Said Mia, Founder and Managing Partner at Stormbreaker Ventures – Interview Series

Said Mia is the Founding and Managing GP of Stormbreaker Ventures, an early-stage venture firm investing at the pre-seed and seed stage in connectivity infrastructure, AI-RAN, Open RAN, edge compute, satellite-cellular convergence, IoT, connected mobility, and U.S. manufacturing modernization. His GP partners include Wade Oosterman (founder of Clearnet Communications, acquired by Telus for $6.6B; former Bell Canada C-suite where he added $50B in market cap), Derek Aberle (former President of Qualcomm for nearly a decade), and Glenn Lurie (former President and CEO of AT&T Mobility and Consumer Operations). Said’s background is in telecom investment banking; he conceived Stormbreaker’s thesis and assembled the team.

You spent years advising major telecom, semiconductor, and infrastructure companies at firms like Citi and PJT Partners before founding Stormbreaker Ventures. What convinced you that the biggest opportunity was no longer advising incumbents, but backing startups building the infrastructure layer for the AI era?

The advisory work gave me a front-row seat to a slow-motion problem. I was working on deals involving some of the most consequential infrastructure on the planet — spectrum, network architecture, semiconductor supply chains — and I kept noticing the same dynamic: the incumbents understood the scale of the AI transition intellectually, but their capital allocation, M&A timelines, and organizational incentives made it structurally difficult for them to move fast enough to own the infrastructure layer that AI actually requires. They were optimizing for the existing network, not the next one.

The more I stress-tested that observation across multiple engagements, the clearer it became that the real value creation wasn’t going to happen inside those organizations. It was going to be built by founders who could operate without legacy constraints — building the connectivity fabric, edge compute architecture, and AI-native network infrastructure that a machine-driven economy genuinely demands. At a certain point, the most intellectually honest thing I could do was stop advising on that transformation and start funding it.

Stormbreaker focuses heavily on the “picks-and-shovels” layer of AI infrastructure. Why do you believe infrastructure companies may ultimately capture more durable long-term value than many application-layer AI startups?

The application layer is exciting, but it’s also structurally fragile in ways that aren’t fully priced in. When the underlying models commoditize — and they will — a lot of application-layer moats evaporate quickly. The companies that trained on GPT-4 now have to reckon with GPT-5, then GPT-6, and the switching costs are low precisely because the apps are built on someone else’s infrastructure.

The physical layer doesn’t work that way. Spectrum assets, radio access networks, edge compute deployments, satellite-to-cellular handoff infrastructure — these things are expensive to build, slow to replicate, and deeply embedded in the operational fabric of the customers they serve. The network effects are real and durable. When an industrial manufacturer integrates private 5G and edge AI into their production floor, they’re not ripping that out because a new model dropped. The infrastructure becomes load-bearing.

There’s also a temporal dimension that matters. We’re still in the first chapter of AI’s physical buildout. The gap between what the AI economy will eventually demand from network infrastructure and what exists today is enormous — and closing that gap will take a decade or more of sustained capital and innovation. That’s a very different risk profile than betting on which chatbot wins the enterprise productivity market in 18 months.

There is growing attention around Artificial Intelligence Radio Access Networks (AI-RAN), edge compute, and satellite-cellular convergence. Which of these shifts do you believe remains the most misunderstood by the broader venture community?

AI-RAN, without question. The venture community has largely processed it as a telecom story — incremental improvement to existing radio access infrastructure — and missed what it actually is: a fundamental architectural shift in how intelligence is distributed across a network.

Traditional RAN is dumb at the edge. AI-RAN changes that. You’re pushing inferencing and real-time decision-making into the radio layer itself, which has profound implications for latency, energy efficiency, and the economics of deploying autonomous AI at scale. This is the infrastructure primitive that enables a robot on a factory floor, an autonomous vehicle on a highway, or a drone in a logistics corridor to operate with sub-millisecond responsiveness without routing every decision through a hyperscaler cloud.

Most generalist VCs don’t have the telecom depth to see that distinction clearly. They understand AI at the application and data layer, but the radio access network has historically been a black box to Silicon Valley. That’s exactly the gap Stormbreaker was built to close — we have operators on our partnership team who’ve run these networks at scale, and we understand both the technical architecture and the commercial dynamics in ways that generalist investors simply don’t.

You’ve argued that legacy networks are poorly suited for a machine-driven economy. What are the biggest infrastructure bottlenecks preventing today’s telecom and industrial systems from supporting autonomous AI at scale?

