Read the context.
GPUs process prompts and context in parallel, preparing the model to produce its next response.
← Capacity reservations
Frontier intelligence.
The Infaio Agentic Grid · SambaNova SN50 RDU
Frontier intelligence.
Purpose-built decode.
Reserve SambaNova SN50 RDU capacity for the next wave of interactive AI and agent workloads.
Put specialized silicon to work where it matters. Infaio’s heterogeneous, disaggregated inference design pairs GPU prefill with SN50 token generation and CPU execution, coordinated by GridRunner.
Heterogeneous, disaggregated inference
Different work.
Different silicon.
Reading a long prompt, generating a response, and running a tool are different jobs. Infaio’s design separates them, so each stage can use the compute suited to it.2
Keep the answer moving.
SN50’s reconfigurable dataflow architecture and three-tier memory are built for token generation, where moving data efficiently matters.
Turn reasoning into action.
CPU services run tools, code, and sandboxes, then return the results for the agent’s next step.
GridRunner coordinates the workload.GridVault is designed to protect data, context, and intellectual property.
Frontier open-weight models
Big models.
Bigger ambitions.
Plan deployments around leading open-weight models such as Kimi K3 and GLM 5.3, with a compute architecture designed for the demands of multi-step agents.3
Kimi K3
Explore long-horizon coding, reasoning, and knowledge-work applications with native multimodal capabilities.
GLM 5.3
Explore complex software engineering and long-horizon agent tasks with an open-weight model built for sustained work.
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 Helios 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 chip-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.
Architecture references
2. SN50’s role as purpose-built decode silicon and its reconfigurable dataflow architecture with SRAM, HBM, and DDR memory are described by SambaNova. The GPU-prefill / specialized-decode split is also described by General Compute. Infaio’s GPU, SN50, and CPU arrangement is a planned architecture, not a claim that General Compute’s measured results apply to Infaio. No model-specific speed, throughput, or savings guarantee is implied. The $8.95 rate is per SN50 chip-hour; GPU prefill, CPU services, and other components are scoped separately in the contract.
Model deployment notes
3. Model examples are deployment targets, not a live supported-model catalog. Model descriptions follow Moonshot AI’s Kimi K3 model card and Z.ai’s GLM 5.3 documentation. These sources establish model characteristics, not verified SN50 compatibility. Model licensing, compiler/runtime support, precision, context limits, system size, and performance must be qualified for the selected deployment before service commitments. Open weights do not imply unrestricted commercial licensing.
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.