SmartCore lands China angel++ round for silicon photonics AI interconnect

SmartCore sits at one of the least glamorous but most critical chokepoints in the race to AGI: moving bits between GPUs. As models scale to trillions of parameters and context windows stretch past 200K tokens, the memory and interconnect bottlenecks become at least as important as FLOPs. A Tsinghua‑affiliated team raising meaningful early capital for silicon‑photonics chiplets in Shenzhen suggests that China is not just chasing Nvidia on core compute, but also on the optical plumbing that makes massive clusters viable.

Silicon photonics is attractive here because it promises far higher bandwidth and lower energy per bit than copper backplanes, especially at rack‑scale and beyond. If SmartCore can deliver robust, manufacturable optical I/O chiplets, it could help Chinese clouds and labs build denser, more efficient clusters without relying as heavily on foreign interconnect IP. That would blunt one of the few remaining levers the US and its allies have in semiconductor export policy.

For the broader ecosystem, SmartCore is another reminder that AGI‑relevant innovation is increasingly distributed across a long tail of component suppliers. Frontier labs and hyperscalers will have to map, and in some cases secure, supply chains not just for GPUs and HBM, but for the optical substrates and packaging technologies that tie it all together.