There are three bottlenecks that I think about constantly.

Latency architecture. The internet and cellular networks were designed for human-speed interactions. Humans don’t notice 50 milliseconds of latency. Autonomous systems do — and in many cases it’s the difference between a safe decision and a catastrophic one. Legacy core network architecture routes traffic in ways that are simply incompatible with the deterministic, low-latency requirements of industrial AI.

The intelligence gap at the edge. Most industrial networks still treat edge infrastructure as passive transport — pipes that move data to a central compute environment for processing. But autonomous AI at scale requires local intelligence. You can’t autonomously operate a factory, port, or logistics hub if every decision has to traverse the WAN. The compute, storage, and inferencing capability needs to live at or near the point of action, and most existing deployments aren’t built that way.

Interoperability and fragmentation. Industrial environments are a patchwork of OT systems, proprietary protocols, and legacy hardware that predates modern connectivity standards by decades. Getting those systems to speak coherently to an AI layer is a massive integration challenge — one that’s easy to underestimate if you haven’t actually tried to deploy in those environments. The startups that crack this problem, that can serve as intelligent connective tissue between the physical operational world and the AI layer above it, will be enormously valuable.

Companies like Qualcomm, Ericsson, and Nokia are increasingly pursuing AI and software acquisitions. Do you see this as the start of a prolonged consolidation cycle across telecom and infrastructure technology?

Yes, and I think we’re still in the early innings of it. What you’re seeing from Qualcomm, Ericsson, and Nokia isn’t opportunistic deal-making — it’s a strategic necessity. These companies understand that the software and AI intelligence layer is where the economics of infrastructure are migrating, and none of them can build everything they need organically at the pace the market demands.

The deeper dynamic is that the AI-native startups in this space are building capabilities that incumbents genuinely cannot replicate internally — not because of talent or capital, but because of architectural freedom. When you don’t have a legacy product line to protect and a $10B customer relationship to preserve, you can design systems from first principles for the AI era. That’s a structurally different output.

From a Stormbreaker perspective, this is a meaningful part of the exit thesis for our portfolio companies. We’re building relationships with the likely strategic acquirers proactively — our GP team has personal working relationships with the decision-makers at several of these organizations, which gives us visibility into M&A roadmaps that most early-stage investors simply don’t have. When Nokia Growth Partners or Ericsson Ventures participate in a financing round, that’s not just capital validation — it’s a signal about where strategic interest is heading.

How important will edge AI become over the next five years, particularly in sectors like manufacturing, logistics, robotics, and industrial automation where latency and reliability are critical?

It will be the defining infrastructure theme of the next five years in those sectors, full stop. The question isn’t whether edge AI matters — it’s whether the companies deploying it have the underlying infrastructure to support it reliably.

Manufacturing is probably the most advanced on this curve. The economics of AI-driven quality control, predictive maintenance, and autonomous production lines are already compelling, and the large manufacturers moving fast are doing so because the ROI is measurable and immediate. The constraint isn’t the AI models — it’s the network and compute infrastructure that makes real-time inferencing on the plant floor viable.

Logistics and connected mobility are slightly earlier but will move fast. The unit economics of autonomous freight, smart port operations, and AI-driven supply chain management are massive. The important nuance for investors is that edge AI isn’t a single infrastructure category — it’s a stack. Compute, connectivity, power management, and software orchestration all have to work together at the edge, and the companies building integrated solutions across that stack, rather than point solutions, are the ones that will be most durable.

Industrial modernization and reshoring have become major policy priorities in both the U.S. and Canada. How much of the current AI infrastructure boom is being driven by geopolitics and national security concerns rather than purely commercial demand?

More than most investors in this space are willing to say publicly. I’ll be direct about it: the national security overlay is a real and structural tailwind for the infrastructure layer, not a talking point.

The U.S. government has concluded — and I think correctly — that AI leadership is not purely an economic question. It’s a sovereignty question. Who controls the physical infrastructure that AI runs on, who manufactures the semiconductors, who builds the private networks inside critical industrial facilities — these are national security decisions as much as commercial ones. The CHIPS Act, executive actions around critical infrastructure protection, the DOD’s investment posture toward edge AI and sovereign compute — these aren’t temporary political phenomena. They reflect a durable bipartisan consensus about the strategic importance of domestic infrastructure capacity.

For Stormbreaker, this creates a very specific category of company that we find compelling: infrastructure businesses where the national security use case and the commercial use case converge. The customers in those environments — federal agencies, defense contractors, critical infrastructure operators — have budget certainty and strategic patience that pure commercial customers often don’t. That combination of durable demand and mission-critical application is exactly what we want in an early-stage infrastructure company.

