Asia supply-chain feedback remains strong, and AI compute demand continues to outpace supply

Overview

This report summarizes the main conclusions from Barclays' meetings with IT hardware supply-chain companies in Asia. Overall feedback was positive: demand tied to AI computing continues to outpace supply, and supply-chain discussions centered on capacity constraints, component shortages, and multi-vendor configurations. Strong segments include TPU, GPU/AMD AI clusters, cloud CPUs, data center switching, power, and cooling; relatively weak segments include PCs, smartphones, campus switching, and WiFi.

Core views

The core view is that AI infrastructure remains the most important growth driver for the IT hardware supply chain. The TPU chain is expanding well, with CLS, FLEX, JBL, and other ODMs participating at different points; Trainium is growing steadily but remains smaller in scale than TPU; NVDA GPUs and AMD clusters are expected to ramp strongly by year-end, driving demand for servers, Ethernet, transceivers, switching equipment, power systems, and rack suppliers. Mainstream cloud CPU demand has improved materially versus six months ago, with unit expectations revised from flat to growth of 20%-40%, alongside higher ASPs. At the same time, device-side businesses are under pressure from rising memory prices and weak consumer demand.

Analysis framework

The report mainly uses supply-chain conference research, inferring demand, capacity, share, and revenue trends from feedback across Asian EMS, ODM, white-box switching, test equipment, power, server, and device supply chains, and mapping those trends to Barclays-covered companies based on business exposure.

Methodology notes

• Supply Chain ResearchAsia Conference Notes

  Validating downstream demand, capacity expansion, and supplier share changes through meetings with supply-chain companies

  This approach is well suited to assessing short-term order trends and capacity trends in the AI hardware chain, but the actual capital spending pace of end customers and long-term share shifts still require follow-up data validation.

• Relative Rating SystemBarclays Equity Rating System

  Overweight, Equal Weight, and Underweight refer to expected 12-month total return relative to the industry coverage portfolio

  The stock ratings disclosed in the report are not simple buy, hold, or sell labels, but a relative-performance assessment versus the analyst's industry coverage portfolio.

• Industry ViewIndustry View

  Positive, Neutral, and Negative are used to describe the fundamentals and valuation trend of the industry coverage portfolio

  The industry view in this report is Neutral, meaning the industry's fundamentals and valuations are broadly stable, but the report text is more constructive on AI-related subsegments and individual stocks.

Asset mapping & comparison

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

• CLS

  Main TPU supplier and potential beneficiary of white-box switching and AI cluster Ethernet

  Strengths

  Seen as a key supplier for TPU and its variants, with 2026 TPU-related revenue expected to grow sharply; expansion in the white-box switching TAM is also favorable for CLS.

  Weaknesses

  Long-term share could be diluted if new suppliers enter.

  Comparison

  Compared with other EMS/ODMs, CLS has a more central position in the TPU and white-box switching chain.

  Risks

  TPU expansion falls short of expectations, customer multi-vendor strategies, and more intense competition from customers such as Meta.

• FLEX

  Second-source TPU supplier, power system supplier, and rack supplier

  Strengths

  Still participates in TPU manufacturing, and has high exposure to the power business; FY26 power revenue is expected to be US$2.1 billion, up 61%.

  Weaknesses

  TPU revenue growth may lag other ecosystem participants.

  Comparison

  It stands out more in power, while in TPU it is more of a second source.

  Risks

  Share loss and AI power business growth coming in below company guidance.

• JBL

  Participant in test equipment, Trainium racks, NVDA switching equipment, and power supply chains

  Strengths

  Growth in test capacity, Trainium rack buildout, and NVDA switching and power businesses all provide growth drivers.

  Weaknesses

  The report does not view JBL as a direct core TPU supplier.

  Comparison

  Compared with CLS, JBL benefits more from adjacent areas such as testing, racks, switching, and power.

  Risks

  Delays in GPU or Trainium ramp-up and volatility in test-equipment demand.

• FN

  Trainium supply-chain participant and NVDA transceiver supplier

  Strengths

  Trainium line expansion is significant; as a leading NVDA transceiver supplier, its 1.6T products could help reignite growth.

  Weaknesses

  Last quarter was slightly below expectations, and Trainium growth is smaller in scale than TPU.

  Comparison

  It has more direct leverage in optical communications and transceivers.

  Risks

  Third-source entries, changes in customer project timing, and slower-than-expected 1.6T ramp-up.

• DELL

  Beneficiary of GPU servers and enterprise servers

  Strengths

  Its GPU-server positioning is strong, helped by SMCI issues and new Texas capacity; enterprise server demand is also holding up well.

  Weaknesses

  Still depends on the GPU-server cycle and a recovery in enterprise demand.

  Comparison

  Its branded-server and GPU-server integration is closer to end customers than that of other EMS players.

  Risks

  GPU supply constraints, competitive recovery, and enterprise demand below expectations.

