The Next Bandwidth Curve
Why 800G & 1.6T Are Reshaping AI Infrastructure
The AI-cluster optics market reached $16.5 billion in 2025 and is on track to hit $26 billion in 2026 — roughly 60% growth for two years running. A market that doubles in two years isn’t a refresh cycle. It’s a structural shift in how AI infrastructure is engineered.
The Optical Roadmap
Market: LightCounting, 2026 forecast01 — The Bottleneck
Bandwidth Is the New Bottleneck
Every new generation of AI accelerators moves more data per second. Larger models, longer context windows, and scaled inference all land on the same constraint: the network must keep up with compute — or GPU utilization pays the price.
In large training clusters, the #1 cause of wasted GPU compute isn’t the model — it’s network instability and link flaps that interrupt distributed communication and force costly re-computation.
At 400G, AI clusters are already straining. The next wave of AI factories is being designed around 800G today and 1.6T on the near horizon — driving new module types (OSFP, OSFP224) built for higher speeds and higher port densities.
02 — For AI-Infra Teams
What 800G / 1.6T Actually Change
Roadmaps Matter Now
The fabric you deploy today must have a credible path to 800G/1.6T without a full re-architecture.
Port Density Rewrites the Rack
Higher-speed modules deliver the same bandwidth in fewer ports — changing rack design, power draw, and cooling.
Power-per-Bit Is the Real Metric
Faster optics must move more data per watt. Optical-layer efficiency is now a first-order design constraint.
Pluggable Isn’t the Only Answer
Co-packaged optics couples the optical engine directly to the switch ASIC — pushing power and stability further than pluggables alone.
The Proof: Our 51.2T CPO Switch
Our industry-first 51.2T co-packaged optics switch — now in volume production — is where the curve bends. Built with Broadcom’s Tomahawk 5 and manufactured in our co-owned Malaysia factory, it redefines what an AI fabric can deliver:
03 — The Method
The Open AI Factory Approach
We treat optics as a strategic part of the architecture, not an afterthought. It’s why the “factory” in our name is literal:
An Evolution, Not a Rebuild
The fabric is designed with the optical roadmap in mind — 400G → 800G → 1.6T is a continuous path, not a fork.
Validated Before It Ships
Transceivers are validated against switching and compute layers before shipping — Optech’s 800G and 1.6T OSFP / OSFP224 modules are qualified in that loop for next-generation AI data centers.
Multi-Vendor Optics
Customers keep choice: multi-vendor optics mean no single supplier owns the speed roadmap.
We Own the Factory
Our co-owned factory in Malaysia — clean-room manufacturing with advanced automation — is what lets us take CPO from validation to volume production without handing the roadmap to a third party.
big news Lands September 15.
We’re at booth #316 with Broadcom showcasing our Open AI Factory story — planning an 800G or 1.6T build? Have it IRL!
Bandwidth is the new compute.
AI factories that plan for 800G and 1.6T today — and control the validation path to them — scale without rebuilding tomorrow.