Your operator network includes former executives from companies like AT&T Mobility, Bell Canada, and Qualcomm. What operational insights do experienced telecom leaders bring that traditional Silicon Valley investors often overlook?

The honest answer is: a deep respect for what it actually takes to deploy and operate at scale in physical infrastructure environments.

Silicon Valley investing culture has historically been optimized for software — high margins, fast iteration cycles, relatively low capital intensity, and customer acquisition dynamics that can be modeled cleanly. Infrastructure doesn’t work that way. Network deployments have multi-year sales cycles with enterprise and carrier customers. Integration with existing OT environments is complex and slow. Regulatory touchpoints are real and can materially affect go-to-market timelines. The operator relationship — carrier acceptance, network integration agreements, spectrum access — is not a formality; it’s a strategic asset that takes years to build.

Having Wade Oosterman, who built Clearnet and then drove $50 billion in market cap creation at Bell Canada, and Glenn Lurie, who ran AT&T Mobility and knows how large carriers think about vendor relationships and technology adoption, as GPs means we can evaluate founders and companies against a standard that most early-stage investors simply can’t access. When a founder pitches us on their go-to-market with a major carrier, we can ask the right questions — not just about product-market fit, but about whether the commercial structure is realistic given how those procurement organizations actually work. That operational depth also means our portfolio companies get more than capital. They get introductions to decision-makers, guidance on navigating enterprise sales cycles, and credibility in rooms where the Stormbreaker name carries weight because of the individuals behind it.

AI infrastructure is becoming increasingly power-intensive, capital-intensive, and supply-chain dependent. What separates startups that can survive in this environment from those that may struggle despite strong technology?

The companies that survive are the ones that have found a way to make the capital intensity work in their favor, rather than treating it as a drag on the business.

Go-to-market architecture. Capital-intensive infrastructure businesses that rely solely on direct enterprise sales are slow to scale and fragile. The ones that develop distribution through carrier partners, system integrators, or government procurement channels can deploy faster and create recurring revenue streams that support their capital requirements. The business model has to be designed for the environment.

Customer profile and contract structure. If you’re building power-intensive edge compute or network infrastructure, you need customers whose buying behavior matches your capital cycle. Long-term contracts, anchor deployments, and strategic partners who co-invest in the infrastructure — these are not nice-to-haves. They’re survival requirements.

Team depth in operations, not just engineering. I’ve seen technically brilliant infrastructure startups fail because they couldn’t manage supply chain complexity, couldn’t negotiate the right vendor relationships, or couldn’t execute a large-scale deployment without burning through capital. The founders who’ve done this before — who’ve actually deployed physical infrastructure at scale — have an enormous advantage over first-time founders with strong engineering credentials alone.

Looking ahead five to ten years, what does the infrastructure stack powering the AI economy actually look like? Do you envision a future dominated by centralized hyperscalers, or something more distributed across edge networks, sovereign compute, and industrial AI systems?

Distributed — but not uniformly distributed, and not without the hyperscalers playing a role. The honest picture is a tiered architecture that’s more complex and more interesting than either the “hyperscaler wins everything” narrative or the “edge replaces cloud” counter-narrative.

The hyperscalers will continue to own the training and large-scale inference workloads. That’s a capital game that favors consolidation, and the hyperscalers have won it decisively. But inference at the point of action — on the factory floor, inside the autonomous vehicle, at the base of the cell tower, inside the sovereign government data center — will be broadly distributed. The physics of latency, the economics of bandwidth, and the politics of data sovereignty all push in the same direction: intelligence needs to live closer to where it acts.

What I find most compelling about the ten-year picture is the emergence of a new infrastructure category that doesn’t have a clean analogy in the prior generation of computing: sovereign and industrial compute networks that are private, purpose-built, and deeply integrated with the physical operational environment they serve. These are networks that a port authority, a defense contractor, a major manufacturer, or a national government builds and controls — not because the hyperscaler option doesn’t exist, but because the data sensitivity, latency requirements, and strategic importance of the workloads demand it. That’s the infrastructure layer we’re building toward at Stormbreaker. The companies that emerge as foundational to that distributed, sovereign, AI-native infrastructure stack will be among the most consequential infrastructure businesses built in this decade — and most of them haven’t been founded yet.

Thank you the great interview, readers who wish to learn more should visit Stormbreaker Ventures.