PhotonWeave was founded to solve a problem the industry had been avoiding for over a decade:
the bandwidth wall inside advanced compute systems.

​

As GPUs scaled faster and data centers grew denser, the interconnect problem quietly became the dominant limiter of performance and power. SerDes power scaled linearly with bandwidth, retimers multiplied latency, and pluggable optics sat too far from the compute to matter. The industry kept optimizing around the edges—but the core architecture remained unchanged.

​

Eric Li saw this firsthand.

​

With a background in physics from Peking University and decades of experience building high-performance communication and packaging systems, Eric had worked across LEDs, optoelectronics, and advanced silicon platforms. What became clear over time was that optics had already won the long-haul battle—but lost the short-reach war for the wrong reasons.

​

The missing piece wasn’t lasers, fibers, or DSP.
It was how photons were generated, timed, and received at the package boundary

​

In 2024, Eric began rethinking optical I/O from first principles. Instead of forcing traditional optical transceivers into shorter distances, he asked a simpler question:

​

                                        What if optical I/O were designed like a spatial fabric—not a serial link?

​

The breakthrough came from combining three ideas that had never been integrated at scale:

• Micro-LED arrays as dense, directly modulatable light sources—eliminating external lasers

• Spatial and phase-aligned signaling that removes the need for power-hungry SerDes

• Package-level integration, bringing optics inside the compute envelope rather than at the edge

This became PhotonWeave.

​

PhotonWeave’s core architecture—PhaseWeave—replaces high-speed serial links with thousands of low-energy, phase-coherent optical channels operating in parallel. Each emitter runs slower, cooler, and simpler—yet the aggregate bandwidth scales by orders of magnitude. The result is 16× higher bandwidth at the same power, with a path to tens of terabits per second per package.

​

The implications go far beyond faster links:

• GPU-to-GPU communication without SerDes bottlenecks

• Optical GPU-to-HBM interconnect inside advanced packages

• A fundamentally new scaling model for AI, HPC, and disaggregated systems

​

PhotonWeave was founded not to make optics smaller—but to make compute scale again.