ATAILA Newsroom · Hybrid · Budapest · 2026-09-11

2026-W37

AWS Outposts has no current-generation GPU. So we are building the hybrid that puts it on-prem.

If you run agents on AWS and your inference must stay on your premises, you have a hardware problem: AWS Outposts offers no current-generation GPU. The only GPU family on Outposts is g4dn — NVIDIA T4, a 2019 card — and only on first-generation racks; the second-generation racks of April 2025 and the single-rack Outposts of September 2026 ship CPU families, with GPU support "coming soon" since launch. AWS's own answer for GPU at the edge is EKS Hybrid Nodes: your hardware, their control plane. That is exactly what ATAILA's GPU nodes are. So we are building it — with AWS solutions architects, on a real customer case, as a fixed-price six-week pilot.

The gap, in AWS's own words

Second-generation AWS Outposts racks (April 2025) brought C7i, M7i and R7i; C8i, M8i and R8i followed in February 2026; the second-generation single-rack Outposts of September 2026 lists the same CPU families plus accelerated-networking instances. No GPU instance appears in any of those announcements. The launch post for the second generation said that support for more instances, "including GPU-enabled instances", is coming soon — and in September 2026 it still is. The one GPU family Outposts does support, g4dn, pairs NVIDIA T4 GPUs with 2019-era CPUs, on first-generation racks only.

AWS's own path for current GPUs at the edge is different: Amazon EKS Hybrid Nodes, generally available since December 2024, join on-premises or edge machines — any hardware, any GPU you install the driver for — to an EKS control plane in the region. AWS's engineers have published GenAI inference on hybrid nodes with customer GPUs, and an NVIDIA DGX Spark joined as a hybrid node. The hardware is the customer's problem to bring. We bring it.

g4dn
The only GPU instance family on AWS Outposts — NVIDIA T4 (2019), first-generation racks only
Dec 2024
Amazon EKS Hybrid Nodes generally available — on-premises GPU nodes under an EKS control plane
Oct 2025
Amazon Bedrock AgentCore generally available — a gateway with guardrails and policy that routes inference to OpenAI-compatible endpoints
6 weeks
The pilot — design, hybrid substrate, agent baseline, decision — fixed price, credited against a Build

What we are building, with whom

This week an AWS solutions architect brought us a customer case: a European AWS customer that wants agents developed in the cloud and inference on-premises, an immutable repository with tests run on on-prem hardware, a smart gateway with inference routing, and one observability and governance plane across cloud and on-prem. We said yes the same morning. This is a technical collaboration between engineers on a customer case — not an AWS partnership, and not an AWS endorsement; when either becomes true we will write it here.

The architecture we are building, block by block:

The deliverable of the pilot is a comparison, not a slide: the same agent, the same test suite, run against the cloud model and against the on-prem GPUs, with latency, cost, quality and data residency on one dashboard. That is what a board decides on.

Amazon Web Services (AWS)
Amazon Web Services and the AWS logo are trademarks of Amazon.com, Inc. or its affiliates. ATAILA is not affiliated with, sponsored by or endorsed by AWS; this is a technical collaboration on a customer case.

1 · Week 1

Design workshop.

Target architecture agreed, model list, the test suite and its acceptance criteria, the network design, the success criteria. Remote, or in our Executive Briefing Center.

2 · Weeks 2–3

The hybrid substrate.

VPN up; an ATAILA GPU node joined to the customer's EKS cluster as a hybrid node; vLLM serving the first model; the gateway chain working end to end; OpenTelemetry into both Grafanas.

3 · Weeks 4–5

The agent baseline.

The customer's first agent containerised, in ECR and Harbor, tested on AWS for the baseline and on-prem for the comparison; guardrails and policy on the gateway.

4 · Week 6

The decision.

A report the board can read: quality, latency, cost and data residency per location, and the go/no-go for production on the customer's own hardware or on ATAILA Cloud. Fixed price for the six weeks, credited against a Build if you continue.

Where this sits in the mission

The Service Provider mission promised connectors to the hyperscalers as an option, never a dependency. AWS is the first, and it is a hybrid rather than a connector because the customer's need is a hybrid: the agents live where the developers are, the inference lives where the data is. Azure, Oracle Cloud and Huawei Cloud follow on the roadmap — each when a customer asks or a vendor commits. Every ATAILA edition runs without any of them.

We call this a technical collaboration because that is what it is. ATAILA is not an AWS partner today and AWS has not endorsed this work; if either changes, this page will say so. The pilot's comparison is handed to the customer whatever it shows — including the case where the honest answer is that the cloud model wins.

Sources

Talk to us

If you run agents on AWS and your inference has to stay on your premises — or you are the service provider whose customers do — write two sentences on the form. It reaches the founder directly.

Talk about a hybrid AI pilot →The Service Provider mission →

← Back to the Newsroom Press inquiries: contact us