[[ZKP/ZKP Base Layer/ZKP Blockchain/Performance/Application Layer|The Application Layer]] handles AI-driven computations, integrating [[ZKP/ZKP Base Layer/ZKP Blockchain/Technical Build Application Layer/Smart Contract Execution Environments EVM|EVM]] pallet and [[ZKP/ZKP Base Layer/ZKP Blockchain/Technical Build Application Layer/Smart Contract Execution Environments WASM|native WASM]] execution with ZK wrappers for [[ZKP/ZKP Base Layer/ZKP Blockchain/Technical Build Application Layer/Privacy-Preserving Computations with ZK Wrappers/Use Cases for Privacy-Preserving Applications|privacy-preserving tasks.]] Under optimal test conditions, assuming each task involves verifying a [[ZKP/ZKP Base Layer/ZKP Blockchain/Technical Build Application Layer/ZK Wrappers in Action|zk-SNARK proof]] in roughly 2 milliseconds [9, 30], the layer can process up to **500 complex AI tasks per second**. This figure reflects efficient parallel processing of proofs and is derived from theoretical zk-SNARK verification benchmarks, such as those outlined by Groth (2016) [30]. Task complexity in these tests assumes operations like matrix multiplications or inference on small neural networks, representing an optimistic projection pending empirical validation. In practice, considering the full lifecycle—including proof generation, validation, and state updates through Substrate's weight-based execution model—initial performance is projected to range between **5-20 AI tasks per second** for simple inference operations, and **0.1-1 per second** for complex models. _**Note:** Our research roadmap targets significant throughput improvements through several breakthrough technologies: (1) specialized hardware acceleration that could reduce proof generation time by 90%, (2) recursive proof composition allowing for the aggregation of multiple AI task proofs into a single verification, and (3) parallel proof generation across distributed validator nodes. These advancements, combined with circuit optimizations, have the potential to increase throughput to the 50-150 AI tasks per second range for simple operations within 1-3 years of development. Proof generation time, which currently scales as T_p = k × c × log c, will benefit most substantially from these improvements, particularly for repeated operations that can leverage proof caching and reuse strategies [30, 47, 49]._ See also: [[ZKP/ZKP Base Layer/ZKP Blockchain/Performance/Storage Layer|Storage Layer]]