Compute & inference infrastructure
Training and inference capacity, colocation and power, networking, and the financing structures that sit under fleets of accelerators.
Sectors
Two markets where the money is real and the underwriting is hard.
Artificial intelligence and space are unusual in the same way: enormous capital requirements, long payback, and value that only makes sense on a horizon most finance is not built to look at. That is precisely why the intermediation is worth something.
01 Sector
Artificial intelligence
The financing question is nearly always the same: who funds the capex, and what is the gross margin on the other side of it.
The interesting AI companies stopped looking like software some time ago. They have fleets, contracts, power constraints, supply agreements and depreciation schedules. A generalist software investor prices them off ARR multiples and arrives at a number that is either absurdly high or absurdly low, and usually cannot tell which.
Our job is to present the business as what it actually is — frequently an infrastructure asset wearing a software company's clothes — and then find the investors who already know how to underwrite that.
Silicon & systems
Accelerators, interconnect, memory-adjacent plays, photonics, and the long tape-out cycles that make conventional venture timelines uncomfortable.
Model labs & foundational research
Frontier and domain-specific labs, where the raise is really a bet on a research trajectory and a compute commitment.
Data, evaluation & tooling
Proprietary data positions, evaluation and reliability layers, and the unglamorous infrastructure that production systems fail without.
Applied AI in hard industries
Healthcare, defense, energy, industrials and financial services — where regulation and procurement cycles are the moat and the obstacle simultaneously.
Robotics & embodied systems
Where an AI thesis meets a hardware bill of materials, and both have to be financed at once.
Capex is not a flaw in the story
It is the story. Capital intensity is a barrier to entry if the utilisation and the contracted demand hold. The work is proving they hold — and then structuring the round so that equity is not paying for assets that debt should be paying for.
Depreciation curves decide valuations
Almost every argument about an AI infrastructure business collapses into one question about the useful life of a piece of hardware. We would rather have that argument openly in the memo than have it happen without us in an investment committee.
02 Sector
Space & aerospace defense
Hardware timelines, government revenue and dual-use classification break most generalist underwriting models. All three are manageable if you know the shape of them.
Space companies get punished twice: once for the capital intensity, and again because a large part of the revenue arrives through a procurement process the average investor has never read. The result is a persistent mispricing that a specific and growing set of funds, sovereigns and strategics have organised themselves to exploit.
There are perhaps sixty institutions worldwide that will lead a serious round in this sector. We are in the business of knowing which sixty, what each of them believes this quarter, and which of them should be in your syndicate rather than merely in your inbox.
Launch & propulsion
Small and medium lift, propulsion systems, engines, and the component supply chain underneath both.
Satellite systems & constellations
Buses, payloads, communications constellations, and the capital structures that let a constellation get to break-even.
Earth observation & sensing
Optical, SAR, RF and hyperspectral — and the far harder question of who pays repeatedly for the data.
In-space services & manufacturing
Servicing, logistics, debris, and manufacturing in microgravity, where the market is real but arrives later than the pitch usually suggests.
Defense autonomy & dual-use
Autonomous systems, C4ISR-adjacent software, and companies whose export classification changes their investor universe overnight.
Ground segment & downstream
Ground stations, mission software, and analytics layers — the parts of the sector with the software margins nobody notices.
Government revenue is not lower quality. It is differently shaped.
Programs of record, IDIQ ceilings, option years and appropriation risk are all underwritable once they are laid out properly. Most decks bury them in an appendix, which reads to an investor as though the founder does not understand them.
Non-dilutive capital belongs inside the raise, not beside it
Grants, contract vehicles, strategic prepayments and equipment finance change the equity story materially. Sequenced properly they reduce the round; sequenced badly they delay it by a quarter.
03 Outside the mandate
Everything else.
Consumer, fintech, marketplaces, crypto, biotech, climate outside the aerospace overlap, and enterprise software with no frontier component. These are fine businesses and we are the wrong people for them. We keep a short list of advisors we would send our own companies to, and we will make that introduction for nothing.
04 Enquiries
Working in one of these two markets?
Tell us what you are building and what you think it needs. The first conversation costs nothing and is usually the most useful one.