Capital for the researchers
who build the next decade of AI.
Asim is a global venture firm built to onboard AI research teams from lab to company. Pre‑seed and seed checks, research‑grade infrastructure, and an operator network across frontier labs — wherever you are in the world.
The next great companies will be founded by researchers.
For two decades, the frontier of computing was set by infrastructure companies, then by application companies. The next decade is being set by research teams — the people writing the papers, training the models, and shipping the eval suites that redraw what is possible every six months.
Most of these teams sit inside a small number of labs in a small number of cities. The capital, infrastructure, and operating knowledge required to spin them out into companies is unevenly distributed and almost never localized to where the talent actually is.
Asim exists to fix that. We back AI research teams at the moment they decide to become a company — pre-product, pre-incorporation, often pre-team — and we keep our hands on the wheel through the first training run, the first eval, and the first paying customer.
Eight surfaces where we believe the next $100B companies are being trained right now.
Teams advancing RLHF, RLAIF, distillation, and preference optimization for the next generation of foundation models.
Test-time compute, planners, verifiers, program synthesis, and the systems that turn raw next-token prediction into reliable thought.
Long-horizon agents, orchestration runtimes, evaluation harnesses, and the reliability layer the field still does not have.
Inference, training, vector and graph stores, simulation environments, and the developer tooling that the next 10,000 AI teams will pay for.
Benchmarks, red-teaming, interpretability, and the assurance stack that decides whether frontier systems are deployable.
Simulators, neural physics, sim-to-real, and the data engines that turn sensors into general-purpose embodied intelligence.
Models for proteins, materials, climate, and chips — where research breakthroughs translate directly into industrial leverage.
Regional foundation models, open-weight ecosystems, and the alternative supply chain to closed frontier labs.
The Asim onboarding program.
A structured 90-day on-ramp from research team to company — built around the actual pain points researchers face when leaving a lab.
Pre-seed and seed checks, written fast.
$250k–$3M into the first or second institutional round. We commit in days, not months — and we are happy to lead, co-lead, or follow strong partners.
Research-grade infrastructure from day one.
Direct access to GPU partners, sovereign clouds, and reserved capacity. We help you size your training and inference budget before you raise the round to pay for it.
An evaluation harness for your model and your company.
Red-team partners, benchmark builders, and an in-house eval studio. If a regulator, customer, or board asks how good your model is, you have a credible answer.
Hand-introductions to the people you actually want.
ML researchers, applied scientists, founding engineers, and infra leads from the labs you respect. We have hired most of them once already.
First five logos, not a Slack channel.
Warm intros to design partners across enterprise, defense, finance, and bio. We build the deck, attend the call, and stay until the contract is signed.
Capital that scales with the model.
Reserved follow-on through Series B and a global syndicate of multi-stage partners that will price your next round on the merits of the work.
A globally distributed firm, by design.
Frontier AI research no longer lives in one zip code. Neither do we. Our investment partners and operating advisors are split across the cities where AI is actually being built — so we can move at the pace of the team, in the timezone of the team.
// Asim is registered as a venture fund and operates through entities in
// Delaware, London, and Singapore. We invest globally with no regional cap.
Four principles we will not negotiate away.
Researchers, not operators-in-residence.
We back the people doing the work, not the people who once advised the people doing the work.
Conviction over consensus.
We are happy to be the only believer in the room. Half of our best calls have looked premature for 18 months.
Eval-driven, always.
We do not invest in vibes. Every thesis we hold has a benchmark we can be wrong about.
No human-in-the-loop theater.
We do not hover, we do not post-mortem your standups, and we do not rewrite your roadmap. We are on call when you need us.
Building something at the frontier?
Send us your paper, your repo, your eval scores, or a paragraph about what you are about to leave your lab to go build. We respond within 48 hours. Always.