9 segments · 130 companies · as of Sep 18, 2026
AI Developer Platforms
The tools and infrastructure that turn models into working software
Where it sits
Fed by: Cloud & AI CapitalFeeds: Applications & Agents · Physical AI
Developer platforms host models, generate code, coordinate agents and measure production behavior. Enterprises pay for reliable inference, private data access, evaluations and controlled software delivery. Ownership increasingly combines cloud infrastructure with the tools developers use every day.
Why now
Agentic coding broadens into application building and software delivery. Inference vendors compete on latency and utilization, while major platforms acquire evaluation tools. China's model clouds and open-source runtime companies expand the field beyond US hyperscalers.
Managed Model Platforms
Large clouds and lab platforms expose model catalogs, private endpoints and enterprise controls. Buyers weigh model access, data residency, committed cloud spending and integration with existing data. Chinese MaaS platforms provide an important parallel market with domestic model distribution.
Bedrock distributes Nova, Claude and third-party models with enterprise controls; AgentCore adds managed runtime
leaderFoundry distributes OpenAI, Claude and open models with Azure security, data and enterprise deployment controls
leaderVertex AI and Gemini APIs combine Gemini 3.8 with partner models and enterprise data services
majorNIM and DGX Cloud Lepton package optimized model services for enterprises, clouds and GPU infrastructure partners
majorOCI Generative AI hosts Cohere and other models near enterprise database workloads
majorwatsonx.ai and Red Hat AI distribute Granite and third-party models across enterprise hybrid infrastructure
majorBailian/Model Studio serves Qwen and external models to enterprise APIs and Qoder developer workflows
majorVolcano Engine Ark serves Seed, DeepSeek and other models to Chinese enterprise and application developers
majorQianfan hosts ERNIE and external models with tuning, knowledge tools and enterprise deployment controls
majorLKE and Hunyuan APIs package model serving, retrieval and agent development for Chinese cloud customers
majorMosaic AI serves foundation and custom models beside Unity Catalog data; MosaicML provides training technology
majorCortex AI hosts partner models and governed inference beside enterprise data and Cortex Search
majorGPT-6 Astra access and GPT-5.6 APIs supply hosted reasoning and coding with enterprise usage controls
majorClaude API and cloud partnerships distribute Fable 5.1 for coding and knowledge-work applications
majorLa Plateforme serves Medium 3.5 and open models; enterprise customers can also deploy privately
challengerModelArts Studio serves Pangu and partner models with MaaS APIs, tuning and enterprise deployment controls
nicheCLOVA Studio exposes HyperCLOVA X models and tuning tools for Korean enterprise application developers
nicheParadigm deploys model-backed search and assistants against private enterprise documents and infrastructure
niche
Sources: menlovc.com, aws.amazon.com, anthropic.com, alibabacloud.com, docs.volcengine.com, databricks.com
Specialist Inference Clouds
Independent model hosts and specialized silicon providers compete on latency, throughput and deployment flexibility. Some operate their own GPU or accelerator fleets; routers instead aggregate providers. Listings and acquisitions are treated separately from technical performance claims.
