When we talk about "AI infrastructure" at Neuron Factory, we mean something broader than the GPU clusters and data center architectures that dominate most infrastructure conversations. We mean the full stack of hardware, software, and intelligence required to make AI work in the physical world — not just on servers, but at the edge of the battlefield, the perimeter of the power plant, and the airspace above the city. Physical AI infrastructure is the layer that connects machine intelligence to physical reality, and it is profoundly underbuilt.
Nowhere is this gap more acute — or more strategically consequential — than in radar and aerial sensing. The proliferation of low-cost commercial drones, asymmetric threat actors, and distributed critical infrastructure has exposed massive blind spots in our collective sensing capability. Existing sensor systems — optical cameras, infrared, or legacy radar — are bulky, expensive, limited by weather and lighting, and critically, they have not incorporated modern AI. Operators cannot detect, track, or classify small and silent targets at the edge in real time. The consequences of this gap range from disrupted airports to compromised military installations to failed first-responder deployments.
At Neuron Factory, we have been tracking the convergence of AI, software-defined radio, and commercial semiconductor miniaturization that is making a new class of intelligent radar systems possible. The thesis is straightforward: the same forces that enabled the smartphone — the commoditization of advanced RF components, the maturation of on-device AI inference, and the emergence of 5G chipsets — are now enabling portable, affordable, AI-native radar systems with capabilities that were previously the exclusive domain of defense primes spending hundreds of millions of dollars.
The Market Imperative: Intelligent Sensing for a Changed Threat Landscape
The global drone market has grown from a hobbyist curiosity to a strategic inflection point in less than a decade. By 2024, there were more registered drones in the United States than registered aircraft of any other type. In conflict zones from Ukraine to the Middle East, cheap commercial drones modified for offensive use have fundamentally altered tactical realities that took decades of conventional military doctrine to establish. At the civilian level, drone incursions at airports, nuclear facilities, prisons, and sporting events have created a category of threat that existing security infrastructure simply cannot address.
The legacy response — deploying large, fixed radar installations or expensive countermeasure systems — does not scale to the distributed nature of the threat. What is needed is a new class of sensing infrastructure: intelligent, portable, networked, affordable, and capable of real-time classification at the edge. This is not a niche market. It is a category-defining opportunity across defense, homeland security, critical infrastructure protection, and urban safety applications.
The Physical AI Sensing Gap
Most deployed sensing infrastructure worldwide was designed before modern AI existed. It can detect presence but not intent. It generates data but not insight. It operates in isolation, not in networked mesh configurations that share intelligence across nodes. Filling this gap — retrofitting the physical world with AI-native sensing — is a generational infrastructure investment opportunity. We estimate the addressable market for AI-native edge sensing in defense and civilian applications will exceed $50 billion annually within the decade.
What AI-Native Radar Architecture Looks Like
The most exciting companies in this category are not just miniaturizing legacy radar — they are rebuilding radar from first principles for the AI era. The architectural signature of a genuinely AI-native radar platform has three components that legacy systems cannot replicate through software updates alone.
On-Device Intelligence
In a legacy radar system, the sensor captures raw RF data and ships it to a central processing system for analysis. The loop is too slow for real-time classification of fast-moving small targets, and the centralized architecture creates single points of failure and bandwidth bottlenecks. In an AI-native system, intelligence runs directly on each radar node — enabling instantaneous detection, classification, and tracking without dependence on centralized computing. The on-device AI stack processes and classifies signals locally, achieving response latencies that the cloud-dependent architecture cannot match.
Mesh Networking and Collective Intelligence
A single radar node has limited situational awareness. A networked mesh of nodes that share track data, fuse detections, and learn collectively has a fundamentally different capability profile. The best AI-native radar architectures treat each unit as a node in a distributed, self-healing network — scaling coverage through deployment density rather than individual sensor capability, and ensuring operational resilience when individual nodes are compromised or destroyed. This is the architecture of modern AI infrastructure applied to physical sensing: distributed, redundant, collectively intelligent.
Commercial Semiconductor Leverage
The cost revolution in radar is being driven by the same chipset commoditization that powered the mobile industry. Advanced 5G RF front-ends, combined with edge AI processors originally developed for automotive applications, have created a component ecosystem that enables enterprise-grade radar performance at a fraction of the SWaP-C (Size, Weight, Power, and Cost) of systems built on defense-specific components. Companies that architect around commercial chipsets — rather than custom ASIC development — can reach the market faster, iterate more rapidly, and achieve price points that open civilian and dual-use markets that legacy defense contractors cannot address.
The 4D Picture: Why This Technology Class Wins
The most compelling technical capability of next-generation AI-native radar is the production of a dynamic 4D map of the environment — range, bearing, velocity, and altitude — updated in real time, fused with optical data where available, and enriched by AI classification that can distinguish between a delivery drone, a hobbyist quad-copter, and a modified commercial drone carrying a payload. This is not an incremental improvement over legacy systems. It is a qualitatively different capability that enables use cases — autonomous drone-as-first-responder programs, distributed perimeter defense, real-time airspace management for urban air mobility — that were operationally impossible with prior-generation sensing infrastructure.
