Angels get into AI startups the same way as any early-stage deal: through founders they know, syndicates, angel groups and by being useful, usually at pre-seed or seed. Before investing, check what the company owns that the next model release can't copy, whether its unit costs improve with scale, and whether the team ships quickly.
AI makes up most of the notable companies in my portfolio: seven of the twelve are AI companies, and Lenny's Newsletter, which ranked me the #1 most active angel investor in 2024, described my investing as primarily in AI and B2B SaaS. This guide covers how an angel gets access to AI deals and the checks worth running before writing the check. It explains; it doesn't recommend any particular investment.
How do angels get into AI startups?
Access depends on the stage.
Pre-seed and seed. This is where most angel money goes, and access works like any other early deal. Founders raise first from people who know them: former colleagues, operators in their field, earlier investors. Syndicates and angel groups bring deals to members, and demo days put many companies in one room. The angels who get the best AI deals tend to bring something specific, such as technical judgement, customers in a vertical, or experience scaling inference costs. The guide to building deal flow as a new angel covers the habits that make founders call you.
Late-stage AI companies. For individuals, access to the large labs and their peers usually comes through special purpose vehicles (SPVs) or secondary sales from employees and early holders. Be careful here. Fees can stack when one SPV invests through another, private companies often require their approval for any share transfer, and some offerings hold no direct shares at all. OpenAI said publicly in 2025 that tokens sold as "OpenAI" exposure were not OpenAI equity and that any transfer of its equity needs the company's approval. Ask any SPV manager how the vehicle holds its position and whether the company has approved it. The article on angel investing in frontier AI companies goes further into that end of the market.
Where AI startups sit, and why it matters to an angel
AI startups fall into rough layers, and each carries a different risk:
Model builders. Training frontier models takes enormous amounts of capital, so angels rarely get in early except through personal connections.
Infrastructure and tooling. Data, training and inference platforms, evaluation tools and developer tooling. Two of the AI companies in my portfolio sit here: Scale AI (data infrastructure for AI) and Hugging Face (an open-source ML platform). The guide to investing in AI infrastructure companies covers this layer, and the case study on how Scale AI built its data labeling business shows one path.
Applications. Products that use models to do a job for a customer: Pika for AI video generation, Artisan AI for AI-powered sales automation, Brainbase for enterprise AI deployment. This is where most new AI startups are, and where the competition is fiercest.
For reference, the angel investment portfolio page lists these companies, including Anthropic, with the valuations the site tracks. A company valuation is not a return, and it says nothing about when or at what price an investor came in.
What should you check before investing in an AI startup?
The usual early-stage questions still apply: team, market, traction and terms. AI adds a few of its own.
What survives the next model release? If the product is a thin interface on a general-purpose model, a better model may make it unnecessary. Look for what the company owns: proprietary data, a workflow customers have built their process around, distribution, or deep integration with a customer's systems.
Do the economics improve with scale? Inference costs money on every request. Ask for gross margin and cost per task now, and how both change as volume grows or models get cheaper. A company whose costs rise with every power user has a problem that growth makes worse.
How dependent is it on one provider? Relying on a single model vendor's pricing, rate limits or policies is a business risk. Ask how hard it would be to switch.
Do the technical claims hold up? Ask for a live demo on data the founders didn't choose, the evaluations they run, and their failure rate. The guide to evaluating AI startups' technical claims lists the questions worth asking.
Is the revenue real? Many enterprises run AI pilots from experimental budgets. Separate paid pilots from signed contracts, and ask how many pilots converted.
Are data rights and regulation covered? Ask where training and customer data comes from and what the contracts allow. Companies selling into the EU should know which of the EU AI Act's obligations apply to them.
Can the team ship faster than the field? The space moves quickly, so speed of iteration is itself a moat. Look at how much shipped in the last three months.
Mistakes angels make with AI deals
Paying for the narrative. Early AI rounds are often priced high. A high valuation is not a reason to pass, but it shrinks the upside, so size the check accordingly.
Overweighting one theme. Ten AI application companies in the same category are less diversified than they look; one platform shift can hit them all.
Ignoring the competition from incumbents. Large software companies are adding AI to products customers already pay for. Ask why a buyer would add a new vendor.
Skipping the documents on secondaries. A late-stage AI deal through an SPV deserves more reading than a seed SAFE, not less.
More guides on stages, terms and portfolio building are in the angel investing section.
Frequently Asked Questions
Q:Can individual angels invest in OpenAI or Anthropic?
Not directly through a public market, since both are private. Individuals typically get exposure through SPVs or secondary transactions, which carry layered fees and need the company's approval for transfers. Verify how any vehicle holds its shares before investing.
Q:Is it too late to angel invest in AI?
New AI companies are founded every month, and most of them are at the pre-seed and seed stages where angels invest. What has changed is price and competition: valuations are often higher and more investors are chasing the same rounds, so selection and terms matter more.
Q:What is an AI wrapper company, and is it always a bad investment?
A wrapper is a product built mostly on top of someone else's model with little of its own technology. It's not automatically bad: some build distribution, data or workflow depth quickly enough to become durable. The risk is a company that never gets beyond the wrapper.
Q:How technical do I need to be to invest in AI startups?
You don't need to train models, but you should understand the basics of how the product uses them, what it costs to run, and how quality is measured. Where you can't judge a claim, ask a technical co-investor or an expert you trust.
Q:How should I size checks in AI startups?
Treat AI like any other high-risk category: many small checks rather than a few large ones, with reserves for follow-ons in the companies that work. Higher entry valuations make diversification more important, not less.