Zyphra and AMD launch open AI cloud powered by Instinct MI355X GPUs

Zyphra has partnered with AMD to launch a new AI cloud platform built around open weight models and AMD Instinct GPUs.

The platform, called Zyphra Cloud, is designed for high speed inference on frontier open models such as DeepSeek V3.2, Kimi K2.6, and GLM 5.1. Zyphra says the service uses custom kernels, long context inference methods, and advanced parallelism to improve throughput and reduce latency.

The goal is to support workloads such as agentic AI, deep research, and long running AI tasks that need fast responses across large context windows.

The platform runs on TensorWave infrastructure using thousands of AMD Instinct accelerators. Zyphra Cloud will use 15MW of compute from TensorWave’s MI355X GPU installation, with plans to expand to future AMD GPUs such as MI450 and later generations.

This is important for AMD because it shows another AI company building a production cloud service on Instinct hardware instead of NVIDIA GPUs. AMD has been trying to grow its role in AI infrastructure, and partnerships like this help prove that its hardware and software stack can support large scale inference.

Here is the main picture:

Zyphra does not want the platform to stop at inference. The company plans to expand it into a broader AI platform with reinforcement learning and fine tuning support. Those future capabilities will use AMD EPYC CPUs and dedicated GPU clusters.

TensorWave says its goal is to give AI native companies dedicated access to high performance AMD compute. The company previously announced plans to build one of the world’s largest AMD GPU clusters, using accelerators such as MI300X, MI325X, and MI350X.

Zyphra is also building its own model lineup. The company has introduced ZAYA1 8B for reasoning, ZAYA1 74B as a mixture of experts model with up to 74 billion parameters, and ZAYA1 VL as its first vision language model.

The comparison to DeepSeek is clear. Zyphra is positioning itself as a US based open AI platform with strong inference infrastructure and open model support. Instead of focusing only on closed models, it is trying to make open weight AI faster and easier to deploy at scale.

For AMD, this is another useful win. The company has powerful AI GPUs, but it still needs more visible cloud deployments, better software support, and more customers proving that Instinct can handle real production workloads.

Zyphra Cloud gives AMD another example of that. If the platform performs well and expands to MI450 and beyond, it could help AMD become a stronger alternative in AI inference, especially for companies that want open model infrastructure without being locked into NVIDIA’s ecosystem.