AMD Instinct MI455X
Compute for larger ambitions.
AMD CDNA 5 GPUs and HBM4 memory form the compute engine for large-scale inference, model training, and fine-tuning.
← Capacity reservations
AMD Helios.
The Infaio Agentic Grid · AMD Helios
AMD Helios.
Instinct MI455X.
Rack-scale compute for the next generation of AI workloads.
Plan long-term capacity around AMD Helios systems powered by Instinct MI455X accelerators, with Infaio’s workload-aware architecture connecting GPU compute, specialized inference, and agent execution.
AMD Helios · Rack-scale architecture
Think beyond the GPU.
Design for the whole rack.
Large models need compute, memory, and networking to work together. AMD Helios brings MI455X GPUs, EPYC CPUs, and Pensando networking into one rack-scale reference design.2
72 GPUsMI455X GPUs in the reference rack
432 GBHBM4 memory per GPU (up to)
31 TBApproximate aggregate HBM4 per rack
AMD EPYC CPUs
Connect the host and GPU work.
The Helios reference design integrates EPYC server CPUs alongside the GPUs, connecting host-side processing with accelerated workloads.
AMD Pensando networking
Keep the system connected.
Scale-up and scale-out networking connect the GPU domain and the wider deployment. Open rack and interconnect standards underpin the reference architecture.
On The Infaio Agentic Grid
Rack-scale capability.
Workload-level intelligence.
Infaio’s design puts GridRunner above the hardware, coordinating the model, processor, and placement for each stage of an agent workload. Helios supplies the GPU foundation; the grid connects it to specialized inference and execution services.
Plan GPU-intensive prefill and batch work alongside SN50 decode and CPU tools as part of a heterogeneous, disaggregated deployment.3
Explore specialized decodeStart with the work
Capacity with
a purpose.
Bring the model, the demand, and the outcome you need. We’ll discuss the capacity plan around them.
Frontier model inference
Give large models room to run.
Plan capacity for larger open-weight models, long contexts, and demanding inference workloads across a connected GPU system.
Multi-agent applications
Plan for many agents at once.
Discuss concurrency, response targets, and workload mix for applications running many model calls and agent tasks.
Training & fine-tuning
Build, adapt, and serve.
Explore a capacity plan spanning model development, adaptation, and serving, with the system configuration matched to each stage.
Start with the work.
Tell us about your models, applications, and expected demand so we can discuss a suitable capacity plan.
Plan where it runs.
Share your preferred deployment location and requirements for data control, networking, and proximity.
Align on the terms.
Discuss configuration, capacity, timing, and contract requirements before a reservation is confirmed.
Start the conversation
Let's plan
your capacity.
Tell us what you have in mind. Our team can discuss the right scope for your reservation.
Explore AMD MI325X This inquiry does not secure capacity. Availability, configuration, and deployment schedules are confirmed through the contract process.Thank you for your inquiry.
Your reservation inquiry has been submitted to Infaio. We’ll follow up at the email address you provided. Capacity is not reserved until a separate agreement is confirmed.
Reservation rate terms
1. Starting rates are quoted in USD per GPU-hour and apply to long-term contracts with non-refundable reservation fees. Subject to contract terms and conditions. Contract duration, allocation size, configuration, included services, fees, and deployment schedules are confirmed in the applicable agreement. A starting hourly rate is not a quote for a complete node or rack and does not imply on-demand availability.
Submitting an inquiry does not secure capacity, charge a reservation fee, or accept a contract. The Infaio Agentic Grid is in development; architecture descriptions express design intent. Hardware-specific availability and service scope must be confirmed separately.
Hardware references
2. Specifications describe AMD’s Helios reference design: 72 MI455X GPUs, up to 432 GB HBM4 per GPU, and approximately 31 TB aggregate HBM4 per rack. Architecture and workload descriptions are also documented in AMD’s CDNA 5 and Helios overview. These are reference-design characteristics, not a commitment to an exact Infaio system configuration, available inventory, or a 72-GPU reservation minimum. The $7.95 starting rate is per GPU-hour, not per rack-hour.
Deployment scope
3. Infaio’s GridRunner, GridVault, CPU-service, and GPU/SN50 integration descriptions express planned architecture. Model compatibility, licensing, precision, runtime support, networking, system size, and measured performance require deployment-specific qualification. Hardware capabilities do not establish Infaio benchmark performance, availability, or service guarantees. GPU-hour rates do not imply an all-inclusive heterogeneous system; additional processors and services are scoped in the agreement.
AMD, AMD Instinct, AMD EPYC, AMD Pensando, AMD ROCm, and the AMD logo are trademarks of Advanced Micro Devices, Inc. SambaNova and its logo are trademarks of SambaNova Systems, Inc. Manufacturer marks identify the hardware discussed and do not imply endorsement or an authorized-partner designation.