Goldman Sachs sees AI infrastructure demand supporting a multi-year semiconductor upcycle, with capacity constraints remaining a key bottleneck.

Overview

This conference-takeaways report synthesizes management commentary from US semiconductor and technology companies. Goldman Sachs presents a constructive industry view centered on AI infrastructure demand, constrained supply and expanding equipment, storage and connectivity content, while noting a more difficult near-term backdrop for IT services.

Core views

The report's central conclusion is that AI-infrastructure spending remains a powerful multi-year driver for digital semiconductors. Nvidia reiterated its estimate of $3tn-$4tn of AI-infrastructure spending in 2030, supported by the transition from retrieval-based to generative computing. Management argued that slower Moore's Law increases the value of tightly co-designed hardware, networking and software systems, including NVLink and scale-up/scale-out architectures. The report notes that both upstream inputs—advanced packaging, wafers and memory—and downstream infrastructure—power, sites and datacenter capacity—remain constrained; Goldman Sachs believes downstream constraints may be more consequential at the margin. Nvidia nonetheless reaffirmed an expectation for approximately 70% year-on-year growth in CY27. Its management also characterized financial guarantees as ecosystem-seeding support rather than circular financing, and highlighted autonomous vehicles, warehouse automation, navigation, reasoning-based manipulation systems expected roughly two years away, and 6G as Physical AI opportunities. Across semiconductor capital equipment, foundry and storage, companies described datacenter buildouts as a structural growth engine, with sharply higher orders and backlog producing unusually strong visibility. Goldman Sachs believes commentary that 8-10 new fabs could come online in 2027 supports accelerating wafer-fab-equipment demand, though companies did not provide aggregate 2027 WFE guidance. Lam expects DRAM to be the largest 2027 WFE growth driver, followed by leading-edge foundry/logic and NAND. It sees GAA, backside power, advanced packaging and increasingly 3D memory as technology inflections that lift process-equipment content; each additional 100,000 wafer starts could expand Lam's serviceable addressable market by roughly $1bn. Cleanroom availability remains a constraint through end-2027, but customer discussions are informing production and supplier investment plans. Lam also expects differentiated mix, local manufacturing and R&D, and pricing to move gross margins toward the mid-50% range over the next several years. Qnity highlighted advanced packaging and interconnect as near-term growth drivers, alongside recovery in trailing-edge nodes. Advanced-node sales grew 20% in the prior quarter, and the migration to 3 nm is viewed as a tailwind. Management expects sequential Semiconductor Technologies growth through the second half and into 2027. Its Interconnect Solutions segment benefits from advanced packaging and AI PCBs, which account for about 30% of segment activity; Qnity guided to mid-single-digit quarter-on-quarter growth in 3Q26. While mix and input costs pressured gross margin relative to expectations in the prior quarter, management expects steady expansion, potentially helped by Interconnect Solutions mix and operating leverage, toward the 50% range over the long term. Storage commentary was also constructive. Seagate links AI-infrastructure demand and limited supply growth to continued price improvement and higher earnings through FY27. Price per exabyte rose about 11% year on year last quarter, and management expects momentum to continue. HAMR volumes are expected to exceed PMR on an exabyte basis by end-CY26 and reach 80%-90% of bit volume within a few years. Management expects sequential revenue and EPS gains through FY27, while Goldman Sachs expects at least 70% incremental gross margins, supported by HAMR, internal-laser cost benefits and pricing. The company plans further high-yield debt retirement and accelerated share repurchases, but remains cautious on supply additions because eventual datacenter bottlenecks could affect demand. For interconnect, Credo expects copper-based AEC solutions to remain the preferred, reliable solution for short links in the medium term, while optics gains share at longer distances. Its optical growth plan includes PICs and ZF optical transceivers aimed at AI-network reliability, expanding AEC deployments and prospective ALC contributions. ALCs are designed to extend reach from about 7m to about 30m with comparable reliability while reducing cable volume by about 75% versus AECs; launch is expected at OCP next month, followed by customer qualification and FY28 volume production. Management guided to about 20% quarter-on-quarter revenue growth in FY3Q and about 30% in FY4Q, expects FY27 gross margin to match FY26, and expects OpEx to grow about 55% year on year before normalizing. EPAM was the principal weaker-demand thread. Management expects muted discretionary spending to persist for the balance of 2026, citing cautious clients, pressure in certain verticals, oil prices, conflict in the Middle East and US sales-execution shortfalls behind its downward 2026 guidance revision. AI-native services remain a source of traction, generating more than $160mn of sales in 2Q and expected to generate at least $600mn in 2026, but clients often seek agentic automation within existing budgets. EPAM expects gross-margin support from pricing and utilization, but forecasts FY27 non-GAAP operating margin of 15%-16%; limited external AI-native talent also requires internal retraining and upskilling.

Analysis framework

Goldman Sachs organizes conference management commentary into sector-wide debates and company-specific implications. It connects AI demand, supply constraints and technology transitions to equipment, storage and interconnect demand, then summarizes management guidance, operating-margin expectations, capital allocation and company valuation frameworks.