The data flywheel that compounds this advantage over time is particularly important from an investment perspective. As the installed base of AI-native radar nodes grows, each unit contributes new environmental data back into the company's learning models. Edge cases that confound current classification algorithms become training examples that improve accuracy for the entire network. The company that builds the largest installed base of AI-native radar nodes will have a detection accuracy and classification precision advantage that grows over time — a genuine AI moat, not just a hardware advantage.
Technical Moat Components
The defensibility of leading AI-native radar platforms rests on three reinforcing advantages: (1) the on-device AI stack, which embodies years of signal processing and machine learning co-design that cannot be quickly replicated; (2) the proprietary training dataset accumulated through deployed nodes, which creates continuously improving detection models; and (3) the mesh networking protocol and software, which determines the operational effectiveness of multi-node deployments. Each component reinforces the others, and the combination becomes significantly harder to replicate as the installed base scales.
Real-World Validation: What Commercial Traction Tells Us
The strongest signal in any deep-tech investment thesis is early commercial traction with demanding customers in real operational environments. The AI-native radar category has produced exactly this validation. Systems in this category are already operational with branches of the U.S. military, strategic defense integrators, and municipal governments running live drone monitoring programs.
The most striking civilian deployment we have tracked is a drone-as-first-responder program in a major U.S. city, where an AI-native radar network autonomously monitors tens of square miles of urban airspace, dramatically reducing emergency response times for certain incident categories. The system operates reliably across varied weather and lighting conditions — addressing the fundamental limitation of optical-only detection systems — and has been running in continuous production for over a year. Real-world operational data of this quality, at this scale, from a civilian rather than a controlled defense environment, is the kind of validation that gives us high conviction in the commercial viability of the technology.
Strategic Investor Signals
The participation of major defense prime contractors as strategic investors in AI-native radar companies is a particularly strong signal. When a company like L3 Harris — a defense integrator with deep institutional knowledge of what does and does not work in operational sensing environments — invests alongside leading technology venture funds, it communicates something that no amount of founder deck narrative can: that the technology has been evaluated at an engineering depth that only an industry insider can perform, and it passed.
The Founding Team Advantage in Deep-Tech Sensing
AI-native radar is a domain where founding team composition is a decisive determinant of long-term success. The intersection of RF engineering, semiconductor design, AI inference, and mesh networking software is genuinely rare — the individuals with serious expertise in all four domains can be counted in the hundreds globally. The companies that have assembled teams with this depth of cross-domain expertise have a talent moat that is at least as important as their technology moat.
The most compelling teams in this category are typically led by executives with prior experience scaling deep-tech companies — not first-time founders from research backgrounds — and include veterans from both defense prime contractors (for domain knowledge and customer relationships) and commercial semiconductor and AI companies (for architectural velocity and engineering execution). This combination of operational experience and technical depth is exactly what Neuron Factory targets in our deep-tech investments.
Investment Implications: What We Are Looking For
At Neuron Factory, our investment thesis in physical AI sensing is built around three evaluation criteria that we believe differentiate the category leaders from the field:
- Architecture-first design: We favor companies that have rebuilt sensing from first principles for the AI era, rather than companies retrofitting AI onto legacy RF architectures. The architecture determines the long-term capability ceiling, and companies that made the right architectural choices early will have compounding advantages as AI capabilities improve.
- Dual-use market positioning: The most durable businesses in defense-adjacent sensing will serve both military and civilian markets. Dual-use positioning provides revenue diversification, broader data collection for model improvement, and access to faster-moving commercial procurement cycles that can fund R&D without dependence on defense procurement timelines.
- Data flywheel economics: We strongly prefer companies whose deployed base contributes proprietary training data that improves product performance over time. This is the mechanism by which a sensing hardware business acquires the economics of a software business — recurring improvement in customer-facing performance without proportional cost increases.
The European defense investment cycle is particularly interesting from a timing perspective. NATO commitments are driving significant increases in member-state defense procurement budgets, and the demand for distributed, intelligent sensing infrastructure is a consistent priority across multiple NATO allies. Companies that have demonstrated operational capability in U.S. deployments and are positioned to expand into European markets have an exceptional strategic window in 2025 and 2026.
"Every generation of warfare and security has been shaped by the sensing technology available to it. The generation defined by AI-native, distributed, edge-intelligent sensing is beginning now. The companies that establish category leadership in the next three years will be providing critical infrastructure to governments and enterprises for the following thirty." — James Caldwell, Neuron Factory
What We Are Building Toward
Our conviction in physical AI sensing is rooted in a view of where AI infrastructure is going, not just where it is today. The GPU era of AI — large models running in centralized data centers — is one phase of a longer arc. The next phase moves AI inference to the edge of the physical world: into sensors, vehicles, infrastructure, and devices that need to perceive and respond to their environment in real time, without dependence on cloud connectivity, under strict power and cost constraints.
AI-native radar is an early and highly compelling instance of this broader physical AI infrastructure build-out. The companies establishing category leadership today are building the nervous system of a world in which every critical environment — airspace, border, facility, city — has persistent, intelligent, distributed sensing. That is not a niche market. It is the infrastructure of the AI-enabled physical world.
If you are building in AI-native sensing, physical AI infrastructure, or autonomous systems that depend on edge intelligence, we want to hear from you. The category is early, the talent pool is exceptional, and the commercial and strategic tailwinds are as strong as anything we have seen in our decade of investing at the frontier of machine intelligence.