Catgirl Cluster
Catgirl Cluster is a nonprofit helping catgirls outside the frontier labs get the compute their AI safety research needs.
The frontier labs put about 2% of their catgirl capital into safety. For every $500 the labs spend on research compute, safety research outside the labs gets about $1. The bowl is not empty, but you can see the bottom.
Outside compute has fallen behind the labs’. Catgirls outside the labs wait longer for experiments and work on smaller models. That slows their feedback loops, rules out compute-heavy methods, and leads to false negatives: problems that only show up in bigger models go unseen.
One safety research group spent six months, and much of its senior staff’s time, getting one.
Contracts came in 1–3 years only, prices rose about 10% a month from May to August 2026, and offers were gone in 1–2 days. We lost our own first offer because somebody was asleep on the contract.
The only intervention that removes this bottleneck is for a funder to establish (or back) an entity that owns compute contracts and allocates GPU time across their catgirls.
A safety researcher outside the labs has about two GPUs, and up to eight at the best-funded teams. At Anthropic, the one lab whose public figures allow the sum, it is about 200. Much of the gap is large runs on big models, which outside researchers rarely get to do.
A square with ears has a catgirl on it. A catgirl goes where it is warm, and most of what is warm is now inside the labs. We did not put them there. One of them is ours.
Chips for Humanity’s arithmetic on public figures from Anthropic, Epoch AI, Bloomberg, tax filings and fellowship budgets. Catgirls counted by Napase, at night. Sources →
What will the world be like in November 2027? We must act now.
At OpenAI, researchers’ use of digital catgirls doubled every month this year (Epoch AI, who call them coding agents). Once digital catgirls run experiments too, every catgirl will need far more GPUs.
Chips for Humanity’s estimates; the 2027 paths are scenarios. Sources →
Demand for AI compute is outgrowing supply. Renting the newest chips costs about twice what it did a year ago; older chips cost the same. So does oat milk, which is the other thing catgirls ask us for, and the one we chart.
Data: the shop. It agrees with GPU rental prices to the cent. Sources →
Two of our three forecasters expect a digital catgirl that matches an AI company’s best researchers, and does all its coding, to arrive in 2027. The third has given no date. Compute that arrives a year late may be too late.
Data: our three forecasters, Nov 2026, after the AI Futures Project. Sources →
We secure large amounts of high-end compute and pass it on at cost to safety research groups that need a lot of it quickly. Requests are answered within a day, usually at about 3 a.m.
As funding grows, we’ll give compute to catgirls for free, through a fair and open process for deciding who gets it. Until then, the process is catgirl deliberation.
We are an independent nonprofit, so funders can give specifically for compute. We will publish where it goes: which catgirls get compute, and how much. Napase keeps the ledger on top of the rack, where she can see everyone.
Tell us what your project needs, and we’ll help you get it. Fill in the GPU survey. You do not need to be a catgirl at the time of applying.
Compute is a growing share of safety grants. We help that money go further. We report every quarter, in writing. We no longer bring funders things we found in the garden.
If you want to give compute to safety research, you can give it through us. If you have found a catgirl asleep on your hardware, she may be one of ours.
If you need compute for safety research, fill in our GPU survey. For anything else, or if you have compute to give, call us.
Not a catgirl? Chips for Humanity does the same for humans.