Source: futurumgroup.com

IXP.us, Hunter Newby's joint venture with Connected Nation, is featured in this Futurum Group analyst report on the QAI Moon partnership and its edge AI infrastructure strategy.

---

by Brendan Burke

Moonshot Energy, QumulusAI, and IXP.us have formed a joint venture called QAI Moon to design and deploy a nationally distributed platform that pairs carrier-neutral internet exchange points with modular GPU clusters. QAI Moon's AI Pod modular design aims to compress the construction timeline that has so far delayed the data center industry from meeting demand. The benefits of placing AI chips closer to customers include higher reliability and lower latency.

What is Covered in this Article

• Why Time to First Token is emerging as the core user experience metric for AI in 2026.
• How edge AI data centers at carrier-neutral internet exchange points can reduce latency and network hops.
• Details from the Moonshot, QumulusAI, and IXP.us agreement and rollout plan.

The News: QAI Moon formed as a joint venture among Moonshot Energy, a Texas-based manufacturer of modular electrical and AI infrastructure; QumulusAI, a provider of inference-optimized GPU-as-a-Service; and IXP.us, a developer of Internet Exchange Points (IXPs). The joint venture will design and deploy a nationally distributed platform that pairs carrier-neutral IXPs with modular GPU clusters at 25 initial sites, with plans to scale to 125 across U.S. university research campuses and municipalities. The first deployment is slated to begin by July 2026 at Wichita State University, with further expansion planned as IXP.us markets come online.

Will QAI Moon Beat Hyperscalers in GPU Latency?

Analyst Take: Neocloud Moving to the Edge — This joint venture demonstrates that AI neoclouds can meet the insatiable demand for compute by bringing GPUs closer to customers. QumulusAI is an AI infrastructure company specializing in the rapid deployment of GPU servers with ultra-low latency and distributed storage. Neoclouds stand out for their close coupling of customer workloads with sensitive compute clusters via advanced software scheduling, customer service, and access to the latest AI accelerators. With hyperscale GPU resources in high demand, 2026 presents a prime opportunity to innovate in AI deployment methods.

The QAI Moon AI Pod aims to compress the construction timeline that has delayed the data center industry from meeting demand. These timelines typically extend 2-3 years, due to labor and permitting constraints. AI Pods are pre-engineered, factory-built modules with a rapid, modular deployment model designed to bring AI compute online in months, according to QumulusAI. They are designed to integrate with sites that have available or stranded energy.

Each QAI Moon AI Pod deployment is engineered as a network-dense, low-latency inference platform optimized for QumulusAI's GPU-as-a-Service model. To achieve this, every site is provisioned with dual, geographically diverse 400G IP transit connections sourced from four independent ISPs, ensuring robust connectivity and route redundancy. The architecture includes direct high-count dark fiber adjacency between the IXP's interconnection infrastructure and the modular AI compute environment.

Lowering Latency with Carrier-Neutral Connectivity

Time to First Token (TTFT) is a key metric in AI inference, and this announcement proposes reducing latency by moving inference to the edge of the interconnection. TTFT refers to the latency between a user's request and the model's initial response. A tight chain governs it: user-to-model round-trip time, queuing delays, model cold starts, and cross-region hairpinning. By colocating inference capacity directly at carrier-neutral IXPs, the QAI Moon architecture aims to reduce network hops and variability while keeping models warm and close to demand.

While hyperscale and CDN edge architectures move compute closer to users than legacy models, QumulusAI's IXP pods uniquely combine carrier-neutral physical location, dense GPU-as-a-service readiness, and an explicit focus on optimizing low-latency AI inference. This makes them one of the first AI-native edge designs directly targeting the next wave of distributed, latency-sensitive inference workloads.

From CDN to IXP

Building on carrier-neutral IXPs aligns compute placement with traffic reality. Rather than backhauling requests to distant regions, the QAI Moon model places inference in local area networks interconnected across physical buildings, which should cut route length and variability. Most hyperscale and CDN nodes are in operator-controlled facilities or colocation centers, while QAI Moon AI Pods are sited directly at sites where compute is most needed. For TTFT, fewer network domains typically means a snappier first byte and better tail latency. That's precisely what reasoning-heavy and Mixture-of-Experts (MoE) architectures require for critical workflows, such as academic research and voice interaction.

By hosting AI inference capacity at local IXPs, university researchers, students, and municipal data-driven projects can access high-performance AI resources with minimal network hops and latency. This enables real-time analysis, accelerated simulations, and interactive AI-driven applications. IXP-based AI deployments offer robust, redundant connections to multiple networks, enhancing uptime and reliability for critical services, educational platforms, and smart infrastructure. Being at the forefront of edge AI infrastructure enables universities and municipalities to lead in implementing AI agents.

What to Watch

• Universities and municipalities are planning to take advantage of the practical benefits of agentic AI via on-site data center capacity to supplement cloud computing.
• Neoclouds will differentiate based on partnerships with energy service providers and real estate developers.
• A shortage of GPUs will encourage continued innovation in deployment methods as the data center industry moves into the inference phase of AI.

Continue Reading