Neurophos, a startup company that has garnered investment from Bill Gates, has made a groundbreaking announcement: it has successfully developed its inaugural Optical Processing Unit (OPU), dubbed the Tulkas T100. When it comes to tackling FP4 and INT4 computational tasks, the Tulkas T100 outperforms NVIDIA's cutting-edge Vera Rubin NVL72 AI supercomputer by a factor of 10, all while consuming the same amount of power. Neurophos has successfully tackled the long-standing challenge of bulky optical transistors prevalent in traditional silicon photonics manufacturing, shrinking their size by an impressive 10,000-fold. This breakthrough paves the way for the seamless integration of extensive photonic sensors directly onto chips. In comparison to the majority of AI GPUs, Neurophos has significantly boosted the scale of the photonic sensor matrix integrated onto the chip, increasing it by roughly 15 times. Its first-generation accelerator is outfitted with a single Tensor core, occupying an area of approximately 25 square millimeters. The Tulkas T100 is capable of reaching a theoretical operating frequency of up to 56GHz, comes equipped with 768GB of HBM, draws between 1 and 2 kilowatts of power under full load, and attains a computational speed of up to 470 petaOPS.