NVIDIA beats expectations, and AI infrastructure bottlenecks spill over into Taiwan supply chain

Overview

This report is Citigroup's quick take on NVIDIA's earnings call, focusing on the implications of NVIDIA's strong results for Taiwan's semiconductor, electronic components, and equipment supply chain. NVIDIA's revenue reached US$81.6bn, and it guided next-quarter revenue to about US$91bn, well above expectations. Citigroup believes AI infrastructure buildout remains in a high-intensity expansion phase, and NVIDIA is evolving from a pure GPU supplier into an AI factory company, with growth now extending to networking, NVLink fabric, and rack-scale systems.

Core views

Citigroup's core view is that the AI infrastructure bottleneck is no longer limited to GPU wafers, but has expanded to full data center execution capacity, including HBM, advanced packaging, optical interconnects, liquid cooling, power, networking, and rack integration. NVIDIA's roughly US$119bn in inventory and capacity commitments indicate that it is more proactively locking up upstream resources and has greater visibility into 2026-2027 supply allocation and deployment plans. This trend is positive for Taiwan's AI hardware ecosystem, especially advanced packaging, backend testing, substrates, thermal management, power, and ODM rack integration.

Analysis framework

The report uses an earnings-call interpretation and supply-chain mapping approach: first it analyzes NVIDIA's revenue, guidance, inventory, and capacity commitments, along with management's remarks on the Blackwell and Rubin product cadence; then it breaks down the AI infrastructure bottleneck into GPU, HBM, CoWoS, substrates, optical interconnects, liquid cooling, power, and rack integration; finally it maps those needs to Taiwan-listed supply-chain companies to identify potential beneficiaries.

Methodology notes

• Industry chain analysisAI infrastructure bottleneck migration framework

  From GPU shortages to data center system execution bottlenecks

  The report argues that the current constraint is not just GPUs, but the combined effect of HBM, advanced packaging, optical interconnects, liquid cooling, power, and rack integration on AI cluster delivery capability.

• Supply chain mappingTaiwan AI hardware supply chain beneficiary chain

  Map NVIDIA's product cycle to Taiwanese suppliers

  Through the mass production, packaging, testing, cooling, power, and ODM system delivery needs of Blackwell and Rubin, the report identifies beneficiaries such as TSMC, ASEH, KYEC, Winway, Delta, AVC, Auras, Hon Hai, Quanta, and Wistron.

• Product cycle analysisObserving the generational transition from Blackwell to Rubin

  Product transition risk and production ramp cadence

  The report believes Blackwell shipments are strong, Vera Rubin samples have already shipped, and the subsequent production plan is relatively stable, which reduces market concerns about product delays and rack integration issues.

Asset mapping & comparison

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

• NVIDIA CORP (NVDA.US)

  Earnings and guidance are the main triggers for this report

  Strengths

  Revenue and guidance beat expectations, AI infrastructure demand remains strong, Blackwell shipments are ramping, Rubin samples have shipped, and supply-chain visibility is improving.

  Weaknesses

  AI system deployment complexity is increasing, which raises dependence on the execution capability of the full data center stack.

  Comparison

  The company is extending its positioning from a GPU supplier toward an AI factory platform company, covering GPUs, CPUs, interconnects, switches, software, and rack-scale systems.

  Risks

  Product transitions, liquid cooling and rack integration, supply-chain bottlenecks, and competition from ASICs and cloud providers' in-house chips.

• TSMC (2330.TW)

  A key supplier of NVIDIA's advanced-node and CoWoS capacity

  Strengths

  It has a strategic position in Blackwell and Rubin advanced-node production and CoWoS expansion.

  Weaknesses

  Capacity expansion and advanced packaging cadence need to match AI demand.

  Comparison

  Compared with other suppliers, TSMC sits at the most critical wafer manufacturing and advanced packaging stage.

  Risks

  Risks from advanced packaging capacity, customer demand volatility, and product transition timing.

• Hon Hai Precision (2317.TW)

  A potential major beneficiary of AI rack-scale systems and ODM integration

  Strengths

  The report sees Hon Hai as a hub supplier as AI deployment shifts toward complete rack-scale systems.

  Weaknesses

  It faces high requirements for large-scale rack integration, validation, and delivery execution.

  Comparison

  It sits in the ODM chain alongside Quanta and Wistron, but the report places stronger emphasis on Hon Hai's hub role.

  Risks

  Rack integration complexity, customer schedule changes, and delivery execution risk.

• Delta Electronics (2308.TW)

  A beneficiary of AI rack power and cooling upgrades

  Strengths

  Rising AI rack power density and greater liquid cooling adoption are supportive of power and thermal management demand.