Methodology notes

• Industry AnalysisSupply-demand framework

  AI infrastructure demand versus upstream component and downstream datacenter-capacity constraints

  The report uses supply and demand conditions to explain why AI spending can support orders, pricing and investment visibility while capacity bottlenecks may limit deployment.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  Transmission from AI compute spending into wafers, packaging, equipment, storage and interconnect

  The report traces how higher compute demand increases content and demand across semiconductor supply-chain layers.

• Valuation methodsP/E and PEG Valuation

  P/E-based target-price frameworks for Qnity, Nvidia, Lam, Seagate and Credo

  The cited targets apply specified earnings multiples to normalized EPS estimates or forecasts.

Asset mapping & comparison

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

• Qnity (Q)

  Advanced packaging, interconnect and node transitions are cited as earnings drivers.

  Strengths

  Advanced-node momentum, trailing-edge recovery, Interconnect Solutions growth and expected margin expansion.

  Weaknesses

  Recent gross-margin pressure from mix and input costs.

  Risks

  Slow wafer starts across served end-markets, tariffs/export restrictions and PFAS litigation liabilities.

• Nvidia (NVDA)

  Primary beneficiary of AI compute infrastructure and Physical AI adoption.

  Strengths

  Co-designed hardware, software and networking stack; ecosystem scale and customer visibility.

  Comparison

  Management argues broader system value is increasing its share of AI capital expenditure.

  Risks

  AI-infrastructure spending slowdown, competitive share or margin erosion, and supply constraints.

• Lam Research (LRCX)

  Equipment demand benefits from DRAM, leading-edge logic, NAND, 3D architectures and packaging.

  Strengths

  Technology-content growth, services momentum, differentiated mix and improving pricing.

  Weaknesses

  China capital expenditure is growing more slowly than other regions.

  Comparison

  DRAM is expected to lead 2027 WFE growth, ahead of foundry/logic and NAND.

  Risks

  NAND upgrade-spending pause, tighter US export restrictions and a shift toward lithography intensity.

• Seagate Technology (STX)

  AI-driven data usage and disciplined supply support storage pricing and HAMR adoption.

  Strengths

  HAMR transition, pricing tailwinds, internal-laser cost benefits and capital returns.

  Weaknesses

  Demand could be affected if datacenter customers encounter other bottlenecks.

  Comparison

  HAMR is expected to overtake PMR by end-CY26.

  Risks

  HAMR qualification difficulties at major hyperscalers and insufficient supply discipline.

• Credo Technology Group (CRDO)

  AI-network densification supports AEC demand while optics expands at longer reach.

  Strengths

  Optical portfolio, vertical integration, reliability-focused solutions and AEC momentum.

  Weaknesses

  Revenue can be lumpy because of customer concentration.

  Comparison

  Copper is expected to remain preferred for short links, while optics gains share at longer distances.

  Risks

  Faster-than-expected optical substitution of copper, AEC competition and customer concentration.

• EPAM Systems Inc. (EPAM)

  AI-native services provide growth potential amid weak discretionary IT spending.

  Strengths

  AI-service traction, pricing/utilization support and record-low attrition.

  Weaknesses

  Muted client spending, sales-execution shortfalls and scarce AI-native talent.

  Risks

  Further demand slowdown, dilutive M&A and increased geopolitical tensions.

Key data

• Nvidia AI infrastructure opportunity$3tn-$4tn by 2030Management's reiterated AI infrastructure spending/TAM outlook.
• Nvidia CY27 growth expectation~70% YoYReaffirmed despite supply tightness.
• Potential new fabs8-10 in 2027Goldman Sachs views this commentary as supportive of WFE acceleration.
• Lam SAM expansion~$1bn per incremental 100k wafer startsManagement estimate tied to GAA and backside-power technology inflections.
• Seagate price per exabyte~11% YoY increase last quarterManagement cited pricing improvement supporting FY27 momentum.
• Seagate HAMR bit-volume mix80%-90% in the next few yearsHAMR is expected to surpass PMR by end-CY26.
• Credo FY3Q/FY4Q revenue growth guidance~20% QoQ / ~30% QoQManagement's near-term growth outlook.
• EPAM AI-native services sales>$160mn in 2Q; at least $600mn in 2026AI-services traction despite broader discretionary-spending weakness.

Impact & implications

The report argues that AI buildouts are broadening demand beyond accelerators into semiconductor equipment, advanced packaging, storage and connectivity, with tight capacity supporting visibility and pricing. It contrasts this with IT-services demand, where AI adoption is growing but has not yet offset weak discretionary budgets.

Risks

• AI infrastructure demand could slow or be constrained by shortages of packaging, wafers, memory, power, sites and datacenter capacity.
• Export restrictions, tariffs and competition could pressure semiconductor companies discussed in the report.
• Datacenter bottlenecks and undisciplined storage supply additions could weaken Seagate's demand or pricing outlook.
• EPAM faces continued discretionary-spending weakness, talent constraints and geopolitical risks.