_analysisAI server supply chain mapping

  The AI server supply chain is mapped from fabless, design services, foundries, packaging and testing, memory, substrates, connectivity, power/thermal, PCB, and ODM to servers/racks.

  This method is used to identify the key hardware links in AI infrastructure buildout and to show that non-semiconductor segments such as PCB, ODM, power/thermal, and connectors are also important beneficiaries of AI expansion.

• use_case_analysisAI adoption across hardware workflows

  AI implementation scenarios are analyzed along three main lines: R&D, manufacturing, and product innovation.

  The R&D side focuses on design efficiency and materials discovery; the manufacturing side focuses on forecasting, automation, maintenance, quality inspection, and logistics; and the product side focuses on experience upgrades in VR/MR, AR/AI glasses, smartphones, and PCs.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• PCB supply chain

  Demand for AI servers and high-speed interconnects increases PCB layer count, complexity, and value content.

  Strengths

  Benefits from AI workloads shifting transmission from the MHz range to the GHz range, as well as demand for high layer counts and high-density routing; AI design tools can also improve design efficiency.

  Weaknesses

  High design complexity and long verification cycles require stronger engineering capabilities and higher yield.

  Comparison

  Compared with traditional consumer electronics PCB, AI server PCB is more complex, has more layers, and carries higher value per board.

  Risks

  Fluctuations in AI server demand, overly rapid capacity expansion, customer concentration, and yield risk.

• ODM and server assembly

  ODM is an important assembly and system-integration link for AI servers and racks.

  Strengths

  AI server volume growth drives revenue and profit growth, while digital twins and robotics can improve manufacturing efficiency.

  Weaknesses

  Gross margin is usually constrained by customer bargaining power, component costs, and delivery timing.

  Comparison

  Compared with traditional PC/consumer electronics assembly, AI server ODM requires higher capability in thermal management, power supply, rack-level integration, and supply-chain coordination.

  Risks

  Order volatility from major customers, technology transitions, delivery delays, and capital expenditure pressure.

• Power/Thermal

  Rising power consumption and heat density in AI servers directly drive demand for power supply and thermal solutions.

  Strengths

  High-power AI racks require more complex AC/DC, DC/DC power and thermal solutions, improving suppliers’ pricing power and technical barriers.

  Weaknesses

  Technology routes may change, and there is uncertainty around liquid cooling, air cooling, and system integration solutions.

  Comparison

  Compared with traditional servers, AI servers have higher requirements for thermal management and energy efficiency.

  Risks

  Customer technology-solution shifts, cost pressure, and energy-efficiency regulatory requirements.

• Connectors and optical interconnects

  AI clusters require higher-bandwidth, lower-latency copper and optical interconnect solutions.

  Strengths

  Expansion of training and inference clusters increases demand for high-speed connectivity.

  Weaknesses

  Products iterate quickly, requiring continuous investment in high-speed signaling, optical modules, and reliability capabilities.

  Comparison

  Compared with ordinary data-center connectivity, AI clusters have higher requirements for bandwidth, latency, and stability.

  Risks

  Technology substitution, price competition, and changes in cloud vendors’ procurement pace.

• Consumer electronics and edge devices

  Agentic AI, AR/AI glasses, VR/MR, and AI phones/PCs may drive a new hardware cycle.

  Strengths

  Local AI models, hybrid cloud-edge architectures, and multimodal interaction can increase demand for SoCs, memory, thermal solutions, batteries, and sensors.

  Weaknesses

  Consumers’ willingness to upgrade, application maturity, and battery life remain bottlenecks.

  Comparison

  Compared with traditional smart terminals, AI terminals rely more on system-level permissions, local model capabilities, and continuous perceptive interaction.

  Risks

  Privacy and security restrictions, platform lock-in, unstable model capability, and rising memory prices suppressing demand.

