Inception Capital for Physical AI

First check in. Then we compress the time to validation.

Physical AI is intelligent systems that operate in the real world. AI models are well funded. We invest in the deployment layer beneath them: chips, power, machines, and the supply chain that feeds them.

We back technical founders between a lab result and a first paid pilot, before the venture ecosystem recognizes them. We write the first check, then work alongside the team on government contracts, strategic partners, and first industrial customers.

We plan milestones against real industrial roadmaps, so each one moves a buyer closer to a purchase order.

Most pre-seed capital is passive. It arrives, then waits for signal. In long-cycle industrial domains, that gap is where companies stall. PaperJet was built to close it.

Chris Kong, Founder and General Partner, Bay Area

$Technology layer over fractured infrastructure$

The thesis

The technology works.
The system around it doesn't.

Physical AI stalls in the two years after the demo. Not because the hardware fails. Because procurement, supply chain, and capital sequencing were never built for what comes next.

Logos impress venture. Roadmaps decide industry. We invest accordingly.

Where Physical AI companies stall

Physical AI lives in factories, power plants, and machines. When intelligence moves into manufacturing, robotics, and energy systems, the build cycles are long, the capital requirements are real, and the path to a credible customer looks nothing like a SaaS pilot. Most venture capital is not built for this.

Money arrives after the build

Software funds wait for traction. A hardware company has none until it has built and tested something, and that costs money first. The check a founder needs most is the one most funds decline to write.

Each pilot starts from zero

A pilot with no follow-on contract ends when the pilot ends. The company keeps the lessons and loses the customer, so cost per unit stays where it started. Anchor customers and repeatable deployments change that.

Founders learn procurement on their first deal

Government contracting, industrial purchasing cycles, and supplier qualification each take years to learn. Most technical founders have not done them. Most investors have not either.

Venture pacing and industrial pacing disagree

Venture rewards a story that moves every quarter. An industrial buyer works from a roadmap that moves every year. Founders who chase the quarterly story can miss the milestone a buyer is waiting for.

Why PaperJet exists

Capital waits for traction. A hardware company cannot produce traction until it has built something. We enter in that gap, after a lab result and before a first paid pilot, then work the system around the company, inside the process and alongside the team.

01Source

Find them before the traction shows

We source from research labs, university programs, and dual-use operators. These teams have a result but no revenue yet, so traditional deal flow does not surface them.

02Underwrite

Test whether it survives a first customer

We check feasibility, whether suppliers can deliver, and what deployment costs. The test is whether the company can survive its first real customer relationship.

03Build

Compress the time to validation

Government contracts, strategic partner introductions, CFO-level financial discipline, our Asia sourcing network, and milestones sequenced against industrial roadmaps. We work side by side with founders.

04Scale

Connect them to the buyers

We introduce portfolio companies to CVCs, OEMs, defense primes, and industrial operators. These buyers decide whether a Physical AI system reaches production or stalls.

Where the thesis is in motion

Each company below required more than capital to reach its next milestone. Government relationships, OEM access, CFO-level discipline, supply chain readiness. That work is what PaperJet provides from day one.

Apollo Atomics

Nuclear

Infrastructure components for next-generation nuclear developers.

First VC fund check. We sequenced early dual-use traction so government credibility preceded commercial outreach. In nuclear, that order matters.

Noble Machines

Robotics

Humanoid robots for rugged industrial environments. Construction, fabs, energy, mining.

First VC fund check. We connected the team to strategic investors who understood industrial deployment before the company needed outside validation to open doors.

Arrakis Technologies

AI Infra

Foundry-scale photonic chip packaging. Solving the hardest version first for defense, then scaling to AI and space.

PaperJet entered before the market recognized the structural gap. Our role is milestone sequencing and ecosystem positioning around the chip industry's packaging transition.

Auriga Space

Space Launch

Hypersonic testing and space launch systems with structural DoD demand.

We worked the DoD relationship actively. In defense-adjacent markets, the difference between an introduction and a contract is sustained follow-through.

H3X

Electrification

The lightest propulsion system for electric aviation.

We embedded CFO-level support through the OEM relationship process. Staying in the room through financial structuring until adoption followed.

Where we invest

Physical AI is three compounding layers. The returns come from how they interact.

Chips

Packaging, photonics, and compute substrate. Models get the funding. The hardware they run on sets how fast anyone can deploy.

Power

Nuclear, electrification, and energy infrastructure. The longest cycle. Where the physical limits on AI infrastructure are actually built.

Machines

Robotics, edge compute, and space systems. The most visible layer and the most crowded. We work one level below the demand curve, on the bottlenecks that decide who actually deploys.

Most investors chase the demand curve. We work one layer below it, on the bottlenecks that decide who actually deploys, and on the supply chain that feeds all three.

Request for startups

Specific gaps we'd fund and are willing to be wrong about.

RFS-1 · Robotics · Open · Updated October 2026

The gear inside a robot's arm

The problem

Many Western humanoids run their arm and wrist joints through a strain wave gearbox. It is typically 30 to 50 percent of each joint's actuator cost.

What's unclaimed

Japan sets the quality standard. Chinese suppliers sell for 30 to 60 percent less. Schaeffler plans to mass-produce a cheaper formed version in Germany from 2027. Two US-owned companies make these gearboxes, and neither has disclosed a humanoid customer.

What we'd fund

A US team building strain wave gearboxes for humanoid joints, with a credible path to Chinese pricing at volume.

What would close this

A US team already pricing near Chinese suppliers
A joint design that shrinks the gearbox's share of cost
A US supplier naming a humanoid customer

The open question

A domestic gearbox has to compete with China, not Japan. A federal rule from July pushes new mobile robots toward mostly US parts, but today a plan satisfies it. New models after January 2028 have to meet it outright.

Building this? Tell us. Think we have it wrong? Email contact@paperjet.vc.

Sources: McKinsey, 36Kr, The Robot Report, FCC

How we think

The fund is built on a specific view of how industrial systems evolve, where capital pressure builds, and what it takes for a Physical AI company to survive its first deployment cycle. These essays, written by Chris Kong, make that thinking explicit.

The M&A Clock

How we think about acquisition timing before the comp exists.

Sep 2026

The Fifth Filter

Why the pre-seed window in Physical AI is not about finding the best product.

Jun 2026

Orbital Wedge

Where AI compute belongs in space, and where it does not. A conditional infrastructure thesis.

Mar 2026

Under-Automated, Overlooked

Why robotics keeps missing the manufacturing capital cycle and what changes when it does not.

Jan 2026

Designing an Advanced Manufacturing Strategy for the AI Era

Why Industrial AI needs policy, industry, and venture to move on the same clock.

Jan 2026

Read all essays

For founders

We write the first check. Then we build the system around you.

Inception capital for Physical AI, between a lab result and a first paid pilot. We back technical founders in chips, power, machines, and the supply chain that feeds them at the moment when the technology is real but the system around it does not exist yet. That is the window we are built for.

Best fit: US companies, and founders abroad who plan to incorporate in the US. We prefer pre-seed, and stage is a preference, not a rule. First customers in the US help. Software fits when it runs a physical system. General B2B software does not.

Tell us what you are building →

A short form. Referred by someone? Email contact@paperjet.vc.

For allocators

Allocated to the physical world?

PaperJet backs Physical AI before consensus forms. The conversation starts at the thesis level.

Review the case →