Inference and fine-tuning cloud serves open models and custom deployments backed by a large GPU fleet
leaderInference and reinforcement-learning services optimize open-model deployment for application developers
leaderTruss and optimized inference engines deploy private model endpoints for production enterprise workloads
majorServerless GPUs run custom inference, batch processing and fine-tuning without customer-managed clusters
majorCS-3 inference supports OpenAI Codex-Spark and a >$20B, 750MW agreement; AWS adds distribution
majorIndependent GroqCloud expands GPU inference after the NVIDIA license and staff transfer; August funding backs deployment
majorSN50 and SambaCloud deliver model inference on proprietary dataflow accelerators for enterprise customers
challengerToken Factory serves open models including Kimi K3 on serverless and dedicated endpoints
majorW&B inference and GPU cloud connect model serving with training, evaluation and OpenPipe reinforcement learning
majorGradient and Inference serve GPT-6 Astra and DeepSeek V4.1 Flash alongside dedicated GPU endpoints
challengerHub and Inference Endpoints connect open-model developers to managed deployment; NVIDIA acquisition is pending
majorUnified model API routes requests across providers; provider choice and fallbacks avoid owning every inference fleet
majorSiliconCloud supplies optimized Chinese and open-model APIs; its planned HK IPO does not yet create a listed line
challengerManaged Inference and MemoryAlloy serve open models; September financing supports expansion beyond GPU rental
challengerCloud Inference combines NVIDIA Dynamo/NIM and NetApp integrations for enterprise model-serving deployments
challengerInference at the Edge serves open-model APIs through distributed GPU locations and dedicated enterprise endpoints
challengerServerless inference APIs and dedicated endpoints host open language, image and audio models for developers
challengerLLM and media inference APIs combine serverless endpoints with managed GPU deployment for application builders
challengerSubscription and serverless endpoints host a broad open-model catalog without customers operating GPU clusters
challengerInference Platform deploys and scales custom model endpoints over managed GPU capacity for application teams
challengerServerless diffusion infrastructure serves image, video and audio generation APIs for creative application developers
challenger
Sources: crusoe.ai, privatemarketview.com, investors.cerebras.ai, groq.com, nebius.com, docs.digitalocean.com
Inference Runtimes & Edge
Runtime software, gateways and distributed networks determine where models execute and how requests reach them. Open-source engine companies commercialize support and optimization without owning the entire community project; edge caching is distinguished from actually executing a model.
Workers AI runs edge inference; acquired Replicate adds model deployment and AI Gateway manages requests
leaderInference Cloud and NVIDIA AI Grid distribute GPU inference across cloud and edge infrastructure
majorAI Accelerator caches semantically similar API responses at the edge; it reduces calls to external model hosts
nicheTensorRT-LLM, NIM and Dynamo optimize and orchestrate distributed inference for GPU operators
majorRed Hat AI Inference commercializes vLLM-based serving on IBM Cloud and enterprise OpenShift infrastructure
majorCommercializes vLLM performance and deployment expertise; the serving engine remains an open-source project
challengerCommercializes SGLang inference and Miles RL infrastructure with NVIDIA, AMD and MediaTek investor support
challengerRay and Ray Serve orchestrate distributed model workloads; announced Nscale acquisition remains pending
majorEnterprise AI 2.8 exposes governed model endpoints, including air-gapped NVIDIA NIM for private deployments
majorAI Gateway unifies model routing, MCP policy and token-cost controls; NGINX supports inference-aware traffic
majorPrime-RL, sandboxes and distributed compute let developers train and serve models with reusable RL environments
challengerAcquired Modular MAX runtime and Mojo compiler optimize model execution across GPU and edge architectures
majorLocal model runtime packages open weights and exposes APIs for laptops, workstations and private deployments
majorONNX Runtime and Foundry Local execute models on device CPUs, GPUs and NPUs with local application APIs
majorMoAI Platform distributes model workloads across heterogeneous accelerators for training and inference
niche
Sources: qualcomm.com, akamai.com, fastly.com, nutanix.com, finance.yahoo.com, businesswire.com
Coding Agents & IDEs
Coding platforms combine repository context, terminal execution and model reasoning to create tested changes. Lab-owned tools compete with independent agents and incumbent IDEs. Chinese developer ecosystems and European tools provide meaningful alternatives to US cloud distribution.
