From 1.6T to 6.4T: tackling reach and speed bottlenecks in AI data-center interconnects
NantOptiFab's roadmap for next-generation AI interconnects, quantum light sources, and free-space optical communication, built on heterogeneous, multi-material photonic platforms.
Two bottlenecks at once: reach and throughput
Optical interconnects inside modern AI data centers are constrained along two physical dimensions. Transmission reach remains limited under current architectures, restricting how flexibly compute and memory can be clustered across a data-hall campus. Data throughput is the second limit: with 1.6T becoming the mainstream deployment baseline, conventional optical transceivers struggle to scale at the pace AI training and inference clusters demand.
In short, links inside today's data centers are too short to connect compute flexibly, and bandwidth is not scaling as fast as demand.
A heterogeneous, multi-material approach
Rather than forcing a single material system to satisfy conflicting physical requirements, NantOptiFab combines complementary photonic materials within an integrated platform architecture.
Silicon photonics delivers sub-micron waveguide confinement and high-density, CMOS-compatible fabrication at scale. Because silicon lacks an intrinsic linear electro-optic (Pockels) response, however, it meets material-level limits at ultra-high modulation speeds.
Thin-film lithium niobate offers an exceptionally high Pockels coefficient. In high-confinement ridge waveguides, TFLN enables ultra-broadband, low-loss electro-optic modulation in a dramatically smaller footprint.
Uniting the two platforms is designed to ease the trade-off between higher line rates and longer channel distances.
That is why NantOptiFab does not rely on silicon photonics alone: integrating thin-film lithium niobate alongside complementary materials targets both higher modulation speeds and longer link spans.
Design targets and development status
While the industry standardizes around 1.6T, NantOptiFab's interconnect architecture targets 6.4T from the outset: a fourfold throughput step.
6.4T is a target product specification, not current volume production; core component-level devices are in verification and validation. Actual link budgets and aggregate data rates depend on each customer's network architecture, advanced packaging (including CPO), and link length. Stated figures are internal architectural benchmarks, not customer field-test data.
Quantum light sources: front-end precision sets back-end fidelity
In photonic quantum computing and networking, the quality of the front-end quantum state directly shapes fault-tolerance thresholds and computational fidelity.
High-purity sources of entangled photon pairs reduce quantum noise and computational error, which makes a clean, stable light source the first building block of any photonic quantum processor.
Built on periodically poled lithium niobate engineered for precise quasi-phase matching, NantOptiFab's quantum sources enable highly efficient nonlinear conversion for quantum key distribution and photonic quantum-computing front ends.
Free-space optical communication for orbital data infrastructure
Beyond terrestrial networks, orbital compute nodes and satellite data constellations require links spanning thousands of kilometers through the atmosphere and the vacuum of space.
Over inter-satellite and ground-to-space links, optical communication is effectively the only transmission technology that can deliver the bandwidth required.
For these mission-critical aerospace requirements, NantOptiFab's LN, ELN™, and TFLN platforms support optical inertial navigation, coherent satellite laser communication, and high-reliability spatial sensing.
See the AI interconnect applications ↗