Layer·Domains·Cloud & compute
The infrastructure layer that sets cost, speed, and availability. Hardware access and cloud contracts often shape what teams can ship faster than model headlines do.
- NVIDIA5GW IREN strategic partnership announced. See Cloud & compute.
NVIDIA
● Moved5GW IREN strategic partnership announced. See Cloud & compute.H100, H200, and B200 GPUs define the hardware standard for frontier training and inference. Supply constraints directly shape which labs and clouds can scale.
Partner-only
NVIDIA
● MovedCold read
NVIDIA designs and sells data-center GPUs — primarily the H100, H200, and Blackwell series — used for AI training and inference. It is a publicly traded US company and the largest data-center GPU vendor by units shipped.
Position read
Announced a strategic partnership with IREN for up to 5GW of AI infrastructure on May 8 — a supply-side commitment at a scale that signals confidence in sustained demand well past 2026. AMD and custom accelerators (MTIA, TPU) are gaining ground, but CUDA's software depth is the primary reason teams don't switch. Meta's four-generation MTIA roadmap is the sharpest public signal yet that NVIDIA's position at hyperscaler scale is on a multi-year exit path.
AWS
Operates Bedrock (multi-model managed API) and SageMaker. Default procurement path for teams already in the AWS ecosystem. GPU and Trainium hardware shaping cost structure.
PlatformDistribution
AWS
Cold read
AWS is Amazon's cloud computing division. It operates Bedrock, a managed multi-model AI service, and SageMaker, a model training and deployment platform.
Position read
Amazon's cloud division hosts foundation models through Bedrock, now including OpenAI's GPT-5.5 and Codex alongside Anthropic and Meta. For most enterprise teams already running on AWS, Bedrock is the path of least resistance for adding AI to existing workloads. AWS is the largest cloud provider by market share.
Microsoft Azure
Primary distribution channel for OpenAI models. Broad enterprise installed base gives it workflow and identity leverage that pure compute providers lack.
PlatformDistribution
Microsoft Azure
Cold read
Microsoft Azure is Microsoft's cloud computing platform. It hosts Azure OpenAI Service, the primary managed access point for OpenAI models.
Position read
OpenAI's original exclusive distribution partner — that exclusivity ended on Apr 27. Azure still carries OpenAI models and remains the default for teams already running Microsoft infrastructure, but it no longer has a unique model access advantage.
Google Cloud
Operates Vertex AI for managed model access. TPU availability and BigQuery integration are its differentiated compute assets for AI workloads.
PlatformDistribution
Google Cloud
Cold read
Google Cloud is Google's cloud computing platform. It operates Vertex AI, a managed model hosting and training service, and has exclusive access to Google's proprietary TPU accelerator hardware.
Position read
The natural choice for teams already running BigQuery or Dataflow, since the data stack integrates directly with Vertex AI. For teams not already on Google infrastructure, the TPU compute advantage matters mainly for training workloads — inference still runs on NVIDIA-based instances for most configurations.
CoreWeave
GPU-native cloud focused on AI workloads. Growing as an alternative to hyperscalers for teams optimizing inference cost and burst capacity.
Platform
CoreWeave
Cold read
CoreWeave is a GPU-native cloud provider focused on AI training and inference workloads. It went public in 2025 and counts Microsoft as one of its largest customers.
Position read
A GPU-native cloud provider focused on AI workloads, growing as an alternative to AWS, Azure, and Google Cloud for teams that need burst inference capacity without hyperscaler overhead. Recently went public. Microsoft is one of its largest customers.