Architect Labs presented Frostbite, a general-purpose processor designed by swarms of AI agents and built around a new instruction set architecture. In simulations of a 7 nm design, the startup measured 4.59 times the performance per watt of a comparable core and 8.77 times the performance per area across 25 workloads. On 15 workloads not used during tuning, however, the advantage fell to 3.83 times. No physical Frostbite chip has yet been fabricated, and all results come from simulations that the company estimates to be within 1% to 5% of real hardware. The startup raised $24 million in June 2026 and plans to publish additional benchmarks, along with RTL, ISA, and FPGA implementations.
Source: Read the original article


