1.6T Optical Transceiver:
Volume Inflection Has Begun

As of February 26, 2026, the 1.6T transceiver industry has moved beyond the "testing phase" and entered the beginning of a "full-scale volume ramp." Coherent announced on February 4th during its FY26 Q2 earnings release that "1.6T is being delivered to customers simultaneously based on three laser sources: SiPh, EML, and VCSEL." The key takeaways are the 5th consecutive earnings beat and the Q3 guidance of $1.70~1.84B — the statement that demand visibility has extended to 2028.

The market structure is clear. Innolight (TeraHop) maintains its leading position with a 50%+ share for NVIDIA, having started initial mass production from Q4 2024. While total 1.6T shipments in 2025 were less than 1M units for the entire market, quarterly deliveries in the millions of units are within sight in 2026. Eoptolink announced its Gen2 1.6T OSFP at OFC in March 2025 — reducing power consumption by 20% with a 3nm DSP, and offering both SiPh and EML versions, putting pressure on Innolight.

Coherent's strategy is clear. As the only Western company with InP lasers (6-inch wafers, expanding the Järfälla fab in Sweden), SiPh integration, and a VCSEL lineup, vertical integration defends the module cost structure. At the same time, it is increasing production capacity in Malaysia (Ipoh, Penang) and Vietnam. Lumentum mentioned that EML laser shipments are hitting record highs, with 200G/lane EML accounting for 10% of revenue in early 2026. Marvell declared vertical integration of DSP+photonics with the acquisition of Celestial AI (December 2025), a key strategic event for 2026.

Supply chain geopolitics has settled into "China+One." Thailand has established itself as a neutral production base for the optics world, with Innolight (TeraHop), Eoptolink, Fabrinet, and Lumentum all expanding their production bases in Thailand. This is an essential route to avoid the US Section 301 tariffs (25%) and to obtain "Made outside China" certification.

The hottest technology debate in 2026 is LPO (Linear Pluggable Optics) versus the DSP-based retimer method. LPO can reduce power consumption by 50% and latency by 90% by eliminating the DSP chip, but it requires strict signal integrity and is optimized for short distances (0-2km). Eoptolink is pressuring Coherent by preempting NVIDIA's 800G LPO orders, and has already secured 1.6T LRO (Linear Receive Optics) orders. Coherent has officially confirmed that it is shipping both LPO and LRO.

Large-scale commercialization of CPO (Co-Packaged Optics) is expected in 2028-2030. Cignal AI stated that "CPO will have no material impact on pluggable shipments for the next 3 years," dispelling excessive concerns about the lifespan of the pluggable transceiver market. OSFP-XD has been confirmed as the de facto standard form factor, accounting for 92% of hyperscaler 1.6T contracts.

The DeepSeek shock actually provided a buying opportunity. The logic that algorithm efficiency does not reduce GPU demand but rather accelerates the construction of AI clusters (more people using cheaper AI more often) actually expands the demand for optical interconnects. COHR's Q2 earnings beat and upward revision of Q3 guidance proved this logic with actual results.

H2 2026 is the real volume inflection point. This is when the full transition to 1.6T for GB300 servers, the completion of hyperscaler verification by Google and Amazon, and the full-scale mass production of Eoptolink's Gen2 all overlap. ASP is expected to maintain a 1.5-2x premium compared to 800G.