Replicate is joining Cloudflare

Replicate, the platform for running AI models, is joining Cloudflare. The two companies have been circling the same problem from different angles: Cloudflare's Workers platform aims to make full-stack application development and deployment frictionless, while Replicate has focused on making AI model deployment as simple as a single line of code. By folding Replicate's tools directly into Cloudflare, the combined platform is positioned to become the default destination for building and shipping AI and agentic workloads.

For existing users of either platform, the change is designed to be seamless. Replicate's APIs and workflows will keep working without interruption, and will gain performance and reliability from Cloudflare's global network. Workers AI users, meanwhile, can expect a substantially larger model catalog and, for the first time, the ability to run fine-tuned and custom models directly on the platform.

From machine learning to the open-source boom

For decades, machine learning was a specialized discipline concentrated in a handful of well-resourced research labs. Models were monolithic, data was proprietary, and the tooling was out of reach for most developers. That changed when the open-source ethos that built the modern internet collided with ML, as researchers began publishing not just papers but model weights and code.

The result was an explosion of iteration. Image generation went from blurry curiosities to photorealistic output almost overnight, powered by open models like Stable Diffusion. Today, Replicate's catalog contains thousands of image models, each building on the last, and similar progress has rippled across video, audio, and language models.

That community-driven acceleration created a practical bottleneck: actually running these models. Each new model carries different dependencies, demands specific GPU hardware, and needs complex serving infrastructure to scale. Developers were spending more time wrestling with CUDA drivers and requirements.txt than building applications. Replicate solved this by abstracting the complexity away — using its open-source tool Cog to package models into standard, reproducible containers — so that any developer could run even the most complex models with a simple API call.

Replicate's catalog now spans more than 50,000 open-source and fine-tuned models. Its marketplace also provides access to leading proprietary models such as GPT-5 and Claude Sonnet through the same unified API. Beyond inference, Replicate built a community hub where developers can discover, share, fine-tune, and experiment with models in a public playground — a space that Cloudflare says it will continue to invest in and grow.

The AI cloud takes shape

Cloudflare's developer platform mission has always been to let developers build full-stack applications without managing infrastructure. AI has changed what those applications look like — three years ago, few were building agents or generating launch videos with AI. The infrastructure they need has shifted accordingly.

Cloudflare has been assembling the pieces of what it calls the AI cloud, designed to run inference at the edge, close to users:

  • Workers AI: serverless GPU inference across the global network
  • AI Gateway: a control plane for caching, rate-limiting, and observing any AI API
  • Data Stack: including Vectorize vector database and R2 for model and data storage
  • Orchestration: tools like AI Search (formerly Autorag), Agents, and Workflows for multi-step applications
  • Foundation: built on Workers, Durable Objects, and the rest of the core developer platform

Replicate brings one of the industry's largest model catalogs and developer communities to that stack. Cloudflare contributes its global network and serverless inference platform. The result, as the companies frame it, is the best of both: the most comprehensive selection of models running on a fast, reliable, and affordable inference platform.

One platform, all models

The near-term roadmap centers on bringing the entire Replicate catalog — all 50,000+ models and fine-tunes — to Workers AI. Developers will be able to run models in Replicate's flexible environment or on Cloudflare's serverless platform from one place.

Beyond catalog expansion, fine-tuning capabilities are coming to Workers AI, powered by Replicate's expertise. Workers AI is also becoming more flexible, with support for bringing custom models to the network. Cloudflare will leverage Cog to keep that process seamless and reproducible.

The deeper integration extends beyond inference. With Replicate's catalog wired into the full Cloudflare developer platform, developers can store model output directly in R2 or Vectorize, trigger inference from a Worker or Queue, use Durable Objects to manage state for an AI agent, or build real-time generative UI with WebRTC and WebSockets. The unified inference platform will also be integrated deeply with AI Gateway, offering a single control plane for observability, prompt management, A/B testing, and cost analytics across all models — whether they run on Cloudflare, Replicate, or any other provider.

With Replicate's team joining Cloudflare, the companies intend to build that future together.