Post by Ryan Lindsay

Tim Ayres Andrew Charlton David Swan George Robertson Karissa A. Breen (KB) Alexandre BLANC Cyber Security Aaron Violi MP Matthew O'Brien Mark Stephens CBE Dr. Jeffrey Funk Tim Wilson The West’s current AI infrastructure strategy is a structural trap. Led by evangelists like Sam Altman and echoed by policymakers, the prevailing playbook is simple: construct massive data centers packed with Nvidia GPUs. But this isn't true innovation—it's herd behavior. As the adage goes, "a copycat strategy is a strategy for losing slowly." We aren't pioneering breakthrough AI architectures; we are merely building expensive, power-hungry furnaces to run someone else’s models, artificially inflating the cost of silicon and high-speed RAM in the process. This brute-force approach faces a hard thermodynamic wall. Pushing electrons through copper wiring generates intense resistance and Joule heating. The result is a crippling inefficiency where massive amounts of electricity are consumed just to compute, and millions more are spent on the cooling infrastructure required to keep the hardware from melting. While the West doubles down on this dead end, China is executing a strategic lane change out of pure necessity. Cut off from advanced Western lithography by strict trade sanctions, Beijing is running a dual-track strategy. While they maintain a conventional GPU track via competent domestic silicon like Huawei’s Ascend processors, their primary long-term focus has shifted to optical computing. Photonic chips change the physics of the race entirely. By processing data via light waves rather than electrons, they eliminate electrical resistance, generating virtually zero heat while operating at the literal speed of light. Chinese researchers have already demonstrated architectures operating at trillions of operations per watt. Tsinghua University’s ACCEL chip has demonstrated an astonishing 160 TOPS/W (trillion operations per second per watt), while deep architectures like SLiM are actively overcoming historical signal error barriers. Crucially, photonics solves the manufacturing bottleneck. Because wavelengths of light are physically larger than electrons, components like ring resonators and waveguides are much larger than silicon transistors. They don't require advanced, restricted EUV lithography; they can be manufactured using mature, unrestricted 45nm to 90nm fabrication equipment—nodes where China possesses massive, self-reliant legacy capacity. Meanwhile, the Western tech ecosystem remains paralyzed by software lock-in, with enterprise developers tethered tightly to Nvidia's proprietary CUDA ecosystem. Rather than architecting the most efficient hardware for the job, we are building infrastructure defined entirely by vendor dependence. The fully photonic AI processor is a thermodynamic no-brainer—yet while the West copy-pastes the GPU playbook, China is quietly rewriting the rules of the track.