The moat is physical
Hardware in orbit, real-world data and hard-won licenses take years to replicate. That is the kind of advantage investors will reward next.
Thesis
AI is leaving the screen. Space is its hardest test.
Software captured the last decade. The money is now moving into atoms: chips, data centers, power, and machines that act in the world. Space is where every layer of that stack runs under the hardest constraints, and it is where we focus our advising and investing.
Follow the money
From bits, to atoms, to orbit.
-
2010s to 2024
Bits
Software eats the world: SaaS, apps and per-seat pricing.
Investors reward Growth and seats
-
2025 to 2026
Bits buy atoms
AI money pours into chips, data centers and power.
Investors reward Revenue and retention, not demos
-
2027 and beyond
Bits drive atoms
AI goes into the real world: robotaxis already on the road, then robots, devices, AI-designed drugs.
Investors reward A physical moat: hardware, real-world data, regulation
-
Our focus
Space
Power, compute, bandwidth and distance, all pushed to their limits.
Investors reward Technology that works, and a team that can sell it
Software alone gets copied in months. Software plus the real world is hard to copy. Space is the hardest real world there is.
The stack in space
Five layers, one system, no slack.
AI depends on energy, chips, infrastructure, models and applications. Push on one layer and the others move. Each layer has a space version, and each one is harder.
The AI industrial stack, in space
Space is the whole AI stack, under its hardest constraints.
- Applications & Embodied AI. In space: Earth observation, comms, servicing.
- Models. In space: Onboard autonomy, far from help.
- Infrastructure. In space: Ground stations, orbital compute.
- Chips & Compute. In space: Radiation-tolerant edge inference.
- Energy. In space: Solar power, every watt budgeted.
Freedom with AI participates in three ways: Build, Advise and Invest.
-
Applications & Embodied AI
On Earth Products are no longer limited to screens. They include robots, vehicles, scientific tools and other systems that sense, decide and act.
In space Imagery, connectivity, logistics and servicing in orbit: products whose customers are on the ground and whose hardware is not.
-
Models
On Earth Frontier, open-source and domain-specific models. Their capability, cost and deployment constraints shape what product teams can build.
In space Spacecraft cannot wait for a round trip to Earth to make every decision. Autonomy and onboard analysis move from nice to have to necessary.
-
Infrastructure
On Earth Data centers, networking, cooling, orchestration and security determine whether models run reliably and affordably at scale.
In space Downlinks, ground networks and compute in orbit decide how much data actually reaches a customer, and how fast.
-
Chips & Compute
On Earth Processors, memory, interconnects and access to supply shape performance, cost and who can compete.
In space Compute has to survive radiation, heat and a tight power budget, and still run useful models at the edge.
-
Energy
On Earth Every workload consumes power and produces heat. Generation, siting and cooling affect where capacity can be built and how quickly it can grow.
In space Power is the constraint under everything on Earth. In orbit it is literally finite, and every subsystem competes for it.
Why space
Where hard technology needs company builders.
Every AI constraint is sharper
Power, compute, bandwidth and latency are the limits of AI on Earth. In space they are tighter, so the companies that solve them build things others cannot.
The business is the bottleneck
Many space companies have the engineering. Fewer have a clear first customer, a sales motion that survives procurement, and a story investors outside the sector can follow.
Where we work
The company-building layer.
We do not design spacecraft, chips or power systems. Our work begins where a technical capability has to become a product, find a buyer and support a durable company. Those questions repeat across the stack, even when the technology does not.
- Capability
- Product
- Market
- Enduring business
Pieces
Behind the thesis.
The five-layer stack builds on Jensen Huang’s framing of AI as infrastructure. The space mapping and the company-building layer are our own. NVIDIA on the five layers of AI