Zyphra Cloud Launches on AMD GPUs to Power Next-Gen Agentic AI

SAN FRANCISCO, CA – May 04, 2026 – AI research firm Zyphra today unveiled Zyphra Cloud, a full-stack artificial intelligence platform poised to challenge the established order of AI development. In a significant partnership, the new platform is powered by AMD's latest Instinct™ MI355X GPUs and hosted on specialized infrastructure from cloud provider TensorWave. The launch marks Zyphra's transition from a pure research entity to a commercial platform provider, aiming to equip developers and enterprises with the tools to build and deploy highly advanced, autonomous AI systems.

The platform's debut service, Zyphra Inference, is a serverless offering designed to run cutting-edge, open-weight models such as DeepSeek V3.2 and Kimi K2.6. It specifically targets what the company calls "long-horizon agentic workloads"—complex, multi-step tasks that require AI to operate autonomously over extended periods. This move signals a growing market shift beyond simple generative AI towards more sophisticated, agent-based systems capable of automating entire workflows.

"Zyphra Cloud is the natural extension of our research," said Krithik Puthalath, Founder and CEO of Zyphra, in the official announcement. "We've spent years building, optimizing, and validating AI systems on AMD infrastructure, and are now bringing that capability to market as a platform for developers and enterprises."

AMD's Strategic Strike in the GPU Market

At the heart of Zyphra's new platform lies the AMD Instinct MI355X accelerator, a strategic choice that underscores the intensifying competition in the AI hardware sector. For years, the market has been dominated by a single major player, but this collaboration represents a significant validation for AMD's role as a powerful alternative for demanding AI inference tasks.

The MI355X, built on AMD's CDNA4 architecture, boasts a standout feature: a massive 288 GB of high-bandwidth HBM3E memory. This is significantly more than competing accelerators, allowing extremely large AI models to reside on a single GPU. This hardware advantage is critical for the long-context, memory-intensive workloads that Zyphra Cloud is designed to handle, reducing latency and improving efficiency by minimizing the need to split models across multiple chips.

"AMD delivers leadership solutions, in combination with open platforms and deep industry-wide collaborations, to power the next generation of AI infrastructure," stated Negin Oliver, Corporate Vice President at AMD. "Zyphra Inference running on AMD Instinct MI355X GPUs...demonstrates how optimized AI software combined with our accelerator architecture can deliver leading AI inference performance in production environments."

This partnership provides AMD with a crucial public-facing use case, showcasing the MI355X's performance in a production environment tailored for the most advanced open-weight models. It reinforces the narrative that a more diverse and competitive hardware ecosystem is emerging, giving AI developers more choice and potentially better price-performance ratios.

The Rise of the Specialized AI Cloud

Equally important to the new platform is the infrastructure provided by TensorWave, a specialized cloud provider that has built its entire business around AMD's GPU technology. This partnership highlights a growing trend in the cloud computing market: the rise of niche providers optimized for specific, high-performance workloads, carving out a space alongside hyperscale giants like AWS, Google, and Microsoft.

TensorWave's exclusive focus on AMD allows it to create a deeply optimized stack, from liquid-cooled hardware to the software layer, designed to extract maximum performance from the Instinct GPUs. For an AI-native company like Zyphra, this offers a compelling value proposition: access to dedicated, fine-tuned infrastructure without the overhead and potential resource contention of a general-purpose cloud.

"TensorWave exists to give AI-native companies like Zyphra the dedicated, high-performance AMD compute they need without compromise," said Jeff Tatarchuk, Co-Founder and Chief Growth Officer of TensorWave. "Powering Zyphra Inference with our MI355X infrastructure is exactly the kind of partnership we built TensorWave for — enabling teams to ship production-ready AI on the latest AMD accelerators at scale."

This model suggests a maturation of the AI market, where one-size-fits-all solutions are giving way to specialized platforms that offer superior unit economics and performance for specific tasks, accelerating innovation by lowering the barrier to entry for companies working on the frontier of AI.

Powering the Next Wave of Agentic AI

Zyphra Cloud's explicit focus on "long-horizon agentic workloads" places it at the forefront of a major industry trend. Agentic AI refers to systems that can operate autonomously to achieve goals, using tools, accessing data, and executing multi-step plans with minimal human intervention. This goes far beyond chatbots that simply answer questions; these are agents that do things.

Examples of these complex workloads include agentic coding, where an AI can write, debug, and implement entire software modules; deep research, where an agent can scour vast datasets and synthesize novel insights; and long-horizon workflow automation, where an AI could manage a complex supply chain, adjusting for disruptions in real time. These tasks require a level of context awareness, planning, and persistence that has been a long-standing goal in the field of AI.

Zyphra Inference is engineered for this future, combining custom software kernels, novel algorithms for long-context inference, and advanced parallelism schemes to deliver the high-throughput, low-latency performance required. The platform's ability to efficiently handle the massive context windows and computational demands of these tasks is directly enabled by the underlying power of AMD's hardware and TensorWave's optimized environment.

From Research to Production

For Zyphra, a San Francisco-based research company that reached a $1 billion valuation in 2025, this launch represents the culmination of years of foundational work. The company has a track record of releasing open models and datasets, including its Zamba and ZUNA models, establishing credibility within the AI research community. Zyphra Cloud is the company's vehicle for commercializing this expertise, providing a bridge from theoretical research to real-world production systems.

The company has already laid out an ambitious roadmap for the platform's expansion. Future capabilities will include distributed post-training services like reinforcement learning and fine-tuning, sandboxed development environments powered by AMD EPYC™ CPUs, and direct access to dedicated GPU clusters. This vision points toward an integrated, end-to-end platform where enterprises can not only deploy advanced AI agents but also build, train, and refine them in a unified environment. With Zyphra Cloud now available, the industry will be watching closely to see how developers leverage its power to build the next generation of intelligent, autonomous systems.