Cursor joins SpaceX in August 2026; its coding IDE and agents connect enterprise repositories to frontier models
majorGitHub Copilot coding agents integrate Claude and Codex with repository issues, pull requests and VS Code
leaderClaude Code uses Fable 5.1 for agentic coding; the acquired Bun runtime supports its developer stack
leaderCodex turns GPT-5.6/GPT-6 capabilities into coding workflows; Cerebras-backed Codex-Spark targets low latency
majorDevin Desktop evolves Windsurf into a cloud/local agent workspace; Devin runs implementation and tests
majorAntigravity 2.0 develops interactive coding agents; Jules handles repository tasks and pull requests asynchronously
majorAgent 4 edits existing repositories, runs code and tests, and deploys applications from a hosted development environment
challengerKiro IDE 1.0 uses specs, hooks and custom agents to plan and execute coding work with AWS distribution
majorJunie works inside JetBrains IDEs and terminal with hosted or local models for repository-aware coding
majorDroids automate engineering tasks across repositories, terminals and enterprise software workflows
challengerVibe uses Medium 3.5 for coding; Devstral 2 weights remain downloadable after the hosted model was deprecated in May
challengerTrae IDE and SOLO agents package model-assisted coding for domestic and international developers
majorQoder and Qoder CN replace the Lingma brand with agentic IDE workflows linked to Alibaba model services
majorCodeBuddy IDE combines Hunyuan and external models with Tencent CloudBase deployment workflows
majorComate supplies IDE plugins and standalone agentic coding for Chinese developers and enterprise deployment
majorCodebase context and coding agents target large enterprise repositories with complex cross-file dependencies
challengerAgentic terminal combines coding, command execution and shared context for software development teams
challengerMalibu coding models target enterprise software work; reported NVIDIA licensing and staff transfers leave the lab independent
nicheCode Search, Deep Search and MCP provide repository context to Claude Code, Cursor, Codex and independent Amp
majorIndependent Amp CLI uses frontier models and Sourcegraph context; editor extensions shut down in March 2026
challengerAcquired Tabnine supplies private, context-aware coding and a context engine for enterprise quality agents
majorPull-request agents review repository changes, suggest fixes and connect code quality checks to developer workflows
challengerNative editor agents and Agent Client Protocol connect repository context, model tools and interactive code editing
challengerOpen-source editor and CLI agents read repositories, edit files and execute approved terminal and browser tools
challenger
Sources: cursor.com, replit.com, github.blog, cognition.ai, blog.google
AI App Builders & Backends
Prompt-driven application builders combine generation with previews, hosting and application state. Design platforms enter the same workflow while backend and deployment vendors capture usage as generated applications reach production. Private startup valuations are kept separate from reported revenue run rates.
Agent 4 turns prompts into applications with design, testing and deployment in a managed development workspace
leaderAgent app builder generates and operates full-stack applications with GitHub and Supabase integrations
leaderBase44 generates hosted full-stack apps; Wix Harmony applies AI models to website creation
majorMake turns designs and prompts into working applications; local-code support connects to existing projects
majorv0 generates interfaces and applications; Vercel hosting, AI SDK and Gateway capture production deployment
majorBolt uses browser-based WebContainers and models to create, run and deploy full-stack applications
challengerManaged Postgres, auth, storage and vector support supply backends for Lovable and other AI-built apps
majorFirebase Studio and Antigravity connect app generation with Google model, database and hosting services
majorPower Apps and Copilot Studio combine generated interfaces with enterprise data and managed agent workflows
majorFull-stack agents build web applications, backend services and deployment workflows from natural-language requirements
challengerVisual Flutter builder adds AI generation, backend integrations and exportable source code for application teams
challengerOpen-source visual workflows combine model prompts, retrieval and tools into deployable AI applications
challenger
Sources: docs.replit.com, lovable.dev, wix.com, investor.figma.com, vercel.com, privatemarketview.com
Observability, Evals & MLOps
Teams need to trace model calls, evaluate quality and monitor the systems around agents. Specialist tools compete with observability suites and training platforms. Completed acquisitions are assigned to parents, while Arize and Hugging Face remain separate pending their announced transactions.
