From 800G to 1.6T: Managing the Transition in High-Density AI Clusters

By Simon Ximen, AVP, Business Development and Sales

The conversation around optical transceivers has shifted noticeably over the past six months. Where 800G was still being positioned as the aspirational next step heading into 2026, it’s now the working assumption for any serious AI infrastructure build. The question most engineering and operations leaders are asking is not whether to standardize on 800G, but how to time the move into 1.6T.  

1.6T Is Moving from Roadmap to Volume Reality
Industry analysts project that shipments of 800G-and-above optical transceivers will grow from roughly 24 million units in 2025 to nearly 63 million units in 2026, reflecting a 2.6× year-over-year increase. That is a genuine volume inflection, not a roadmap projection. 

Regarding form factors, OSFP has become the dominant choice for 1.6T deployments. Its integrated thermal design and support for high-lane-count configurations—particularly as we transition to 200G-per-lane electrical interfaces—make it the practical default for high-density AI cluster deployments. While OSFP-XD remains a point of interest for specific next-generation applications, the primary ecosystem for 1.6T production deployments today is firmly built around the standard OSFP form factor. 

800G Supply May Tighten Faster Than Expected
As hyperscaler demand for 1.6T pulls manufacturing attention and capital, the supply picture for 800G is not guaranteed to be as straightforward as it is today. Engineering teams planning multi-year optical roadmaps should factor supplier manufacturing capacity into their vendor assessments, not just current product availability. 

AOI, for example, is on track to achieve a combined 800G and 1.6T production capacity exceeding 500,000 units per month by the end of 2026. With a further expansion of our Houston-area manufacturing footprint underway, this level of capacity visibility and U.S.-based scale is critical when signing agreements with 18- to 24-month delivery horizons.

Aligning Optical and GPU Procurement Timelines
One of the most underappreciated challenges in AI infrastructure planning is aligning GPU delivery with optical readiness. GPU lead times remain elevated, and the qualification window for optical transceivers adds additional complexity. A module qualified for one switch and NIC combination may require extensive validation work when the platform changes—which it frequently does between planning and deployment.

The practical recommendation is to treat optical qualification as a parallel workstream to compute procurement, not a downstream dependency. Work with vendors who have established, current qualification relationships with the switch and NIC platforms you are deploying.

Vertical Integration as a Supply Chain Advantage
Vertical integration at the component level, specifically in-house laser production, provides insulation against the supply chain volatility that fully outsourced module assemblers are more exposed to. Vendors without in-house laser fabrication are exposed to this constraint in ways that vertically integrated manufacturers are not. 

AOI’s model spans the entire lifecycle: from laser growth to finished transceiver production. Our Sugar Land, Texas facility is expanding specifically to support this large-scale 1.6T output. The combination of TAA compliance and in-house manufacturing depth addresses procurement requirements that vendors dependent on external laser supply or offshore-only production simply cannot meet for certain customer segments.

Aligning Optical and GPU Procurement
For network planning teams trying to translate all of this into actual decisions, a few principles tend to hold up:

• Start 1.6T procurement conversations now if you are planning a cluster for 2027. 
• Treat supplier manufacturing footprint as a first-class evaluation criterion. 
• Align optical generation to compute requirements.

The 800G-to-1.6T transition is real, and it’s accelerating. But it’s also nuanced enough that the planning decisions made now will compound over multiple deployment cycles. Getting the timing right, the procurement relationships established early, and the supply chain risk assessed clearly is where engineering and operations leaders can make the difference between a smooth infrastructure scale-up and an expensive recovery situation.