Key data

• EBIT CAGR of AI hardware subsegmentsApproximately 45%-75% (2023-2026E)The report says subsegments such as PCB, ODM, power/thermal, and connectivity achieved strong profit growth over the three-year period.
• Market-cap expansion of leading AI hardware companiesApproximately 5x-8x (2023 to April 2026)The report says leading AI hardware companies saw significant market-cap expansion over the same period.
• Enterprise adoption rate of generative AI88%A mid-2025 McKinsey survey shows that 88% of organizations use generative AI in at least one function.
• Expected enterprise adoption of Agentic AI23% already using, 74% expected to adopt within two yearsA Deloitte survey from August-September 2025 shows strengthening early adoption momentum for Agentic AI.
• ChatGPT paying usersNearly 14 million (December 2025)The report says ChatGPT paying users more than doubled in 2025.
• 32-layer PCB design complexity10k+ components, around 2k+ pins per ASIC, design time close to 3,000 hoursThe Cadence case is used to illustrate the rising complexity of AI server PCB design.
• PCB design efficiency improvementTurnaround time reduced by more than 10xCadence says Allegro X AI can significantly compress PCB design cycles.
• Acceleration in geometric relaxation for materials discoveryUp to 100xThe NVIDIA example shows that AI models can accelerate the geometric relaxation process in materials and chemistry discovery.
• Effect of Sunny Optical’s AI factory30% shorter development cycle, 10% higher UPH, 100% defect detection in key processesThe report uses this as a case study of platformized AI application by a consumer electronics manufacturer.
• Effect of Foxconn digital twinExpected improvement in server manufacturing efficiency, lower costs, and more than 30% reduction in annual energy consumptionFoxconn and NVIDIA are using virtual factories to optimize robotic-arm layout and training.

Impact & implications

The investment implication is that AI hardware should not be viewed only through the main chain of GPUs and semiconductors; segments such as PCB, ODM, power/thermal, connectors, optical modules, equipment, packaging, memory, and server assembly also benefit from AI infrastructure expansion. At the same time, hardware companies’ own adoption of AI can improve R&D efficiency, manufacturing yield, inventory turnover, and automation levels, potentially improving margins and capital efficiency. On the end-device side, Agentic AI, AR/AI glasses, and VR/MR devices may drive a new round of edge-device upgrades, placing higher demands on high-performance SoCs, large-capacity memory, thermal solutions, and low-power design.

Risks

• A slowdown in AI infrastructure capital expenditure or order volatility from cloud vendors could affect demand across hardware supply chains such as PCB, ODM, power/thermal, and connectivity.
• Agentic AI still faces privacy, security, and system-permission constraints, and blockages by super apps or government institutions may affect implementation speed.
• AI agent technology is not yet mature, and third-party LLM calls may lead to inefficient context management and excessively high token costs.
• A weak consumer electronics market and rising memory prices may suppress short-term demand for AI phones, PCs, AR/AI glasses, and VR/MR devices.
• Valuations in the AI hardware supply chain have already expanded sharply; if earnings growth falls short of expectations, valuation pullback risk will be high.
• AI applications in manufacturing require data, system integration, and process transformation, and implementation results may diverge depending on companies’ digitalization foundations.

What to watch

• Changes in 2026E orders, margins, and inventories across the AI server supply chain, especially in PCB, ODM, power/thermal, and connectivity.
• Whether enterprise adoption of generative AI and Agentic AI can move from pilots to scaled deployment.
• Whether actual efficiency gains from AI design tools in PCB, industrial design, and material development can translate into cost advantages.
• Continued implementation progress by companies such as Foxconn, Luxshare, Lenovo, Sunny Optical, Huawei, and Xiaomi in AI manufacturing and AI factories.
• Launch schedules and sales feedback for new AR/AI glasses and VR/MR products from Meta, Apple, Rokid, Samsung, Google, and ByteDance.
• Whether AI phones and PCs will drive stronger demand for local computing power, memory, and thermal solutions due to Agentic AI.