Alan Fagan: Welcome to the Critical Lowdown. At EPS Global, we have been involved in open networking for over a dozen years, and we have partnered with Coherent to sell transceivers for over 25 years. We have never seen demand like we are seeing now, driven primarily by AI and, more specifically, NeoClouds.

NeoClouds are a new class of cloud providers purpose-built for AI workloads, differing from general-purpose hyperscalers like AWS or Azure. They focus heavily on GPU-dense, high-performance computing clusters. Customers such as AI startups and enterprises that need massive GPU capacity and cannot obtain it quickly from the traditional providers are turning to these NeoCloud companies.

As we address the NeoCloud market, we are bringing on new partners. One of those partners is Maia Edge. Tim Lieto, the AVP of Sales at Maia Edge, joins us today. Tim, welcome. Let's start with your career background and what brought you to Maia Edge.

Tim Lieto: Thank you for having me, Alan. About 20 years ago, I started my career with a team that set out to change how voice networking worked. What I learned then has had a profound impact on my career: the team you assemble is critical to your success. If you combine the right team with the right mission, you can achieve a great deal.

That is what brought me to Maia Edge. Some of the people I met 20 years ago at Acme Packet are involved in this venture today. The founders, Abilash Menon and Tim Ziemer, are professionals I highly respect and trust. Combining this team with our mission—to change how private connectivity is established in a distributed AI world—made this a natural fit.

Alan Fagan: Looking at the market, NeoCloud revenue cleared $25 billion in 2025, and Synergy forecasts it will reach $400 billion by 2031. You interact with these operators weekly. What changes are you seeing in how they build infrastructure now compared to 12 or 18 months ago?

Tim Lieto: The pace of change and innovation is incredibly fast. Twelve to 18 months ago, the market was dominated by single-tenant infrastructure—one massive data center built for a single hyperscale client. Today, operators need to:

• Scale across the market and acquire enterprise customers
• Support multi-tenancy and service multiple clients
• Add multiple locations, driven by power constraints

This distributed nature of AI infrastructure is driving significant change.

Alan Fagan: The growth is indeed explosive. We have been involved in open networking for a long time, which was a slow and steady market for us. In contrast, the expansion of NeoClouds is happening incredibly fast.

 A line on the Maia Edge website resonates with me: "Your AI infrastructure runs at the speed of its worst connection." Most industry discussions around AI and NeoClouds focus on GPUs, power, and cooling. Where does the network currently sit in that conversation for NeoCloud operators, and where should it sit?

Tim Lieto: Historically, the priorities have been:

1. GPU allocation
2. Power availability
3. Cooling technology
4. The network

This hierarchy is becoming outdated as we enter the distributed AI era.
AI infrastructure began with massive, centralized data centers designed for training workloads. While network challenges can be mitigated during training, we are now shifting toward inference, where network performance becomes critical. This is further compounded by the rise of agentic AI, where a single agent might call upon ten different inference pods. If even one connection point experiences poor performance or lacks the correct topology, the entire AI workflow is disrupted. Consider use cases like robotic surgery, which requires an inference-based AI cluster situated close to a hospital, or autonomous driving. These applications have vastly different network requirements than a research-based training model.

Alan Fagan: The team behind Maia Edge has a strong track record, having previously built Acme Packet and 128 Technology. When we first engaged with Maia Edge less than a year ago, this deep experience was obvious. Could you explain what the team is setting out to accomplish with this venture, and why this is the right problem to solve now?

Tim Lieto: At both Acme Packet and 128 Technology, the goal was to solve a similar underlying challenge: connecting disparate networks, whether for voice services or connecting applications to public clouds. AI is now impacting infrastructure on a much larger scale. While traditional data centers remain, this new AI infrastructure is introducing vastly more endpoints that require interconnection. Our team recognizes this challenge and is leveraging past experience to address it.

Alan Fagan: For NeoCloud operators listening, could you describe what is currently required to connect distributed AI sites and onboard tenants? What does the process look like before and after implementing Maia Edge?

