AMD and Intel, move over: AI designed an all-new CPU — and claims it is 4.6x more efficient

US AI startup Architect Labs says its AI systems have designed a new general-purpose CPU architecture from scratch, including its instruction set, microarchitecture, RTL, verification, and final GDS layout. The company claims its new design delivers 4.59 times the performance per watt and 8.77 times the performance per unit of area of a conventional CPU core across 25 workloads. However, the results come from pre-silicon simulations rather than a physical processor.
Unlike conventional CPUs, Frostbite is built around a mathematical model called interaction nets, first introduced in 1990. Instead of relying on a single large core, it uses a grid of small computing tiles that execute programs represented as interaction nets: graphs in which computation proceeds through local rewrites of connected nodes. Each tile first processes its own tasks and can take on additional work from neighboring tiles when it becomes idle. If no work is available, it can enter an inactive state to save power. This distributed approach eliminates the need for a centralized scheduler and is intended to make the architecture easier to scale.
Architect Labs says this approach also reduces the need for shared mutable memory, complex cache hierarchies, and cache-coherence mechanisms. Each value has a single owner, while computation takes place through local interactions between connected nodes. Programs are written in the company’s Vine language, which has syntax similar to Rust, and are compiled into an intermediate representation called Ivy.
The company says its AI systems explored the architecture with “zero architectural priors,” iterating across the compiler, instruction set, architecture, microarchitecture, RTL, verification, and final GDS layout. The process used formal, machine-checkable specifications and multiple agent swarms, with 58 proposed ideas considered across four searches and a final round of tests. Architect Labs says 37 ideas were retained.
For the performance evaluation, Architect Labs synthesized Frostbite using a 7 nm process. The design reached a frequency of 800 MHz during synthesis and 500 MHz in its final physical layout. One Frostbite Quad, consisting of four computing tiles arranged in a 2×2 grid and paired with four L3 nodes, occupies 0.792 mm2. The comparison CPU core, which the company calls TCPU, occupies 0.895 mm2 using the same process technology. Architect Labs has not disclosed the identity of the competing processor for legal reasons.
Across 25 workloads, Frostbite achieved a geometric mean of 4.59 times the performance per watt of the reference CPU core. Its advantage reached 6.02 times across the 10 workloads used during optimization and 3.83 times across 15 additional workloads that were held back until the final evaluation. The company also reports an 8.77-fold advantage in performance per unit of area.
Architect Labs says the gains come from reducing memory traffic and instruction overhead. The results varied considerably depending on the workload. Frostbite performed particularly well in highly parallel tasks, although Architect Labs says it also outperformed the reference core in branching and dynamic workloads. The company plans to expand testing to include industry-standard SPEC and PARSEC benchmarks, as well as comparisons against larger, wider, and out-of-order CPU designs.
Despite the promising results, Frostbite remains an experimental architecture rather than a production-ready processor. Architect Labs has not announced a manufactured Frostbite chip or commercial availability. It plans to release the RTL and instruction set architecture (ISA), build an FPGA-based software playground, develop a preliminary operating system and benchmark larger Frostbite configurations against multicore CPUs using industry-standard benchmarks such as SPEC and PARSEC.













