Accelerate Innovation in the Cloud

Perplexity built pplx-api using NVIDIA A100 Tensor Core GPUs on AWS and NVIDIA TensorRT™-LLM, achieving up to 3.1x lower latency and 4.3x lower first-token latency compared to other platforms.​ The startup slashed inference costs by 4x—saving $600,000 annually—while scaling to hundreds of GPUs, with NVIDIA H100 GPUs delivering 50% lower latency and 200% higher throughput than A100s.​

Noetik, a member of the NVIDIA Inception Program, uses NVIDIA Hopper™ Tensor Core GPUs on AWS SageMaker HyperPod to train multimodal foundation models for precision cancer immunotherapy. This enables processing of 1 petabyte of human tumor data—profiling over 200 million cells—to accelerate therapeutic discovery and unlock treatments tailored to individual patients.

Fireworks.ai built a lightning-fast, cost-optimized generative AI inference solution using Amazon EC2 P5 instances powered by NVIDIA H100 Tensor Core GPUs. The platform delivers 4x higher throughput per instance than open-source solutions, cuts latency by up to 50%, and reduces overall costs by 4x for some customers. Developers can run, fine-tune, and customize foundation models including Llama 2, Stable Diffusion XL, and StarCoder while meeting HIPAA and SOC2 Type II compliance standards.

A-Alpha Bio accelerated drug discovery by deploying NVIDIA BioNeMo™ on AWS, achieving 12x faster inference and processing 108 million protein-binding predictions—10x more than initially planned.​ Using Amazon EC2 P5 instances powered by NVIDIA H100 Tensor Core GPUs, the biotech startup reduced experimental cycles by 1–2 iterations, cutting costs while discovering superior monoclonal antibody candidates for therapeutics.​

Synthesia transformed AI video production by deploying NVIDIA GPU-powered Amazon EC2 instances, achieving a 30x improvement in ML model training throughput.​ Using Amazon EC2 P5 instances with NVIDIA H100 Tensor Core GPUs and P4 instances with NVIDIA A100 GPUs, the AI startup reduced training time for voice models from days to hours while supporting 456% user growth.​

Innophore accelerates drug discovery using NVIDIA BioNeMo to analyze protein structures with its Catalophore technology.​ The platform completed mapping of the entire human organism’s protein structures in two weeks—a task that previously took over a year.​ This improves accuracy in predicting off-target drug effects by 30% within top-ranked hits.