NVDA earnings beats reinforce AI semiconductor momentum, with Asian supply chains still the primary beneficiaries

Overview

This report focuses on NVDA's 1QFY27 results and 2QFY27 guidance and distills their implications for Asian technology and the AI semiconductor supply chain. The core conclusion is that data center demand, GB300 orders, networking revenue, and AI infrastructure capex all point to an ongoing uptrend, supporting beneficiaries such as TSMC, ASE, Alchip, Unimicron, Samsung Electronics, SK Hynix, and Hon Hai across Asia; however, China revenue uncertainty, tight capacity in advanced process nodes and CoWoS, competition from ASICs versus GPUs, and the sustainability of AI infrastructure financing are the key risks ahead.

Core views

The report's core views are as follows: first, both NVDA's April quarter results and July quarter guidance were better than expected, with data center revenue growth as the primary driver. Second, GB300 demand remains strong, and JPMorgan recently raised its 2026 estimate for NVDA CoWoS consumption by 7%, mainly reflecting stronger Blackwell GB300 orders. Third, Vera Rubin is expected to start shipping in 3Q26 and ramp in 4Q26 as planned, but VR racks may still account for only a small share in 2026. Fourth, networking is growing faster than compute, and demand for NVLink, Spectrum-X, Quantum, CPO, and Neocloud may continue to sustain revenue momentum. Fifth, the narrative that GPUs are gaining share in inference still needs more evidence to prove that NVDA's growth can outpace ASICs in 2027, especially as Rubin's competitiveness versus TPU v8 Sunfish and Trainium 3 is watched closely. Sixth, overall AI semiconductor sentiment is likely to remain supportive, with the main beneficiaries concentrated in the Asian AI semiconductor supply chain.

Analysis framework

The report combines earnings interpretation, management guidance, supply chain validation, and capex forecasting: it first compares NVDA's revenue, data center, and networking performance with JPMe and market consensus, then combines checks on GB300, Vera Rubin, CoWoS, advanced process nodes, GPU rental prices, CSP capex, and supply chain shipments to derive the impact on the Asian semiconductor supply chain.

Methodology notes

• Earnings and guidance comparisonOutperformance vs. expectations analysis

  Compare NVDA's actual revenue and management guidance with JPMe and market consensus.

  This approach is used to determine whether the results constitute a positive catalyst. The report points out that April quarter revenue was about 3% above JPMe and market consensus, while July quarter revenue guidance of $91.0 billion was about 4% above market expectations.

• Supply chain validationUpstream shipment and capacity checks

  Validate management commentary using supply chain data such as GB300, Vera Rubin, CoWoS, advanced process nodes, and CPU shipments.

  The report uses feedback from the Asian supply chain to assess demand and supply constraints, such as about 2 million Rubin upstream shipments in 2026, 60,000 to 70,000 NVL72 racks, about 3 million to 5 million Grace and Vera CPUs, and a CoWoS unmet demand gap of about 15% to 20%.

• Competitive landscapeGPU vs. ASIC growth comparison

  Compare the growth outlook of NVDA GPUs versus ASIC paths such as TPU and Trainium in inference and training.

  The report believes NVDA is gaining share in inference, but ASIC supply chains may still grow faster in 2027 from a lower base; the market needs more evidence on Rubin model performance and tokens-per-watt.

• Capex cycleTracking AI infrastructure build intensity

  Infer the persistence of semiconductor demand from CSP, Oracle, Neocloud, sovereign AI, and enterprise AI capex.

  The report cites the US team's forecast that total capex for the top four US CSPs has been revised up to more than 70% YoY after the latest earnings reports, and total capex for the top five US CSPs in 2027 may approach $1 trillion.

Asset mapping & comparison

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

• NVIDIA CORP（NVDA.US）

  The core company discussed and a bellwether for AI infrastructure demand

  Strengths

  High data center revenue growth, July quarter guidance above expectations, strong GB300 demand, networking growth faster than compute, and an expanding inference ecosystem and partner base.

  Weaknesses

  China data center revenue is excluded from guidance due to uncertainty around H200 imports; supply constraints may persist throughout the Vera Rubin lifecycle; the 2027 growth advantage of GPUs versus ASICs still needs to be proven.

