Lightelligence Chairman Visit: AI cluster upgrades drive opportunities in optical interconnect and optical computing

Overview

This report is a meeting note from Goldman Sachs after hosting the chairman of Lightelligence during its China Tech Tour in Shanghai on 22 May 2026. The discussion focused on scale-up expansion, the company's competitive advantages, and opportunities in optical computing. The report argues that AI infrastructure upgrades, migration toward supernodes, and the development of higher-integration forms such as NPO/CPO are driving demand for optical interconnect and optical computing solutions.

Core views

The core view is that rising requirements for higher bandwidth, lower latency, and lower-power interconnects in AI clusters will expand the addressable market for optical interconnect in scale-up and scale-out architectures. As an early entrant in optical computing and optical interconnect, Lightelligence has accumulated experience in complex optoelectronic chip design, PIC/EIC design, co-packaging, and commercialization partnerships, and may benefit from the rise in AI networking opportunities.

Analysis framework

The report is mainly based on management discussion notes and industry read-across, with qualitative analysis centered on product architecture, technical pathways, commercialization capabilities, and AI infrastructure demand. Its emphasis is not on financial forecasts or valuation models, but on mapping the potential growth direction of the optical networking industry through the company's products and management views.

Methodology notes

• Industry MappingRead-across to optical industry

  Derive optical networking opportunities from company management's views

  The report maps Lightelligence's positive statements on scale-up and scale-out opportunities to Goldman Sachs' view that AI infrastructure expansion, rising rack-level compute density, and a larger optical-networking TAM are expanding.

• Technology Architecture AnalysisScale-up / Scale-out architecture review

  AI cluster interconnect architecture upgrade

  The analysis centers on technical directions such as AI servers, GPU supernodes, inter-rack interconnects, reduced optical-electrical conversion, and highly integrated NPO/CPO solutions.

• Disclosure FrameworkGS Factor Profile / M&A Rank / Quantum

  Goldman Sachs standard disclosure and research tool description

  The body is followed by Goldman Sachs' general disclosures on factor profiles, M&A ranking framework, and the Quantum database, but the report does not assign 1879.HK a specific factor score, M&A rating, or valuation conclusion.

Asset mapping & comparison

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

• Guangzhi Technology / Lightelligence (1879.HK)

  The report's subject company, providing optical interconnect and optical computing products

  Strengths

  Early entrant in optical computing and optical interconnect; has complex optoelectronic chip design, PIC/EIC design, and co-packaging capabilities; works with foundry, OSAT, and EDA partners to support commercialization and large-scale shipments.

  Weaknesses

  The report does not provide quantitative financial data such as revenue, profit, orders, customer validation progress, or mass-production scale; Goldman Sachs does not cover the stock.

  Comparison

  Compared with traditional electrical interconnect or electrical packet-switching solutions, the company's optical interconnect and optical circuit-switching solutions aim to reduce optical-electrical conversion, lower power consumption, and improve bandwidth and latency performance.

  Risks

  Commercialization pace, customer adoption, AI cluster architecture choices, competitive landscape, yield, and supply-chain stability could all affect implementation.

• AI optical networking industry chain

  The industry read-across object for the company's business

  Strengths

  AI infrastructure expansion and higher rack-level compute density are driving greater interconnect demand, which may expand the optical-networking TAM.

  Weaknesses

  Industry demand still depends on AI capex, data-center architecture evolution, and customer procurement cycles.

  Comparison

  The report believes that different configurations may benefit from growth, but optical interconnect is more attractive in high-bandwidth, low-latency, and low-power scenarios.

  Risks

  If improvements in electrical interconnects are sufficient to meet demand, or if NPO/CPO penetration is slower than expected, the commercialization of optical solutions may be delayed.

Key data

• Report Date2026-05-23The report header shows Equity Research 23 May 2026.
• CompanyLightelligence / Guangzhi Technology (1879.HK)The title identifies 1879.HK; Goldman Sachs status is Not Covered.
• Meeting Location and TimeShanghai, 2026-05-22The report says the chairman was hosted in Shanghai during the China Tech Tour.
• Core ProductsLightSphere X, PACE2, PACE3LightSphere X is used for GPU supernode interconnect; PACE2 is a 3D TSV packaged optoelectronic compute card; PACE3 is the next-generation large-scale commercialization solution.
• PACE2 Scale128×128 optical matrix, more than 40,000 photonic devicesThe report says PACE2 supports optical computing clusters and reflects complex optoelectronic chip design capability.
• PACE3 Target256×256 optical matrixManagement said PACE3 is aimed at broader AI scenarios including AI inference, and is designed to deliver better latency and energy efficiency, suitable for LLM decoding.
• Current Price DisclosureHK$575.00A Lightelligence-related price appears in the company-specific disclosure section, but the report does not provide a target price.

Impact & implications

For investment research, the implication is that as AI infrastructure expands from single-server compute toward rack-, cluster-, and supernode-level scaling, interconnect bottlenecks may become a key constraint, increasing the strategic value of optical interconnects, optical switching, and optical compute chips. If the relevant architectures are validated by customers and reach volume shipments, the company could benefit from a larger AI networking TAM; however, since the report is a Not Covered meeting note, it should not be interpreted as Goldman Sachs' formal investment rating.

Risks

• The report is a meeting note and 1879.HK is Not Covered, so there is no formal rating, target price, or earnings forecast.
• Management views are positive, but the meeting note provides limited verification of commercialization progress and customer demand.
• Next-generation solutions such as PACE3 are still aimed at future large-scale commercialization, and carry R&D, yield, customer validation, and production ramp-up risks.
• AI infrastructure capex, GPU cluster architecture, NPO/CPO penetration, and the adoption pace of optical interconnects may be slower than expected.
• Unstable collaboration across optoelectronic chips, PIC/EIC, OSAT, foundry, and EDA ecosystems could affect large-scale delivery capability.

What to watch

• Customer validation and order progress for LightSphere X in GPU supernode and inter-rack interconnect scenarios.
• The actual commercialization path and shipment pace of scale-up EPS, scale-up OCS, and NPO/CPO solutions.
• PACE3 R&D progress, 256×256 optical matrix performance, LLM decoding validation, and energy-efficiency/latency metrics.
• Changes in penetration of optical interconnect, optical switching, and co-packaged optics within AI data-center network architectures.
• Whether the company discloses verifiable financial data, customer lists, capacity plans, or mass-production milestones.