Citi: Optical Interconnect Market to Reach $92B by 2028 with 3-Year CAGR of 65%
Alina Collins
Citi's Hong Kong team published a 40-page deep dive forecasting the global AI optical interconnect market will reach $92 billion by 2028 at a three-year CAGR of 65% — signaling optical communications is shifting from a supporting role to a core pillar of AI infrastructure, with target prices on three A-share names sharply raised.
A $92 billion market — what's driving the growth?
Citi's first-ever 2028 forecast: global optical interconnect market at $92 billion, 2025–2028 CAGR of 65%.
Shipment volume rises from 110 million units in 2025 to 300 million in 2028 (CAGR 40%); average selling price climbs at an 18% CAGR over the same period.
This means → this cycle is a volume-and-price double lift — units are rising and each unit is getting more expensive as transceivers move to higher speeds.
Data-communications share jumps from 71% in 2025 to 89% in 2028. In plain terms = AI data centers are swallowing the entire optical interconnect market; telecom is being sidelined.
Why are 800G-and-above transceivers the core driver?
800G+ transceivers' share of data-com shipments is projected to leap from 37% in 2025 to 89% in 2028.
This reflects every new generation of AI servers (GPU/ASIC) pushing bandwidth demand higher — optical transceivers must keep pace.
The ramp timeline: 800G/1.6T transceivers ship 60 million / 22 million units in 2026; 3.2T starts at just 4 million in 2027, then surges to 35 million in 2028.
What are CPO and NPO, and when do they actually scale?
Co-packaged optics, or CPO — integrating optical engines directly inside the chip package to shorten signal paths — and near-packaged optics, or NPO — placing optical engines right next to the chip — are expected to ramp from 2027.
Shipment cadence: 400,000 units in 2026, 18 million in 2027, 56 million in 2028. This means → 2026 is still a pilot phase; 2027–2028 is the real proving window.
The driver is AI networks scaling in both directions — horizontally (more nodes) and vertically (higher bandwidth per node) — outstripping what traditional pluggable transceivers can deliver on power and density.
How much will silicon photonics and optical chip demand grow?
Silicon photonics — using light instead of electricity to move data between chips, for higher speed and lower power — is forecast to rise from 29% penetration in 2025 to 60% in 2028.
Three forces converge: 800G+ transceiver adoption, CPO/NPO migration, and a shortage of EML chips (a type of directly modulated laser chip) pushing the industry toward silicon-photonics alternatives.
Total optical chip demand hits 1.705 billion units in 2028 (CAGR 62%); CW laser chips — the light source for silicon-photonics modules — reach 987 million units at a 114% CAGR, the fastest-growing segment in the entire chain.
How did Citi reprice the three A-share names?
Dongshan Precision: target price raised from RMB 225 to RMB 350 (+56%). Citi expects the company to capture share in 800G/1.6T modules and 100G EML/CW laser chips; AI optical-communications profits are forecast at RMB 5.4 bn / 25.7 bn / 52.7 bn for 2026/27/28, with the optical-chip business valued at 50× P/E via SOTP.
Eoptolink: target price raised from RMB 353.57 to RMB 701 (+98%). The largest increase — driven by a three-year earnings CAGR of 190%, mainly from 3.2T transceivers and NPO contribution, valued at 20× 2027 P/E.
O-Net Technologies: target price raised from RMB 318.57 to RMB 419 (+32%). Near-term earnings trimmed on EML supply tightness reducing optical-engine shipments, but a 2028 net-profit forecast of RMB 19.1 bn is introduced, benefiting from CPO and 3.2T optical-engine ramp, valued at 34.3× 2027 P/E.
The three upgrades differ sharply — what does that tell us?
Target-price increases: Eoptolink +98% > Dongshan Precision +56% > O-Net +32% — a nearly threefold gap.
This reflects Citi's sharply different views on when each company benefits most from the transceiver upgrade cycle and CPO/NPO penetration: Eoptolink is seen as the highest-beta play, while O-Net faces a near-term EML bottleneck.
In plain terms = Citi's core thesis is that 2027–2028, when CPO truly scales, is the payoff window. What investors are buying today is an expectation about that window — whoever benefits most clearly in those two years gets the biggest target-price lift.
Content is for reference only, not financial advice.