US Networking & Communications Equipment: Bernstein initiates US networking and optical equipment with a positive AI-driven sector view

Overview

This initiation argues that US networking and communications-equipment companies have become structural beneficiaries of AI infrastructure spending. Bernstein expects the fastest gains to accrue to suppliers exposed to higher-speed switching, optical modules and data-center interconnect, while emphasizing differentiated company-level opportunities and risks.

Core views

Bernstein’s central thesis is that AI infrastructure remains in a higher-for-longer investment cycle. It sees near-term visibility because demand continues to exceed supply and consensus expects data-center capex to rise from about $259B in 2024 to $1.3T in 2027E. For the medium term, the report tests whether hyperscalers can support continued spending using an approximately 15% ROI hurdle; it estimates that end-user AI spending would need to reach about $730B in 2027E, roughly three times current ARR estimates. Bernstein considers this demanding but plausible given rapid adoption. Longer term, it uses continued model-performance scaling as its leading indicator: if more capable models unlock new use cases, excess capacity can be redirected toward training and demand can continue to absorb supply. It nonetheless acknowledges the risk of an eventual AI overbuild and much lower visibility beyond 2028. Within AI infrastructure, networking is identified as the third-largest spending category after compute and memory/storage. Compute performance has improved more rapidly than bandwidth: the report cites GPU FLOPs increasing 17x from H100 to VR300 while SerDes speed only doubled. This creates a “bandwidth wall,” where accelerators wait for data movement, particularly in inference workloads. Operators respond by increasing networking and memory specifications, moving to 1.6T and higher-speed switches. While prices for a fixed-speed switch generally fall over time, Bernstein expects the mix shift toward faster, higher-ASP equipment to dominate; it forecasts a 28% switch ASP jump in 2027 during the 1.6T cycle. Chiplet-adjusted accelerator volumes are expected to double from 2025-28E, with switch-port volumes growing proportionally and switch revenue outpacing accelerator shipments because of pricing. The report also expects Ethernet-based open ecosystems to take share from Nvidia’s proprietary networking technologies. It believes Arista and Celestica can gain in AI scale-out as InfiniBand is supplanted by Ethernet, and potentially in scale-up if UALink displaces NVLink. Arista’s AI back-end switching share is estimated to have risen from 3% in 2024 to 6% in 2025 and is projected to reach roughly 9% in 2026E. Bernstein forecasts 2027 Arista revenue of $18.6B versus $16.3B consensus, supported by its data-center-heavy mix and share gains. Optics is the other main structural theme. Higher data rates shorten copper’s usable reach, making optical interconnect increasingly necessary. Bernstein models transceiver revenue growing at a 40% CAGR from $57B in 2026 to approximately $300B in 2030, while laser-module supply remains constrained. This shortage is expected to preserve favorable economics for Lumentum and Coherent despite the adoption of silicon photonics. The report considers CPO yield concerns real but argues that a shift to near-packaged optics would preserve similar economics for the two companies, making a selloff on CPO-yield concerns unwarranted in its view. It prefers Lumentum to Coherent because Lumentum has purer optical exposure and greater incremental-margin potential; Coherent’s vertical integration offers more geopolitical protection but lowers incremental margins. Outside core AI clusters, Bernstein sees an incremental scale-across opportunity as land and power shortages force operators to connect separate data-center sites into one cluster. Management teams have cited an $8-10B addressable market; Bernstein estimates about $8B of new-build spending and a potential $9B retrofit opportunity. Ciena is viewed as the principal beneficiary, with Dell’Oro-estimated 70% share. Near-term supply constraints in ceramic packaging, pump lasers and memory limit its upside, but Bernstein expects availability to improve beyond 2027 and models 2029 revenue of $14B versus $13.6B consensus. Enterprise networking is also recovering cyclically, with spending up 16% from 2024 to 2025 versus a 7% 2020-25 CAGR, while stagnant service-provider spending is becoming a smaller drag. The company calls reflect this industry structure. Celestica is the top pick because Bernstein believes consensus materially understates contributions from Google TPU, OpenAI Jalapeno, AMD Helios and 1.6T switching. It sizes the potential 2027 incremental revenue opportunity at roughly $25B, comprising about $8B from Google TPU, $11.5-12B from Jalapeno and $5-6B from Helios and broader high-performance switching. After execution haircuts, it models Connectivity & Cloud Solutions revenue rising about $20B to $36B, total 2027 revenue of $40.0B and EPS of $23.79, versus consensus of $35.5B and $19.69. Its $520 target applies about 22x estimated FY2027 EPS of $24; the October 27 analyst day is identified as a catalyst. Lumentum is rated Outperform with a $1,220 target based on 35x estimated FY2028 EPS of $35, and Coherent is Outperform with a $350 target based on about 22x estimated FY2028 EPS of $16. Both calls depend on sustained transceiver pricing and operating leverage. Ciena is Outperform with a $440 target, valued at 16x estimated 2028 EPS of $28; Bernstein regards its near-term numbers as supply-capped but its longer-duration scale-across opportunity as underappreciated. Arista is Outperform with a $250 target, based on 35x estimated 2028 EPS of $7.14; valuation is described as full but consistent with its history, while earnings can compound through data-center exposure and share gains. Cisco is Market-Perform with a $110 target, based on about 20x estimated 2028 EPS of $5.60. Bernstein argues that Cisco’s pivot toward cybersecurity and observability has been poorly timed: security growth slowed to 2% in FY2026 and observability to 4%, while the company has ceded hardware share and lags software specialists. It notes possible tactical near-term upside but sees a fair valuation for an approximately 8% grower. Corning is Market-Perform with a $140 target, based on 25x estimated 2028 EPS of $5.65. The report expects strong AI optical-content growth but sees automotive and Glass Innovations as growth drags, calls Corning’s long-term framework aggressive, and judges AI upside largely priced in absent further scale-up content gains. Bernstein argues that valuation should be assessed against growth rather than by a mechanical DCF alone. It finds that AI-related beta inflates CAPM-derived cost of equity and WACC, potentially distorting DCF results, so it relies primarily on P/E multiples against growth, checked against EV/FCF, EV/EBIT and EV/revenue comparables. The report also emphasizes high stock dispersion within networking, arguing that company selection matters even with a positive sector backdrop.

