A Citi TMT conference analysis discusses how AI agent infrastructure is reshaping semiconductor and optical interconnect demand. Storage capacity needs are expanding exponentially as AI models require massive KV caches, driving NAND TAM projections from $50B to $500B by 2027, while optical interconnects emerge as the next supply constraint with laser and fiber components becoming critical bottlenecks.
DeepJIT is a header-only C++20 JIT runtime that provides a unified interface for compiling and caching GPU kernels on NVIDIA CUDA and Huawei Ascend devices. It enables kernel libraries to share runtime infrastructure including caching, lazy initialization, and distributed filesystem support across both backends.