Goldman Sachs Raises Innolight Target Price to RMB 2,581
Claire Weston
Goldman Sachs on July 17 nearly doubled its 12-month target for Innolight (300308.SZ) from RMB 1,187 to RMB 2,581, citing Nvidia's AI-server roadmap as the engine behind accelerating demand for high-speed optical modules — a signal that the world's largest optical-interconnect supplier is entering its biggest upgrade cycle ever.
Why did the target nearly double in one move?
Goldman maintains a Buy rating, lifting the 12-month target from RMB 1,187 to RMB 2,581 — a 117% increase.
This means → Goldman believes the current share price has yet to price in the full AI optical-interconnect upside.
Three drivers behind the upgrade: ① AI-infrastructure capex consistently beating expectations; ② Innolight's capacity expansion outpacing prior forecasts — overall capacity set to more than double in 2026; ③ entry into new markets including optical engines, NPO/CPO, and OCS optical switches.
How big are the earnings upgrades?
Revenue forecasts for 2026–2028 raised by 57%, 100%, and 118% respectively, implying revenues of RMB 123.6 bn, 249.6 bn, and 334.0 bn.
Net-profit forecasts raised 65%, 108%, and 119% over the same period.
In plain terms = Goldman more than doubled its three-year revenue outlook — this is not a trim, it is a complete redrawing of the growth curve.
How fast is the 800G-to-3.2T upgrade unfolding?
The global optical-module market is forecast to grow from $34.2 bn in 2025 to roughly $69.1 bn in 2028.
Shipment ramp: 800G modules at ~34.18 million units in 2026; 1.6T modules at 25.5 million in 2026, jumping to 45.72 million in 2027; 3.2T modules entering volume in 2027 at 13.07 million, rising to 28.16 million in 2028.
This means → each Nvidia AI-server generation (GB200 → GB300 → Rubin → Rubin Ultra) pushes optical-module speeds up one tier, creating staircase demand rather than linear growth.
Why is silicon photonics the key moat?
Silicon photonics — a technology that moves signals between chips using light instead of electricity, enabling higher speed and lower power — is projected to reach 80% and 100% penetration in 1.6T and 3.2T products by 2026.
Industry-wide SiPh revenue share is expected to rise from 28% in 2025 to 62% in 2028.
Innolight is among the first globally to achieve large-scale commercialization of high-speed SiPh modules, with next-gen capabilities spanning LPO, LRO, XPO, and CPO/NPO.
This reflects a tightening supply of core components combined with rising technical barriers — replicating this advantage is becoming harder for new entrants, pointing to further share gains for the leader.
Can profitability keep pace with expansion?
Gross margin is forecast to rise from 42% in 2025 to 50% in 2028; net margin to 33%.
ROE climbs from 44% in 2025 to 52%–61% across 2026–2028.
2026 capex is expected to surge 155% YoY to RMB 7.03 bn, yet free cash flow remains robust: RMB 8.5 bn, 62.3 bn, and 111.5 bn for 2026–2028.
In plain terms = expansion is expensive, but revenue is growing faster — cash flow is not being crushed by capex.
Where is the biggest uncertainty?
Goldman explicitly flags: whether 3.2T modules can ramp on schedule is the critical verification point for all the above forecasts.
This means → if 3.2T mass-production slips, the 2027–2028 revenue step-up gets deferred, and the "forward premium" embedded in today's valuation needs repricing.
AI-server rack counts are projected to grow from ~55,000 in 2026 to 163,000 in 2028 — that deployment pace is itself another variable to track.
Content is for reference only, not financial advice.