Analysis framework

Bernstein begins with AI capex, adoption and model-scaling indicators, then translates the bandwidth bottleneck into switching, optical and scale-across demand. It builds industry and company forecasts using supply-chain checks, company disclosures, expert conversations and market data, compares its estimates with consensus, and sets targets mainly through growth-adjusted P/E multiples cross-checked with enterprise-value multiples and DCFs.

Methodology notes

• Industry AnalysisSupply-demand framework

  AI infrastructure supply-demand and capacity-bottleneck analysis

  The report assesses whether AI demand exceeds available compute, networking, laser and component supply, and uses bottlenecks to explain pricing, volumes and company earnings.

• Industry AnalysisVolume-price decomposition

  Switch-port volumes and higher-speed ASP mix

  Bernstein separates volume growth from price/mix effects to explain why switching and transceiver revenue may grow faster than accelerator volumes.

• Valuation methodsP/E and PEG Valuation

  P/E multiples benchmarked against earnings growth

  Price targets are primarily based on estimated EPS and multiples judged against growth profiles across hardware, networking and semiconductor peers.

• Valuation methodsDCF (Discounted Cash Flow)

  DCF used as a cross-check

  The report uses DCFs but gives them less weight because elevated AI-related beta raises CAPM cost of equity and WACC, which it believes can distort results.

Asset mapping & comparison

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

• Celestica (CLS)

  Top pick and AI rack, switching and custom-ASIC beneficiary

  Strengths

  Potential upside from Google TPU, OpenAI Jalapeno, AMD Helios and 1.6T switching; October 27 analyst day catalyst.

  Weaknesses

  Complex new-program ramps create execution uncertainty.

  Comparison

  Bernstein sees its estimates materially above consensus and views valuation as undemanding.

  Risks

  ODM/EMS competition, Nvidia share gains, AI-demand deceleration and delays in Jalapeno, TPU or Tomahawk 6 ramps.

• Lumentum (LITE)

  Outperform optical-module beneficiary

  Strengths

  Pure optical exposure, laser-module economics and stronger expected operating leverage.

  Weaknesses

  Higher sensitivity to the AI optical cycle.

  Comparison

  Preferred over Coherent because of cleaner exposure and higher incremental margins.

  Risks

  CW+SiPh adoption could weaken InP scarcity; Chinese export restrictions could disrupt indium-phosphide supply.

• Coherent (COHR)

  Outperform optical-module beneficiary

  Strengths

  Vertical integration, transceiver exposure and operating leverage from growth.

  Weaknesses

  Its industrial business represented about 26% of revenue and lowers AI exposure and incremental margins versus Lumentum.