Weights & Biases tracks experiments and Weave evaluates agents; OpenPipe adds serverless reinforcement learning
leaderLLM Observability traces agents and evaluates outputs; Eppo and Metaplane add experimentation and data quality
leaderLangSmith combines traces, evaluations and deployment for LangGraph agents and other application frameworks
majorOpenTelemetry-based LLM and agent tracing in Elastic Observability correlates model calls with application telemetry
majorFoundry evaluations, tracing and monitoring test agent quality, safety and production performance on Azure
majorAX and open-source Phoenix evaluate and trace agents; the Dynatrace acquisition remains subject to closing
majorEvaluation datasets, trace analysis and prompt experiments measure production-agent quality and regressions
majorAcquired Langfuse adds open-source LLM tracing and evaluation to ClickHouse's analytical data platform
majorDavis and full-stack observability monitor enterprise AI applications; announced Arize integration is prospective
majorSplunk Observability and acquired Galileo combine application telemetry with agent evaluation and AI monitoring
majorAcquired Chronosphere supplies cloud-native telemetry and AI observability after its January 2026 close
majorAcquired Observe analyzes application logs, metrics and traces on Snowflake for AI and software operations
majorGrafana Cloud OpenLIT uses OpenTelemetry to monitor LLM calls, vector databases, GPUs and MCP servers
majorAI Agent Monitoring adds agent maps and trace drill-down to existing application performance telemetry
majorSeer Agent investigates production failures using errors, spans, logs and code context inside developer workflows
majorAutomated evaluations and guardrails test agent reliability and hallucinations before and after deployment
challengerAI Observability and model monitoring track quality, drift and guardrails across predictive and generative systems
nicheStatsig brand and customer business add experimentation and feature flags; original staff remain at OpenAI
majorMLflow and Mosaic AI combine experiment tracking, agent tracing and evaluations with governed enterprise data
majorLLM Mesh and Govern evaluate, monitor and control model workflows across enterprise teams
majorHub, evaluation tooling and model artifacts support reproducible model development across an open ecosystem
majorAgent evaluations and trace grading test output quality; Promptfoo security-testing acquisition remains announced
majorLLM observability correlates prompts, model latency and distributed traces to diagnose production AI behavior
majorOpik traces and evaluates LLM applications; experiment tracking connects model changes to reproducible results
major
Sources: coreweave.com, ir.dynatrace.com, newsroom.cisco.com, investors.amplitude.com, reuters.com, datadoghq.com
Agent & Voice Infrastructure
Agent frameworks manage tools, state, permissions and long-running work. Real-time voice adds audio transport, speech models and turn handling. These platforms sell reusable infrastructure to developers, distinct from finished contact-center or enterprise application products.
Claude Agent SDK executes tool-using agents; MCP defines interoperable connections to external data and actions
leaderAgentKit, Realtime API and Apps SDK supply agent workflows, low-latency voice and tools inside ChatGPT
leaderLangGraph manages agent state and tool loops; LangSmith Deployment operates long-running production agents
leaderBedrock AgentCore supplies runtime, memory, identity and tool access for enterprise agents on AWS
majorAgent Framework and Foundry connect tools, state and enterprise identity across Azure agent deployments
majorAgent Development Kit and managed Gemini agents connect tools and execution to Google Cloud services
majorAgents framework and real-time transport power voice applications including OpenAI voice experiences
majorNova-3 speech recognition, Aura-2 synthesis and Voice Agent API provide low-latency audio intelligence
majorAgents Platform combines Eleven v3 speech with telephony, tools and real-time conversational applications
majorMaestro orchestrates AI agents, robots and people across business processes; agentic testing validates execution
majorwatsonx Orchestrate and DataStax Langflow coordinate enterprise agents, tools and governed data access
majorNeMo Agent Toolkit profiles and connects agent workflows to NVIDIA models and inference infrastructure
majorLlamaCloud and agent frameworks parse documents and retrieve enterprise context for tool-using agents
majorMulti-agent CrewAI orchestrates collaborative agents and production flows with enterprise deployment controls
challengerDurable execution resumes long-running agent workflows; OpenAI, Cursor and Netflix use Temporal Cloud
majorTyped Python agents use Pydantic AI; Logfire traces model calls and tool execution for developers
challengerConversationRelay connects live phone calls with model responses, speech and developer-controlled agent logic
majorConversational AI Engine combines real-time audio/video transport with model and speech partners
challengerAI SDK and AI Gateway connect TypeScript agents to models, streaming, tool calls and deployment
majorHoundify and Amelia 7 license voice recognition and agent actions to automotive OEMs, devices and enterprise brands
majorxUI embeds voice and agent runtimes in vehicle systems, combining local execution with cloud model access
majorSonic streaming speech and voice-agent tooling supply low-latency audio generation for real-time conversations
challengerVoice-agent API orchestrates telephony, speech providers, model responses and tool calls for developers
challengerVoice-agent platform provides phone-call orchestration, turn-taking and evaluations for production deployments
challenger
Sources: uipath.com, platform.openai.com, platform.claude.com, langchain.com, temporal.io, quantlogix.ai
DevSecOps Platforms
Established software-delivery platforms absorb AI into version control, CI, testing, artifacts and security. The buyer remains the engineering organization. Governance of generated code, model artifacts and MCP tools creates demand beyond the initial coding-assistant subscription.
