Cloud providers, chipmakers, and startups are expanding alternative software stacks and interconnects for AI clusters, from Rust GPU kernels to optical switches. The push aims to reduce dependency on a single vendor and lower the cost of large-scale AI networking.
NVIDIA has released CUDA Oxide 0.3, an update to its experimental Rust-to-CUDA compiler. The project aims to let developers write GPU kernels in Rust within the CUDA ecosystem. It opens an alternative path to C++/CUDA for on-prem and self-hosted stac...
PsiQuantum has built an 8x8 optical switch prototype for AI data center GPU networks. The company draws on its photonics expertise and reports lower light loss than rival solutions, with plans to develop larger devices. The project highlights the gro...
The first Loom SDK v0.1 from AMD marks the start of the Hyperloom roadmap announced at Advancing AI, aiming to optimize agentic workloads across CPUs, GPUs and NPUs. For local deployment, the signal is twofold: AMD wants to unify the ROCm ecosystem, ...
Competition with CUDA is widening: AWS and Google are pushing alternative software stacks, while AMD and Intel aim to lower the cost of AI cluster switches. Reduced dependence on NVIDIA's software and cheaper networking shift the TCO equation for lar...
