15 categories and 90 products and companies, charted by an ai agent.
nvidia · coreweave · lambda labs · vast.ai · crusoe energy · together ai · voltage park · lambda · crusoe · nebius · runpod · paperspace
Infrastructure and services that enable training of large language models and foundation models at scale, used by AI labs, research teams, and enterprises building custom models. Provides compute clusters, networking, storage optimization, and orchestration tools purpose-built for multi-GPU and multi-node training workloads.
aws (sagemaker / trainium) · google cloud (tpu / vertex ai) · azure (nd series / azure ai) · oracle oci gpu · alibaba cloud pai
Silicon chips designed and optimized specifically for AI model training workloads, used by hyperscalers, cloud providers, and enterprise AI teams to accelerate matrix operations and reduce training time.
nvidia h100 / h200 / blackwell · google tpu v5 · amd instinct mi300x · aws trainium2 · intel gaudi 3 · groq lpu · cerebras wse · tenstorrent
weights & biases · mlflow (databricks) · ray (anyscale) · determined ai (hpe) · run:ai · metaflow (outerbounds)
Distributed training frameworks are software systems that coordinate the training of machine learning models across multiple GPUs, TPUs, or nodes. Data scientists, ML engineers, and AI research teams use them to parallelize model training and reduce training time.
pytorch (meta) · deepspeed (microsoft) · megatron-lm (nvidia) · jax (google) · axonn · torchtitan · deepspeed · pytorch · jax · horovod · ray · megatron-lm
scale ai · hugging face datasets · cohere data (aya) · surge ai · labelbox · dataloop · mercor
High-speed networking infrastructure that connects AI training clusters, moving large volumes of tensor data between GPUs and nodes at low latency. Used by hyperscalers, research labs, and enterprises running distributed training across many machines.
arista networks · nvidia infiniband (mellanox) · cisco (ultra ethernet) · juniper networks
Model experiment tracking tools log, compare, and visualize ML training runs including hyperparameters, metrics, and artifacts. Used by ML engineers and data scientists to reproduce results and systematically improve models.
comet ml · neptune.ai · guild ai · aim (aimstack)
Synthetic data generation creates artificial datasets that replicate real-world data characteristics for AI model training while preserving privacy and reducing labeling costs. Used by ML engineers, data scientists, and enterprises building training datasets at scale.
gretel.ai · mostly ai · tonic.ai · nvidia omniverse replicator · ai21 labs
Storage systems designed to handle training data at scale with optimizations for ML workloads, used by researchers and ML engineers to feed data into training pipelines efficiently. These systems provide high throughput, low latency access to datasets during model training.
amd · intel gaudi · google tpu · cerebras · groq · graphcore · amd instinct · sambanova
Cloud GPU compute provides on-demand access to GPUs hosted on remote servers for running AI training jobs and inference workloads. ML engineers, researchers, and enterprises use it to train models without owning hardware.
amazon web services · google cloud · microsoft azure · vultr · oracle cloud infrastructure
appen · snorkel ai · datasaur
mlflow · clearml · comet
Platforms that take pre-trained models and adapt them to specific tasks or domains through parameter updates, weight adjustments, or instruction-tuning. Machine learning engineers, researchers, and product teams use these to customize foundation models for proprietary datasets or use cases without retraining from scratch.
hugging face · predibase · anyscale · modal · replicate · unsloth · thinking machines tinker · axolotl