Duo Agent Platform spans coding, testing and security within GitLab CI; usage-based credits monetize agent work
leaderGitHub Actions, Copilot and Advanced Security connect generated code to governed enterprise delivery pipelines
leaderArtifactory and AI Catalog govern model artifacts, MCP servers and agent skills through the software supply chain
majorBitbucket, Rovo Dev and acquired DX connect coding agents with delivery workflows and engineering measurement
majorHashiCorp Terraform/Vault and Red Hat automate infrastructure and secrets for enterprise AI deployments
majorAI-assisted delivery, testing and feature management govern software changes from code through production
majorDeveloper security and AI-code scanning detect vulnerabilities in dependencies and generated application code
majorModel Runner and MCP tooling package local models and tool servers alongside container development workflows
majorAPI development and Agent Builder connect, test and govern the services that model agents call
majorDevOps Aviator embeds generative and agentic AI in Core Software Delivery and enterprise testing workflows
nicheAIOps and incident automation translate production signals into triage and remediation for engineering teams
majorTelerik developer tools and AI-assisted testing support application creation and delivery for software teams
nicheAgentic Quality Engineering combines Tosca testing with acquired Tabnine context to validate enterprise software
majorAI Code Assurance in SonarQube scans generated code for security, reliability and maintainability before release
majorHosted and self-hosted CI agents run model-generated code through build, test and deployment pipelines
niche
Sources: about.gitlab.com, docs.jfrog.com, atlassian.com, newsroom.ibm.com, techshotsapp.com, nasdaqprivatemarket.com
Agent Tools & Sandboxes
Agents need isolated computers, browsers, search and authenticated business actions to complete work. Developers buy this execution layer separately from models; session reliability, permissions, usable web data and integration coverage determine adoption.
Cloud sandboxes isolate agent-generated code and computer tasks behind short-lived execution APIs
majorManaged browser sessions and Stagehand let agents navigate, extract information and act on websites
majorSandbox APIs provision isolated computers and file systems for agent code execution and long-running tasks
challengerSandboxes run untrusted agent code in isolated containers with programmable compute and network controls
majorOpen browser agents and hosted sessions turn model instructions into website navigation and form actions
challengerCrawl, scrape and extraction APIs convert web pages into structured, model-ready context for agents
challengerSearch and Contents APIs retrieve relevant web pages and fresh evidence for autonomous research agents
challengerAcquired Tavily provides search, extraction and web-data tools alongside Nebius agent infrastructure
challengerActors and hosted web automation supply reusable scraping and browser tools through agent integrations
challengerManaged OAuth and tool integrations let agents execute authenticated actions across business applications
challengerOAuth-aware tool execution and MCP gateways enforce user permissions when agents access connected applications
challengerBrowser Rendering, Agents SDK and Sandbox containers supply web actions, state and isolated code execution
majorMCP and app integrations expose authenticated business actions across a large SaaS connector catalog
majorWorkflow nodes, MCP connections and credentials turn agent tool calls into repeatable business-system actions
challenger
Sources: e2b.dev, docs.browserbase.com, daytona.io, developers.cloudflare.com, arcade.dev, nebius.com
Glossary
- Inference
- Running a trained model to produce an answer, prediction or tool call.
- MaaS
- Model as a service: a provider hosts models behind a managed API and charges for usage or capacity.
- MCP
- Model Context Protocol standardizes how AI applications connect to tools and external data.
- Evaluation
- Repeatable tasks and checks measure model or agent quality, safety and regressions.
- Durable execution
- Stores workflow progress so an agent can resume safely after failures, waits or human approvals.
Heat, spend, exposure and shares are editorial estimates. How to read the map ↗