  Comparison

  Relative to TPU, Trainium, and other ASIC routes, NVDA emphasizes inference share gains and ecosystem advantages, but the report believes ASIC supply chains may still grow faster in 2027 from a lower base.

  Risks

  Supply bottlenecks, tight advanced process and CoWoS capacity, restrictions in China, ASIC competition, and AI infrastructure financing pressure.

• TSMC（2330.TW）

  A key beneficiary of advanced process nodes and CoWoS

  Strengths

  Benefits from GB300, Rubin, advanced process nodes, and the long-term migration of LPUs to TSMC; N3 expansion and accelerated N2 ramp support long-term demand.

  Weaknesses

  Utilization at N5 and below remains above 100%, and tight capacity may limit near-term shipments.

  Comparison

  Compared with OSAT partners, TSMC benefits from both leading-edge wafer fabrication and CoWoS, but also faces more pronounced bottlenecks at advanced nodes.

  Risks

  Expansion pace lagging demand, CoWoS clean room constraints, and changes in outsourcing cadence for non-GPU/ASIC projects.

• ASE Technology Holding（3711.TW）

  An Asian beneficiary of AI semiconductor packaging and OSAT demand

  Strengths

  The report lists it as a continued beneficiary of the AI supply chain, potentially benefiting from advanced packaging and outsourcing demand.

  Weaknesses

  Specific orders and earnings leverage are not quantified in the text.

  Comparison

  Compared with TSMC, ASE is more exposed to the OSAT segment and benefits from spillover in advanced packaging and supply chain division of labor.

  Risks

  Packaging demand timing, customer outsourcing ratios, and utilization volatility.

• Alchip Technologies（3661.TW）

  A beneficiary of ASIC and AI chip design services

  Strengths

  The report remains constructive on it as a beneficiary of the AI semiconductor supply chain; if ASICs continue to grow faster in 2027, the related design services chain has upside.

  Weaknesses

  The report does not provide new company-level financial forecasts.

  Comparison

  Relative to the NVDA GPU chain, Alchip is more sensitive to the ASIC growth narrative.

  Risks

  ASIC project timelines, customer concentration, and shifts in the GPU versus ASIC competitive landscape.

• Unimicron（3037.TW）

  A beneficiary of AI server and high-end substrate demand

  Strengths

  Listed as one of the favored beneficiaries in the AI semiconductor supply chain, benefiting from AI servers and high-bandwidth interconnect demand.

  Weaknesses

  The report does not disclose specific order or capacity data.

  Comparison

  Unlike OSAT and foundry segments, Unimicron is more exposed to the substrate and interconnect materials chain.

  Risks

  Substrate pricing, capacity expansion, and fluctuations in AI server demand.

• Samsung Electronics（005930.KS）

  An early foundry partner for LPU production and a participant in the memory supply chain

  Strengths

  The report says LPU production will start with Samsung's foundry, and it may benefit as a key foundry partner.

  Weaknesses

  Over the long term, production may move to TSMC starting around the LP40 generation, leaving long-term share uncertain.

  Comparison

  It has a near-term role in the LPU foundry segment, but longer term it may be reallocated in favor of TSMC.

  Risks

  Slow LPU commercialization, migration to TSMC, and foundry execution risk.

• SK Hynix（000660.KS）

  A beneficiary of AI memory and HBM demand

  Strengths

  The report continues to list it as a beneficiary of the AI semiconductor supply chain, supported by AI compute and memory demand.

  Weaknesses

  This report does not expand on company-level HBM pricing or share details.

  Comparison

  Compared with foundry and packaging companies, SK Hynix is more driven by AI memory demand and the HBM cycle.

  Risks

  Memory cycle volatility, customer concentration, and intensifying competition.

• Hon Hai Precision（2317.TW）

  An AI server ODM and key ODM partner for LPU

  Strengths

  The report says Hon Hai is a key ODM partner for LPU and lists it as a beneficiary of the AI supply chain.

  Weaknesses

  VR rack share in 2026 may still be small, and near-term uplift depends on the production ramp.

  Comparison

  Compared with chips and packaging, Hon Hai is more directly tied to server system assembly and ODM demand.

