GoogleCloud & AI
MicrosoftCloud & AI
AmazonCloud & AI
MetaSocial & AI
SpaceXSpaceXAI / xAI
OpenAIModel developer
AnthropicModel developerThe full-stack advantage
Google, Microsoft, and Amazon connect cloud platforms with their own models, applications, and chip designs. That gives them more ways to shape how AI is built and delivered.
THE AI LANDSCAPE 01 / MARKET STRUCTURE
Who’s building
the AI stack?
The race goes far beyond models. Explore how seven major players connect the apps, intelligence, infrastructure, and chips behind AI.
THE BIG PICTURE
AI Stack Map
Follow each company from the application down to the silicon.
No matching companies
Try a company name, a product like Claude, or a broader group.
A selected-company snapshot, not a performance ranking. Companies can own a layer and still depend on partners. On smaller screens, swipe across the map.
READ BETWEEN THE LAYERS
One stack. Different strategies.
Our interpretation of the map.
Owning doesn’t mean alone
Custom chips still involve manufacturing partners. AI applications can use outside models. Read each cell as one part of an ecosystem, rather than a claim of total independence.
The boundaries keep moving
OpenAI’s chip program and Anthropic’s compute partnerships show different routes into the infrastructure layer. Announcements, operating capacity, and ownership are different things.
TRANSPARENT BY DESIGN
The sources behind
the stack.
Company documentation and announcements, reviewed on September 22, 2026. This page is an editorial snapshot and is updated manually.
Suggest a correctionHow the map is classified
We selected seven major AI companies to make their stack positions comparable. The map is not exhaustive. “Platforms” groups companies operating cloud or large-scale compute platforms; “Model labs” groups OpenAI and Anthropic for navigation, not as an exclusive industry taxonomy.
Own product. An available product, model, operated compute platform, or deployed chip designed by the company. It can still use partners and outside components.
In development. An announced program, expansion, or early hardware deployment. This does not imply that every part is already available at scale.
Partner supplied. The examples shown are supplied by other companies. This describes sourcing, not product quality or a lack of access.
Product names are representative, and the colors are AIstify’s editorial classification of the cited evidence. They do not measure market share, model quality, or financial strength.
Google4 layers
Google / Applications
Gemini
Gemini is Google’s consumer AI assistant. This row shows a representative application, not every Google AI product.
Google / Models
Gemini · Gemma
Google develops the Gemini and Gemma model families. Gemma provides open weights under its own terms; availability and licenses vary by model.
Google / Infrastructure
Google Cloud
Google operates a cloud platform offering AI compute, including Cloud TPU infrastructure.
Google / Silicon
TPU
Google designs Tensor Processing Units for AI workloads. Own design does not mean that Google fabricates the chips itself.
Microsoft4 layers
Microsoft / Applications
Copilot
Microsoft owns the Copilot product family. Owning an application does not mean all models powering it are developed in-house.
Microsoft / Models
MAI
Microsoft develops MAI models spanning reasoning, coding, image, voice, and transcription. This is a product-presence label, not a frontier capability ranking.
Microsoft / Infrastructure
Azure
Microsoft operates Azure, including its cloud infrastructure for AI workloads.
Microsoft / Silicon
Maia 200
Microsoft announced Maia 200 deployment in January 2026. It is an in-house accelerator for inference, alongside hardware from external suppliers.
Amazon4 layers
Amazon / Applications
Kiro · Amazon Q
Kiro and Amazon Q are examples of Amazon’s AI developer applications. Their underlying model choices can include third-party models.
Amazon / Models
Nova
Amazon develops Nova foundation models and makes them available through AWS. This entry is separate from third-party models offered through Bedrock.
Amazon / Infrastructure
AWS
Amazon Web Services operates the cloud infrastructure that provides its AI services and accelerator instances.
Amazon / Silicon
Trainium · Inferentia
AWS designs Trainium and Inferentia accelerators. These sit alongside external hardware in its wider compute offering.
Meta4 layers
Meta / Applications
Meta AI · Muse
Meta offers Meta AI and introduced the Muse personal agent in September 2026. Access and availability differ between products and regions.
Meta / Models
Muse Spark · Llama
Meta develops Muse Spark and Llama. The families have different distribution and licensing approaches; this map does not treat every Meta model as open weight.
Meta / Infrastructure
AI data centers
Meta operates AI infrastructure for its own services. This is not a general-purpose public cloud equivalent to AWS, Azure, or Google Cloud.
Meta / Silicon
MTIA
Meta deploys MTIA chips and is expanding the family. Its MTIA 300 training chip targets recommendation models; this does not establish parity across all generative AI workloads.
SpaceX4 layers
SpaceX / Applications
Grok · Cursor
Grok is offered by SpaceXAI. Cursor confirmed that its acquisition by SpaceX closed in August 2026.
SpaceX / Models
Grok
SpaceXAI develops the Grok family. The company was previously branded xAI.
SpaceX / Infrastructure
Colossus
SpaceX operates Colossus infrastructure. Anthropic also contracts capacity at Colossus 1, illustrating how infrastructure ownership and model access can overlap.
SpaceX / Silicon
NVIDIA
The documented Colossus hardware includes NVIDIA GPUs. This label describes that deployment’s sourcing and makes no claim about every future chip initiative.
OpenAI4 layers
OpenAI / Applications
ChatGPT · Codex
ChatGPT and Codex are OpenAI applications. The map uses product families so a release-version change does not imply a change in stack ownership.
OpenAI / Models
GPT
OpenAI develops GPT models and offers access through its products and API. This selected family is not a complete inventory of OpenAI models.
OpenAI / Infrastructure
Stargate
Stargate is an expanding infrastructure program involving OpenAI and partners. Some capacity is operating; the entry does not imply that OpenAI solely owns the facilities or operates a public cloud.
OpenAI / Silicon
Jalapeño
OpenAI and Broadcom unveiled Jalapeño in June 2026, with engineering samples running in the lab. The cited announcement describes planned deployment; it does not establish broad production availability.
Anthropic4 layers
Anthropic / Applications
Claude · Claude Code
Anthropic offers the Claude assistant and Claude Code application, built around its own model family.
Anthropic / Models
Claude
Anthropic develops the Claude model family. Availability varies by model and distribution platform.
Anthropic / Infrastructure
AWS · Google · Azure
Anthropic contracts compute across partners including Amazon, Google, Microsoft, and SpaceX. Dedicated capacity commitments are distinct from owning a public cloud platform.
Anthropic / Silicon
Trainium · TPU · NVIDIA
Anthropic says it trains and runs Claude on AWS Trainium, Google TPUs, and NVIDIA GPUs. Partner supplied is a sourcing description, not a capability score.