• ANET

  Beneficiary of front-end and back-end networking for AI clusters and front-end networking for cloud CPUs

  Strengths

  Improving cloud CPU server demand benefits front-end networking, and AMD AI cluster front-end and back-end networking also favors ANET; limited white-boxing activity among emerging cloud customers supports high-quality branded vendors.

  Weaknesses

  Long-term competition from white-box switching remains a structural headwind.

  Comparison

  Compared with white-box suppliers, ANET benefits from demand for high-end branded network equipment.

  Risks

  Faster adoption of customer-designed white-box strategies, changes in AI networking standards, and intensifying competition.

• AAPL

  A smartphone demand watch item

  Strengths

  iPhone continues to sell well in China, and promotions are helping to sustain momentum.

  Weaknesses

  The report questions whether this momentum is sustainable.

  Comparison

  Weakness in premium Android makes iPhone perform relatively better, but promotional support may not be sustainable.

  Risks

  A pullback in China demand, unsustainable promotions, and rising memory prices compressing margins.

• CSCO

  Participant in branded networking equipment opportunities

  Strengths

  When the white-box transition is not moving quickly, branded networking vendors also see opportunities.

  Weaknesses

  Its growth optionality is weaker than that of networking names with purer AI exposure.

  Comparison

  Compared with ANET, CSCO has opportunities, but the report's tone is softer.

  Risks

  Cloud customers moving toward white-boxing, and weak demand for campus switching and WiFi.

• MSI

  Beneficiary of the video security and camera ecosystem

  Strengths

  The security business has stronger long-term growth characteristics, ASP improvement supports revenue, and the report believes MSI may gain share.

  Weaknesses

  The camera industry still faces some pressure.

  Comparison

  Compared with PCs and phones, video security is more of a structural-growth area.

  Risks

  Slower end-market demand, intensifying competition, and ASP improvement falling short of expectations.

Key data

• CLS 2026 TPU-related revenue forecastOver US$2.8 billionBarclays expects this to rise from about US$1 billion in 2025 to more than double.
• Trainium unit growth viewAround 30% or moreGrowth is solid but still below TPU, and it could be higher if adopted by more cloud providers.
• Mainstream cloud CPU unit demand changeGrowth of 20%-40%This is a clear improvement from the flat expectation six months ago, and it comes with higher ASPs.
• FLEX FY26 power revenueUS$2.1 billion, up 61%The report believes FLEX has the highest exposure in power and could be a share gainer.
• JBL annual power revenue estimateAbout US$600 million, up more than 30%Hanley's addition is expected to further strengthen this business.
• Full-year PC unit volume outlookDown 5%-10%ODMs expect a rebound in the later part of the second quarter, which is slightly better than Barclays' industry model.
• Number of Barclays-covered companies1851 companiesAmong disclosed ratings, 51% are Overweight, 34% are Equal Weight, and 14% are Underweight.

Impact & implications

The investment implication is that AI data-center capex is still spreading into server and accelerator supply chains, switching networks, power, and cooling. Companies more directly tied to AI computing and cloud infrastructure may continue to see revenue upgrades and better order visibility, while consumer devices, PCs, smartphones, and traditional campus-networking chains lack comparable growth momentum. Supply constraints and multi-vendor strategies, on the one hand, support revenue ramp-up for existing suppliers; on the other hand, they also create share dilution and execution risk.

Risks

• AI accelerator ramps may be delayed; the report mentions potential delays in both NVDA GPU and AMD projects.
• The commercialization pace and actual production scale of TPU, Trainium, Maia, and Meta's in-house accelerators remain uncertain.
• Multi-vendor adoption may reduce the share and pricing power of existing suppliers.
• Component shortages, capacity constraints, and bottlenecks in power and cooling may affect delivery.
• Weak demand for PCs, smartphones, campus switching, and WiFi could drag on the non-AI hardware chain.
• Rising memory prices have a larger impact on device-side businesses and may compress margins or suppress demand.
• Barclays has market-making, investment banking, or other service relationships with multiple covered companies, so investors should pay attention to potential conflicts of interest disclosures.

What to watch

• Whether TPU expansion in 2026 continues to meet or exceed plan, and how share shifts play out among CLS, FLEX, and JBL.
• Whether Trainium is adopted by more cloud providers, pushing growth above the 30%+ view.
• Whether NVDA GPU and AMD AI cluster ramps materialize on schedule by year-end.
• Whether scale-up Ethernet becomes a new growth driver for the data center switching ecosystem in 2027.
• Whether 20%-40% unit growth in cloud CPUs and higher ASPs translate into revenue for EMS and networking companies.
• Whether FLEX and JBL power business growth tracks the rise in AI rack power consumption and cooling demand.
• Whether the second-half rebound in PCs and the full-year decline outlook of 5%-10% improve or worsen.
• Whether iPhone momentum in China and weakness in premium Android persist.