Cirrascale Cloud Services launches production release of the Cirrascale Inference Platform, delivering a complete enterprise AI inference stack. New release pairs turnkey enterprise AI capabilities and governance with support for model routing, fine-tuning, and accelerator selection across NVIDIA, AMD, Qualcomm, and Tenstorrent hardware. Sep 16, 2026 Prev Next 1 of 43,633 Cirrascale Cloud Services, the expert neocloud built for Private AI, announced the production release of the Cirrascale Inference Platform, a complete software stack for enterprise-grade AI inference. From a single serverless platform, enterprises can run open-source models, their own private models, and closed model ecosystems, including Google Gemini delivered on premises through Google Distributed Cloud and operated by Cirrascale. "Cirrascale Inference platform gives customers working applications, spend controls, and agent guardrails on day one, running privately on the accelerator that makes the most sense for their workload." - Alex Nataros, CTO at Cirrascale Cloud Services The platform is built for the workloads enterprises are deploying today: search, chatbots and copilots, agentic AI services, coding assistants, document intelligence, and video generation. Every pipeline is managed from a single web console and connects securely where existing data and services exist, in on-premises and hyperscaler environments. Its web front application, inference AI capabilities, and security layer remove the months of integration work that usually stand between raw infrastructure and a tool employees can use. Enterprises get a turnkey private chat experience connected to their own knowledge base, built-in controls to manage AI spend across teams, and governance guardrails for agentic workloads, all aligned with HIPAA, SOC 2, and FedRAMP requirements where required. "Enterprises do not struggle to stand up a model endpoint anymore. They struggle with everything around it: governance, cost control, and getting a secure application in front of employees," said Alex Nataros, CTO at Cirrascale Cloud Services. "This release closes that gap. Customers get working applications, spend controls, and agent guardrails on day one, running privately on the accelerator that makes the most sense for their workload." The optimization layer of the Cirrascale Inference Platform is built to get the most out of every accelerator. It maximizes throughput, keeps latency stable as demand spikes, and lets customers run larger models on existing hardware. The result is more tokens per GPU dollar and predictable cost per token across every workload. Additionally, the model and hardware selection layer ends the tradeoff between model choice and infrastructure commitment. The platform automatically routes each request to the right model and runs it on the best available accelerator, whether that be NVIDIA, AMD, Tenstorrent, or Qualcomm, with no code changes required to switch, giving organizations flexibility to deploy leading open models on the most appropriate hardware. Teams can also fine-tune models on their own private data without that data ever leaving their environment. That combination sets Cirrascale apart from both hyperscalers and other GPU clouds. No other provider pairs automated, multi-vendor accelerator selection with Private AI delivery of closed models like Gemini and data centers connected close to where customers already operate. "Hyperscalers give you their models on their hardware. Most GPU clouds give you one vendor's silicon and leave the software to you," said Dave Driggers, CEO and co-founder of Cirrascale Cloud Services. "We built the Cirrascale Inference Platform so enterprises never have to make that choice. They pick the model and we put it on the best hardware for the job, in a private environment, at a price their CFO can plan around."

Cirrascale Cloud Services announced support for AMD's Helios rackscale solution and Instinct MI400 Series GPUs across its AI innovation cloud. The platform brings AMD's rack-scale AI infrastructure, powered by Instinct MI455X GPUs, to customers running large-scale inference and frontier-model training. AMD Helios is a 72-GPU rack-scale platform combining AMD Instinct GPUs, 6th Gen EPYC CPUs, Pensando networking, and ROCm software. At rack scale, it delivers up to 2.9 exaFLOPS of peak 4-bit compute and 31 TB of HBM4 memory in a single rack. The platform scales using UALink over Ethernet and standards-based Ethernet aligned with the Ultra Ethernet Consortium. Cirrascale will offer the capacity as dedicated, bare-metal services when the platform reaches general availability.

Cirrascale advances open AI infrastructure with AMD Helios rackscale solution and AMD Instinct(TM) MI400 Series GPUs. Business Wire Cirrascale brings AMD Instinct(TM) MI400 Series GPUs to its AI innovation cloud, spanning the AMD Helios rackscale solution for frontier AI and AMD Instinct(TM) MI430X GPUs for sovereign AI and HPC, all on open, standards-based infrastructure SAN DIEGO, July 14, 2026-(BUSINESS WIRE)-Cirrascale Cloud Services, the expert neocloud built for Private AI, today announced support for the AMD Helios rackscale solution and AMD Instinct(TM) MI400 Series GPUs across its AI innovation cloud. The addition brings AMD rack-scale AI infrastructure, powered by AMD Instinct(TM) MI455X GPUs, to Cirrascale customers running large-scale inference, frontier-model training, and fine-tuning, and it extends Cirrascale's open, multi-vendor approach to accelerated computing. AMD Helios is a rack-scale AI infrastructure platform for frontier AI development and AI factory deployments. It brings AMD Instinct GPUs, 6th Gen AMD EPYC(TM) CPUs, AMD Pensando(TM) networking, and AMD ROCm(TM) software together in a unified, 72-GPU rack-scale platform built from OCP ORW-aligned 4-GPU compute trays. Defense-in-depth security, virtualization, and rack-scale software acceleration round out a system designed on open standards, so customers can deploy and scale interoperable AI infrastructure from a single rack to cluster scale. At rack scale, AMD Helios delivers exaflop-class AI performance, with up to 2.9 exaFLOPS of peak 4-bit (OCP MXFP4) compute and 31 TB of HBM4 memory in a single rack. It scales up inside the rack over UALink(TM) over Ethernet (UALoE) and scales out across racks and clusters on standards-based Ethernet aligned with the Ultra Ethernet Consortium (UEC), giving customers a path to grow without proprietary fabric lock-in. The broader AMD Instinct MI400 Series also includes AMD Instinct(TM) MI430X GPUs, built for sovereign AI and HPC. Combining leadership performance for simulation and modeling with advanced AI capabilities in open, trusted systems, MI430X GPUs give governments, research institutions, and national labs a path to scale private AI compute with control and confidence. As the expert neocloud, Cirrascale delivers these platforms as dedicated, bare-metal capacity backed by hands-on support, backed by hands-on expert support. That approach gives customers the freedom to match accelerator technology to the workload rather than commit to a single vendor or a closed stack, reflecting how AMD built AMD Helios: on open standards and standards-based Ethernet, designed for interoperability and choice. "Our customers are building at a scale where infrastructure choices carry real weight, and they want to match the accelerator to the workload instead of being locked into one path," said Alex Nataros, chief technology officer at Cirrascale Cloud Services. "The AMD Helios rackscale solution fits how we think about the cloud: open, standards-based, and built to run the most demanding inference and training jobs without forcing customers into a closed system."