2022
The first check
Gothic was founded in New York with $64M and one researcher on staff. The first three investments were made before the firm had an office; the diligence memos from that year still set the template we use today.
Fund I · $64M
Built to outlast the cycle.
Gothic Holdings began in New York in 2022 with two convictions: that AI-native companies would need investors who could read the code, and that the firm doing it should be built to still be here in thirty years.
The firm
Four years, deliberately slowly.
We are structured as a conventional fund but managed with a permanent-capital orientation. The partnership's own money is a meaningful share of every vehicle, and nobody here is compensated for recycling a portfolio quickly.
2023
The lab becomes a department
Research moved from one person's side project to a funded group with its own compute budget and a mandate that had nothing to do with marketing. Eleven model audits that year, all internal, all uncomfortable reading for somebody.
11 model audits
2024
Fund I fully deployed
Seventeen positions, the first meaningful set of follow-ons, and the serving stack that took our own inference cost from $4.10 to $1.40 per million tokens — the work that convinced us the lab was not a hobby.
17 positions
2025
Fund II opens
A second vehicle brought the firm to $180M under management, twenty-three companies and eleven people. Serving cost reached $0.62 per million tokens and the lab began publishing internal technical notes on a monthly cadence.
$180M AUM
2026
Where we are
Deploying Fund II at five to seven new positions a year, with a lab researcher on the technical diligence for every one of them. Same office, same standard, more compute.
5–7 per year
Why "Gothic"
Engineering and ornament, the same act.
The name is not a mood. Gothic building was the first European architecture in which the structure and the decoration were a single decision: the rib that carries the vault is the line you are looking at. Nothing is applied afterward to make the engineering palatable.
It was also built on a timescale that has since gone out of fashion. Chartres was rebuilt in twenty-six years after a fire. Cologne took six hundred and thirty-two, with a pause of nearly four centuries in the middle — and the masons who resumed work in 1842 built to the original drawings. Almost nobody involved expected to see the result.
Both halves of that are the point. We back companies where the technical decision and the product decision are indistinguishable — where you cannot separate what the model does from what the company is — and we underwrite them on a horizon long enough that the second and third model generations arrive while we are still shareholders.
The mark is a lancet arch: two arcs meeting at a point, each holding the other up. It is the shape that let builders go higher with less stone. We are aware that an investment firm comparing itself to a cathedral is a large claim. We would rather be measured against that than against a quarter.
The team
Eleven people, one floor.
Investors and researchers work in the same room rather than in adjacent buildings with a liaison between them. It is the only organizational decision we have made that we would defend without qualification.
11
People
All in New York. No satellite offices, no scouts, no venture partners with a day job elsewhere.
4
Research
Full-time researchers working on efficient inference, small domain models and evaluation tooling.
5
Investment
Two partners, two principals and an analyst. Three came from multi-stage funds; two came from building things.
2
Platform & operations
Hiring, LP reporting and the compute relationships that make subsidized GPU-hours possible.
Five of the eleven have shipped production machine-learning systems. One is a founder in residence — a former operator whose company was acquired — who works alongside portfolio teams through their first enterprise deployments and then gets out of the way.
We do not publish individual biographies. The work that matters here is done by the firm, and a founder should choose us for what we do rather than for whose name is on the page. If we meet, you will know exactly who is reading your code and why they are qualified to.
Principles
Three sentences we are willing to be judged on.
-
Conviction over consensus
A crowded round is not evidence. We have passed on companies that everyone we respect was funding, and led rounds nobody else would price. Both decisions were made the same way: by writing down what has to be true, and then finding out.
-
Technical truth over narrative
A demo is a claim, not a result. We would rather read your evaluation harness than your deck, and we would rather be the investor who found the data leak than the one who discovered it eighteen months later on a board call.
-
Decades over cycles
Capital that behaves well in a good year is common. We built the firm — its reserves, its pace, its compensation — so that it behaves the same way in a bad one. We underwrite decades, not demos.
Contact
Two addresses. Both are read.
Founders come first: a partner reads every introduction and replies either way, usually within a business day. Limited partners considering Fund II can request the current materials directly.
The firm is in New York. We travel for a first meeting when it matters and are candid when it does not.
Introduce your company
Two paragraphs is enough: what the model does, and what could not be built without it.
Fund II materials
Strategy, portfolio construction and terms, available on request to qualified institutions.