AI investors: who actually funds AI startups
Every fund says it invests in AI, which makes the label almost useless when you are building a list. The useful split is what kind of AI company they write cheques into. A fund backing a foundation model is underwriting compute cost and a research team. A fund backing a vertical AI application is underwriting ordinary software economics and wants to know which data you own. Pitching one as though it were the other is the fastest way to waste a first meeting.
Reviewed Full-year 2025
The state of AI investment
The direction matters more than the totals, and it is not the one most founders assume. North American funding rose 46% in 2025 while deal count fell about 16% to just under 10,500 rounds, so the money grew and the number of companies raising shrank. The OECD reports the same concentration inside AI: mega deals took about 73% of AI investment value and the early-stage share has been falling since 2023. A record year for the sector is not a record year for the median seed round.
The four kinds of AI investor
This is the table worth keeping. A fund tagged "AI" sits in one of four columns, and the columns want different companies, on different timelines, with different economics. Work out which one you are in before you build a list, because the same deck cannot serve two of them.
| Foundation models & compute | Applied & vertical AI | AI infrastructure & devtools | AI-enabled services | |
|---|---|---|---|---|
| Typical entry stage | Seed, but at growth-round size | Pre-seed to Series A | Seed to Series A | Pre-seed to seed |
| Cheque shape | Venture plus compute credits and strategic capital | Ordinary venture | Ordinary venture | Venture, occasionally revenue-based |
| Wants to see | A research team and a training recipe others cannot copy | Proprietary data, or a workflow competitors cannot reach | Developer adoption and a usage curve | Gross margin that improves as volume grows |
| Time to revenue | 2-4 years, if revenue is the point at all | 6-12 months | 12-18 months | Immediate, though margin takes longer |
| Biggest risk they underwrite | Compute cost, and being leapfrogged by the next release | That a general model absorbs the use case | Open source, or a cloud vendor bundling it for free | That it is a consultancy asking for a software multiple |
| Who else must be in the round | A cloud or silicon strategic, sooner rather than later | Nobody in particular | Angels with genuine developer reach | Nobody in particular |
AI investors, grouped by the cheque they write
AI investors we hold in the Causo catalogue, grouped by the cheque they actually write. Open any of them to see the partners, the stage and the recent deals. This is not every AI investor in the market, and no catalogue is.
AI specialists
Firms whose entire mandate is AI, rather than generalists who added the word in 2023.
Pre-seed and seed
Firms that write the first institutional cheque into an AI company.
Series A and later
Firms that lead once there is a product, a usage curve and revenue to underwrite.
Corporate and strategic
Balance-sheet investors whose parent is also a plausible compute provider, distribution channel or customer.
What AI investors need to see
The gates that are specific to this sector, and that a generalist fundraising guide will not tell you about.
- Say which column you are in, in the first two linesThe most common mistake on this page. A vertical AI application pitching a foundation-model investor is asking for a cheque that firm does not write, and the reverse wastes even more time. Name the category before you name the product.
- Show what data you ownThe first question after the demo is almost always where the data comes from and whether anyone else can get it. A model trained on public data with a thin interface is the weakest position in the sector. An exclusive feed, a proprietary workflow or a customer who lets you learn from their operations is the strongest.
- Bring your evals, not just your demoDemos are cheap and every founder has one. Investors who have been in the sector since before 2023 will ask how you measure quality, what you measure it against, and how it has moved. Having a real eval suite is now a proxy for engineering seriousness.
- Know your inference cost per queryGross margin is the question that separates a software business from a reseller of somebody else’s compute. Know the cost of serving one customer today, and how it changes as usage grows.
- Answer the next-model question before it is askedEvery AI investor is now underwriting the risk that the next general model release absorbs your product. You will be asked what survives that. The credible answers are data, distribution, workflow depth and regulatory position, not model quality alone.
- Check they were investing in AI before 2023A very large number of funds added AI language to their site during the last two years. What tells you is whether they led an AI round before the current cycle, and whether their partners can talk about architecture rather than adoption.
Where an AI investor’s real track record is published
- Fund portfolio pages and announcement posts
- arXiv papers and their author affiliations
- Model and dataset releases on Hugging Face
- GitHub repositories and contribution history
- Patent filings
- Funding announcements
- Partner essays, podcasts and conference talks
- Cloud and silicon partner directories
- Why open work matters more here than in other sectorsAI is unusually legible in public. Papers, model cards, benchmark submissions and repositories show who was working on a problem years before a portfolio page mentions it, and which investors were already in the room when they were.
Reading one AI investor’s actual record
One worked example of what reading those sources produces, from the Causo catalogue with the identity removed.
A personal angel vehicle that was planned as an institutional fund and then deliberately narrowed into a book of direct early-stage cheques. The concentration is artificial intelligence, deep tech and health tech, with adjacent activity in education technology, immersive learning and industrial AI. The stated mission is bridging regional technical talent to global capital.
Questions founders ask about AI investors
Who are the main AI investors?
AI-dedicated firms include Radical Ventures, Zetta Venture Partners, AIX Ventures, AI Fund, Basis Set Ventures and Air Street Capital. Corporate and strategic investors include Nvidia NVentures, Gradient Ventures, Intel Capital, Salesforce Ventures and the OpenAI Startup Fund. Most large generalist firms also now invest in AI, so which of them is relevant depends far more on whether you are building a model, an application or infrastructure than on fund size.
How much venture funding does AI get?
AI firms captured 61% of global venture capital in 2025, more than double their 30% share in 2022, according to OECD analysis. In North America alone about $168bn went to AI-related companies, roughly 60% of all startup funding there. It is the largest concentration of venture capital into a single category in the industry’s history.
Is it harder to raise for AI now that everyone is doing it?
The totals and the median moved in opposite directions in 2025. Funding rose sharply while deal count fell about 16%, and roughly 73% of AI investment value went into mega deals. More money entered the sector and fewer companies raised, so the bar for an ordinary seed round went up rather than down.
Do AI investors still fund application-layer startups?
Yes, and for most founders that is the realistic column. Applied and vertical AI raises ordinary venture rounds on ordinary software economics, and investors underwrite whether you own data or a workflow that a general model cannot reach. Foundation model rounds are a small number of very large cheques and are not a template for anybody else.
What do AI investors look for at seed?
Evidence that the product works on real data rather than a demo, a clear answer on where that data comes from and who else can get it, an eval methodology showing quality is measured and improving, and an inference cost per query that supports a software gross margin. Team credibility in machine learning still matters, but it no longer substitutes for any of those.
How do I tell whether a fund is genuinely an AI investor?
Look at what they led before 2023 rather than what their site says now. A very large number of funds added AI language during the current cycle. Check whether they have led an AI round, whether their partners write about architecture rather than adoption, and whether recent announcements are in your sub-sector rather than the sector generally.