Tim Lieto: Consider a newly built AI data center. Once the physical infrastructure, GPUs, and cooling are established, you need to connect it to the outside world. This typically involves securing carrier connections to a meet-me room and accessing service providers. For a single site, the process is relatively straightforward.

As customers join, you introduce multi-tenancy, which requires segmenting traffic using VRFs, VLANs, or VXLANs. While this introduces some complexity, it is manageable within traditional IP networking.
However, adding a second site—whether driven by power constraints or the need for localized workloads—exponentially increases this complexity. These sites must connect directly to one another. Setting up private dark fiber or MPLS networks is both time-consuming and expensive. As the network scales with more customers, sites, and inference points, traditional IP networking becomes a bottleneck.

Maia Edge addresses this by deploying a Path Border Controller from day one, enabling private connectivity anywhere and bypassing the complexity of traditional interconnection. This provides an AWS-like direct-connect experience, simplifying infrastructure connection and customer onboarding.

Alan Fagan: What makes this so urgent today? Why is the network between AI sites suddenly a layer that NeoClouds and sovereign operators cannot ignore?

Tim Lieto: Two main factors drive this urgency:

• Power availability: Data center power is constrained, forcing operators to build in geographically distributed locations where power is accessible.
• Inference-based AI workloads: These require proximity to end-users to minimize latency.

The combination of searching for power and distributing inference workloads naturally creates a highly distributed environment, which demands more agile networking.

Alan Fagan: The transition from training to inference is a significant shift that is not yet fully recognized. While training requires massive centralized data centers, inference relies on a larger number of smaller, localized facilities closer to the users.

Tim Lieto: Exactly. Inference is highly sensitive to latency and demands much higher network performance than training. Providing technology that enables high-performance, private connectivity across multiple sites is key.

Alan Fagan: Additionally, while traditional hyperscalers have large engineering teams to solve these problems, many operators building inference sites operate with leaner teams. They require simplified solutions, which is where Maia Edge and EPS step in.

How did this partnership come together, and why is a value-added distributor like EPS the right model for you, compared to going direct?

Tim Lieto: It is built on trust and existing relationships from our past roles. EPS actually began working with Maia Edge before I joined. This is a true partnership where we collaborate on manufacturing, components, and go-to-market strategies. EPS provides vendor-agnostic, best-of-breed solutions globally. Instead of relying on a single vendor for an entire stack, operators can select the best technology for each use case, and EPS integrates it. That model offers great value to our customers.

Alan Fagan: Maia Edge is a key piece of a larger validated stack we are developing, which encompasses optics, switching, power, cooling, and installation under a single relationship. From a customer's perspective on day one, how does this integration benefit them?

Tim Lieto: It allows customers to access the best technology for their specific needs without being locked into a single vendor for every component. They can utilize Maia Edge for WAN connectivity alongside chosen switching or optics vendors. This provides a unified experience supported by global logistics, integration, and support.

Alan Fagan: With substantial capital entering this space, including sovereign builds across Europe and the Middle East alongside massive hyperscale commitments, where do you see the primary bottlenecks over the next 12 to 18 months? Is it power, optics, networking, or another factor?

Tim Lieto: It is likely a combination of these factors. Power and networking remain major challenges. However, these constraints are driving significant innovation. We are seeing clients explore:

• Liquid immersion cooling
• Offshore tidal-powered data centers
• Carbon capture integration

The pace of innovation to bypass power limitations and logistical delays is remarkable, making this a highly dynamic space.

Alan Fagan: For a NeoCloud operator listening who recognizes this as a challenge they need to address, what is the best first step?

Tim Lieto: An initial conversation to understand their specific challenges, current projects, and how we might assist. They can engage through the EPS team or reach out to us directly. Discussing the transition from single-tenant training environments to multi-site, inference-led architectures is typically a very productive starting point.

Alan Fagan: Thank you, Tim. This has been a valuable discussion, and we appreciate the insights into Maia Edge. We look forward to seeing how the space evolves over the next year or two. Thank you again for joining us.

Tim Lieto: Thank you for having me. I appreciate the opportunity.