  Risks

  AI server order timing, margins, and any slowdown in customer capex.

Key data

• April quarter NVDA revenue beatAbout 3%Above JPMe and market consensus.
• Data center revenue growth+92% YoY, +21% QoQThe main source of growth in the April quarter.
• July quarter revenue guidance$91.0 billionAbout 4% above market expectations, excluding China data center revenue.
• 2026 NVDA CoWoS consumption estimate revision+7%Mainly reflects stronger Blackwell GB300 orders.
• NVDA networking revenue$15.0 billion, about 3x YoYApril quarter networking revenue grew faster than the 77% YoY growth rate in the compute business.
• 2026 NVL72 rack shipment estimate60,000 to 70,000 unitsMost are expected to be GB300, with a small share from VR racks.
• 2026 Rubin upstream shipment estimateAbout 2 million unitsThe report says this is consistent with supply chain feedback.
• CPU long-term TAM target$200 billionThe long-term CPU market size given by NVDA management.
• 2026 standalone CPU revenue target$20 billionThe report believes this is above market consensus, but additional capacity validation is needed.
• 2026 capex growth for the top four US CSPsAbout 70% YoYHigher than the prior 60% to 65% expectation.
• 2027 total capex for the top five US CSPsAbout $1 trillionConsistent with NVDA's comments on the intensity of AI infrastructure buildout.
• Long-term outlook for 2030 AI infrastructure capex$3 trillion to $4 trillionNVDA's more aggressive long-term outlook implies that high double-digit growth must still be sustained through 2029/2030.
• Change in GPU rental pricesH100 about +20% YTD, A100 about +15% YTDReflects tight supply and demand in AI infrastructure.
• Global semiconductor revenue growth forecastAbout +18% in 2026, about +11% in 2027The report expects the AI upcycle to last at least through 2027.
• CoWoS unmet demand gapAbout 15% to 20%Constrained by clean room capacity and strong demand.

Impact & implications

In investment terms, the report strengthens the medium-term view that the AI semiconductor supply chain remains in an upcycle. Signals from NVDA's data center, networking, GB300, Vera Rubin, and AI infrastructure capex all support continued earnings upgrades across the Asian tech chain, especially for companies tied to advanced foundry services, advanced packaging, OSAT, ABF substrates, HBM/memory, ODMs, and ASIC design services. At the same time, investors should not focus solely on NVDA's standalone results; they also need to monitor CSP free cash flow, financing capacity, the expansion of advanced process nodes and CoWoS supply, Rubin's performance versus ASICs, and whether China data center revenue can recover.

Risks

• China data center revenue is affected by uncertainty around H200 imports, and the July quarter guidance has already excluded the related revenue.
• Supply constraints for Vera Rubin may persist until around 2028.
• Utilization at N5 and below remains above 100%, and the N3 expansion and N2 ramp still need time.
• CoWoS faces an unmet demand gap of about 15% to 20%, constrained by clean room capacity and strong demand.
• The 2027 growth advantage of GPUs over ASICs has not yet been fully proven, and TPU v8 Sunfish and Trainium 3 may create competitive pressure.
• AI infrastructure buildout may be constrained by CSP free cash flow burn and financing difficulties.
• The current addressable market for LPU is narrow, and the near-term attach rate may remain low.

What to watch

• Whether NVDA's data center revenue continues to beat expectations, especially whether China-related revenue can recover.
• Whether GB300 orders, CoWoS consumption, and NVL72 rack shipments continue to exceed expectations.
• Whether Vera Rubin starts shipping in 3Q26 and ramps production in 4Q26 as planned.
• Performance evidence for Rubin models and tokens-per-watt relative to TPU v8 Sunfish and Trainium 3.
• The pace of CPO switch adoption in scale-out and scale-up GPU fabrics.
• Whether capex for US CSPs, Oracle, Neoclouds, sovereign AI, and enterprise AI continues to grow strongly.
• Whether GPU rental prices such as H100 and A100 continue to rise or begin to ease as supply and demand normalize.
• Progress on TSMC advanced process nodes, CoWoS, OSAT outsourcing, and VSMC interposer/silicon bridge cooperation.