  Weaknesses

  Demand depends on the pace of AI data center buildout and customer capex.

  Comparison

  Compared with pure assemblers, Delta has more direct exposure to power and thermal bottlenecks.

  Risks

  Changes in liquid cooling solutions, intensifying competition, and data center deployment delays.

• ASE Technology Holding (3711.TW), King Yuan Electronics (2449.TW), WinWay Technology (6515.TW)

  Beneficiaries of advanced packaging, backend testing, and test-interface ecosystems

  Strengths

  Higher advanced packaging intensity and rising AI chip complexity are favorable for backend testing and related suppliers.

  Weaknesses

  Business performance is affected by the production schedules of NVIDIA and major AI chip customers.

  Comparison

  They sit in the packaging and testing chain after TSMC, so the benefit is more concentrated in back-end processes.

  Risks

  Advanced packaging capacity timing, testing demand volatility, and customer concentration.

• Asia Vital Components (3017.TW), Auras Technology (3324.TWO)

  Beneficiaries of liquid cooling and thermal management

  Strengths

  Higher AI rack power density drives demand for liquid cooling and thermal management systems.

  Weaknesses

  Product specifications change quickly, requiring them to keep pace with customer platform upgrades.

  Comparison

  Compared with power suppliers and ODMs, cooling vendors benefit more directly from higher liquid-cooling penetration.

  Risks

  Technology route changes, price competition, and customer adoption timing.

• Quanta Computer (2382.TW), Wistron (3231.TW)

  Beneficiaries of AI servers and rack ODM demand

  Strengths

  The report believes both companies will also benefit as AI deployment expands to rack-scale systems.

  Weaknesses

  Delivery complexity and customer project cadence may create volatility.

  Comparison

  They are in the same ODM chain as Hon Hai, but the report expresses a stronger view on Hon Hai's hub position.

  Risks

  Changes in order allocation, system validation difficulty, and supply-chain constraints.

Key data

• NVIDIA reported revenueUS$81.6bnThe report says NVIDIA beat expectations again.
• Next-quarter revenue guidanceabout US$91bnAbove market expectations.
• Inventory and capacity commitmentsabout US$119bnShows that NVIDIA is actively reserving capacity for chips, HBM, CoWoS, substrates, optics, and system components.
• Supply-chain visibility2026-2027Citigroup believes NVIDIA has better visibility on supply allocation and deployment plans.
• Rubin production cadenceRubin-series chip production started in 1Q26, and larger-scale rack ramp is expected to proceed as planned in 4Q26The report believes the product schedule is more stable than the market feared.

Impact & implications

If NVIDIA's demand strength and capacity lock-ins continue, the beneficiaries across Taiwan's AI supply chain will expand from advanced process and CoWoS into backend testing, substrates, optical interconnects, cooling, power, and ODM system delivery. Citigroup is especially constructive on Hon Hai and Delta as downstream technology suppliers, because AI deployment is shifting toward complete rack-scale systems and data center infrastructure execution. TSMC remains strategically critical because of its advanced-node and CoWoS expansion, ASEH, KYEC, and Winway benefit from higher advanced packaging intensity, AVC and Auras benefit from liquid cooling and higher power density trends, and Quanta and Wistron are also likely to benefit from AI server and rack demand.

Risks

• The AI infrastructure bottleneck may shift from GPUs to HBM, advanced packaging, optical interconnects, liquid cooling, power, and rack integration; constraints in any one link could affect delivery.
• If Blackwell or Rubin generational transitions run into thermal design, rack integration, or system validation issues, supply-chain ramp-up could be delayed.
• Competition from ASICs and cloud providers' in-house chips could weaken NVIDIA's long-term pricing power or alter supply-chain share.
• If AI data center capex comes in below expectations, visibility for Taiwan supply-chain orders will be affected.
• The report also notes general investment risks such as overseas securities exposure, exchange rates, liquidity, disclosure, and market volatility.

What to watch

• NVIDIA's future quarterly revenue guidance and Blackwell system shipment progress.
• Whether Vera Rubin samples, mass production, and 4Q26 rack ramp-up proceed as planned.
• How NVIDIA's roughly US$119bn in inventory and capacity commitments translate into orders for HBM, CoWoS, substrates, optics, and system components.
• TSMC's CoWoS expansion cadence and advanced-node capacity allocation.
• Changes in power and liquid cooling orders tied to higher AI rack power density for Delta, AVC, and Auras.
• Hon Hai, Quanta, and Wistron's order share and delivery capability in NVIDIA's rack-scale systems.