AI Capital Is Leaving the Cloud and Moving Into the Real World

Over the past year, venture capital has quietly reoriented from pure software and foundation models to the physical backbone of AI: chips, robotics, energy, satellites, and secure infrastructure. This shift is especially visible where defense and commercial use cases converge – exactly the territory of dual‑use investors.
‍

From Bits to Atoms: The New AI Stack

For most of the 2020s, AI investing meant one thing: models and software. That era is ending. The new wave of capital is flowing into what many now call the “physical AI stack” – data centers, energy‑efficient hardware, spatial sensing, and robotic systems that allow AI to act in the real world, not just in the browser.

This “atoms rotation” is driven by three structural forces:

• Defensibility: physical assets, proprietary telemetry, and custom hardware create deeper moats than yet another SaaS wrapper on a public model.

• Energy and compute constraints: scaling intelligence now depends on energy, cooling, and semiconductor capacity as much as on algorithms.

• Dual‑use pull: militaries and governments want deployable capabilities – sensors, autonomy, resilient infrastructure – not just better dashboards.

For dual‑use VCs, this is where the real leverage lies: the same sensing, robotics, or infrastructure that powers industrial automation or climate monitoring can also underpin ISR, logistics, and deterrence.

‍

Defense Tech Supercycle and “Speed to Field”

Global defense budgets have entered a structural up‑cycle, but the more important change is qualitative. Procurement cultures that were historically slow and risk‑averse are pivoting to a “speed to field” mindset: get good‑enough autonomous and AI‑enabled systems deployed quickly, then iterate in the real world.

That shift is reshaping what is fundable in defense and dual‑use:

• Autonomous platforms that can be upgraded with new software and payloads across missions.

• Sensor fusion layers that integrate space, airborne, maritime, and ground data into a coherent operating picture.

• Agentic AI for planning, targeting support, logistics, and cyber defense that acts, not just alerts.

This is a fundamentally different environment for startups. Instead of waiting a decade for a single program of record, dual‑use companies can now land smaller, faster pilots and scale with both sovereign and commercial demand.

‍

Orbit as Dual‑Use Ground Truth

Space is becoming one of the clearest illustrations of the dual‑use thesis. High‑resolution thermal imaging constellations can, in the same orbit, track emissions, energy efficiency, industrial activity, and military logistics. Very‑low‑Earth‑orbit (VLEO) architectures promise higher‑quality data and more responsive sensing at significantly lower cost per satellite.

At the same time, new power concepts – from more efficient solar arrays to radioisotope‑based generators – are emerging to keep military and critical satellites online in contested environments. This is about more than connectivity; it is about resilient, persistent ground truth for both regulators and commanders.

For venture investors, the implication is clear: space is no longer a niche “space tech” vertical. It is an enabling layer for climate, industrials, insurance, and defense – and the teams that can design truly dual‑use architectures will have the most robust revenue mix.

‍

Agentic AI and Security: Identity Becomes the Perimeter

As AI systems move from passive assistants to autonomous agents, security architecture is being rewritten. The traditional idea of a network perimeter breaks down when you have thousands of machine identities – models, agents, services – all making decisions and taking actions.

Two linked trends are emerging:

• Agentic cybersecurity: systems that continuously hunt, prioritize, and remediate threats, rather than simply flagging alerts for human analysts.

• Identity‑centric security: consolidation around the idea that human and machine identity is the real control plane in AI‑native environments.

For dual‑use investors, this opens a rich seam at the intersection of cyber, DevSecOps, and defense. Critical infrastructure, defense networks, and industrial OT environments all require architectures that can securely accommodate autonomous agents without sacrificing auditability or control.

‍

Sovereign Capital and Geopolitical “Geofencing” of Technology

The new AI and deep‑tech cycle is not being driven by traditional venture alone. Sovereign funds in the Middle East, China, and multi‑country vehicles like NATO‑backed initiatives are actively shaping the map of what gets built, where, and by whom.

This creates a de‑facto “geofencing” of the technology stack:

• Semiconductor, AI infrastructure, and space supply chains are increasingly aligned with specific blocs.

• Standards, export controls, and security classifications split markets long before products reach scale.

• Liquidity paths (IPOs, strategic M&A, secondary markets) are diverging across regions.

For dual‑use VCs, portfolio construction is now inherently geopolitical. It is no longer enough to ask “is this a great company?” You must also ask “in which ecosystem can this company safely and sustainably scale?” The most successful funds will deliberately map their exposure across alliances, supply chains, and regulatory regimes – not just across sectors and stages.

‍

What This Means for Dual‑Use Investors

Putting these threads together, a new playbook for dual‑use capital is emerging:

• Prioritize the physical AI stack – robotics, sensing, power, edge compute, secure infrastructure – where commercial and defense pull are both strong.

• Underwrite “speed to field” as a core value driver: teams that can ship, learn, and harden in real‑world deployments will outperform.

• Treat space, thermal sensing, and other orbital capabilities as cross‑sector infrastructure, not exotic side bets.

• Back security and identity layers that make agentic AI safe to deploy in high‑stakes environments.

• Build portfolios with explicit geopolitical constraints and exit paths in mind from day one.

The AI cycle is not over – it is just moving closer to the physical world. For investors focused on dual‑use technologies, that is not a headwind. It is the opportunity.

___

Based on Semantic Visions data research.

‍