  Comparison

  Less preferred than Lumentum, though vertical integration offers greater geopolitical protection.

  Risks

  Optical demand, pricing or supply-chain conditions could weaken; Chinese transceiver restrictions could trigger retaliatory input restrictions.

• Ciena (CIEN)

  Outperform scale-across and optical-transport beneficiary

  Strengths

  Estimated 70% scale-across share, technology leadership and longer-term supply improvement.

  Weaknesses

  Ceramic packaging, pump-laser and memory bottlenecks cap near-term growth.

  Comparison

  Bernstein sees consensus as overly discounting its longer-term growth framework.

  Risks

  Easier land and power availability could reduce scale-across need; supply-chain execution may remain constrained.

• Arista Networks (ANET)

  Outperform AI switching share gainer

  Strengths

  Data-center-heavy mix, Ethernet transition and projected share gains in scale-out and potentially scale-up.

  Weaknesses

  Valuation is characterized as somewhat full.

  Comparison

  Expected to materially outgrow industry growth and consensus revenue estimates.

  Risks

  Broadcom Tomahawk supply constraints and intensifying ODM white-box competition could pressure demand or pricing.

• Cisco (CSCO)

  Market-Perform networking incumbent

  Strengths

  Potential tactical upside from conservative guidance, Acacia traction and enterprise exposure.

  Weaknesses

  Software businesses are decelerating, hardware share has been ceded, and cybersecurity capability is seen as lagging specialists.

  Comparison

  Lower AI-networking exposure may cushion downside in a digestion cycle but limits upside relative to preferred names.

  Risks

  Further software deceleration and share losses; upside depends on stronger Acacia traction or AI switching share gains.

• Corning (GLW)

  Market-Perform optical-fiber supplier

  Strengths

  AI accelerator optical-content growth and a diversified business mix.

  Weaknesses

  Automotive and Glass Innovations are growth drags; capital intensity limits upside from higher growth.

  Comparison

  Bernstein sees optical upside as more fully priced and its long-term growth framework as aggressive.

  Risks

  AI adoption could slow; upside would require faster optical adoption, improved CPO yields or stronger smartphone and automotive demand.

Key data

• Industry growth forecast18% growth in 2026-28Includes 33% cloud and AI growth, 9% enterprise growth and 4% service-provider growth.
• AI data-center capex$1.3T in 2027EConsensus expectation, up from about $259B in 2024.
• Accelerator volumesDouble from 2025 to 2028EChiplet-adjusted volumes; switching revenue is expected to outgrow them.
• Switch ASP change28% increase in 2027Driven by the 1.6T speed transition before ASP growth moderates.
• Transceiver-market forecast$57B in 2026 to about $300B in 2030Approximately 40% CAGR in Bernstein's model.
• Scale-across opportunity$8-10B management estimate; about $9B retrofit opportunityBernstein considers management sizing potentially conservative.
• Celestica 2027 forecast$40.0B revenue and $23.79 EPSVersus consensus of $35.5B revenue and $19.69 EPS.

Impact & implications

The report concludes that AI networking and optics can support above-consensus revenue and earnings for selected suppliers as bandwidth constraints force higher-speed, higher-value equipment. It favors companies with direct exposure to switching share gains, optical scarcity or scale-across demand, but regards generic AI exposure alone as insufficient because valuation, execution and competitive positioning vary sharply.

Risks

• AI adoption could decelerate, reducing AI infrastructure investment and revenue for AI-exposed suppliers.
• More intense ODM, EMS or white-box competition could lower win rates, market share and pricing power.
• Greater adoption of CW+SiPh lasers could ease the InP bottleneck and weaken Lumentum and Coherent pricing power.
• Chinese retaliation against US transceiver restrictions could restrict indium-phosphide inputs and disrupt optical supply.
• Continued CPO yield problems could delay optical adoption.
• Easing land and power constraints could reduce the scale-across opportunity for Ciena.
• Supply constraints, including Broadcom Tomahawk availability, could limit switching demand capture.

What to watch

• AI capex guidance and revisions in the October earnings season.
• Celestica's October 27 analyst day and its 2027 guidance.
• Evidence that model scaling, AI adoption and hyperscaler spending remain intact.
• The pace of 1.6T switching deployment and open Ethernet/UALink adoption.
• Optical-module pricing, laser capacity and CPO/NPO manufacturing progress.
• Ciena supply availability for ceramic packaging, pump lasers and memory.
• March 2027 catalysts including OFC on March 7-11 and Nvidia GTC on March 14-18.