NVIDIA has built tools to accelerate biomolecular structure prediction and co-folding workflows, addressing bottlenecks in Multiple Sequence Alignment (MSA) generation, co-folding inference, serving, and multi-GPU scale-out. Speed and memory-efficiency are critical for drug discovery tasks like virtual screening, where millions to billions of compounds are screened, and for predicting large molecular assemblies, where co-folding model runtime scales cubically with the number of residues. Single GPU memory limits the size of complexes that can be predicted. The NVIDIA BioNeMo Agent Toolkit gives agents access to tools to accelerate these workflows on NVIDIA B300 and H100 GPUs. To remove the MSA bottleneck, MMseqs2-GPU moves homology search onto NVIDIA GPUs, adding Hopper and Blackwell specific optimizations, including efficient support for larger-than-GPU-memory database search on NVIDIA Grace systems and speedups from improved Blackwell DPX instructions available from CUDA 13.2. These contributions have been upstreamed into the main MMseqs2 repository, and the MSA Search NIM uses MMseqs2-GPU, whose Nature Methods paper reports up to 177× faster alignment.