Software & digital3 min read
Best structure for an AI startup: LLC, C-corp and investor readiness
The entity question for an AI company is the SaaS question with the dial turned up: capital-intensive, investor-heavy, and with a cost of goods sold that behaves like a utility bill rather than like software.
The short answer
Delaware C-corp if you are raising, which most AI companies are. LLC if you are genuinely bootstrapping. The difference from ordinary SaaS is that the capital requirement usually answers the question for you.
Published
Most of what matters about structuring an AI company is in the SaaS guide — the LLC-versus-C-corp trade-off, Section 174A, sales tax on software. This page covers the three things that are actually different.
Difference one: you probably are raising
Inference costs money per request, and models cost money to train or fine-tune. That is a materially more capital-hungry shape than a traditional SaaS product, where the marginal cost of a user rounds to nothing.
The practical consequence is that the C-corp branch of the decision applies to more AI companies than software companies generally, and earlier. If you are going to raise a priced round from funds, form the Delaware C-corp now rather than converting mid-term-sheet.
Difference two: cost of goods sold is real
Model API spend scales with usage, which changes several things at once:
- Gross margin is a number you have to manage, not a rounding error. A flat-rate subscription over a variable inference cost is a business that can lose money on its best customers.
- Provider price changes move your margin overnight, in either direction, and are outside your control.
- If you resell provider capacity to clients as a pass-through, the accounting question is the same one agencies face with media spend — see the agency guide.
- Section 174A treats domestic research and experimental expenditure as currently deductible for tax years beginning after 2024, while foreign R&E is amortised over 15 years. Where your engineering sits matters; where your inference runs is an ordinary operating cost.
Difference three: IP and data are the diligence questions
An acquirer or an investor will ask questions here that they would not ask of a CRM company, and the answers need to exist before they are asked:
| Question | What good looks like |
|---|---|
| What is your training data and where did it come from? | A documented provenance for anything you trained or fine-tuned on, and licences where required. |
| Who owns the outputs? | Terms that say so explicitly, consistent with your upstream provider's terms. |
| What do you do with customer data? | A clear position on whether customer inputs are used for training, and a contractual commitment matching it. |
| What happens if your model provider changes terms or pricing? | An abstraction layer, or at least an honest answer about the dependency. |
Regulation is arriving unevenly
There is no single US AI statute, but there are sector rules that already apply — employment, credit, insurance and healthcare decisions made or assisted by automated systems attract existing law. Several states have enacted their own requirements, and the EU's regime reaches products offered to EU users regardless of where the company sits. If your product makes consequential decisions about people, that is a compliance question now, not later.
If you are not a US person
- No S-corp election — Section 1361 bars non-resident alien shareholders. A C-corp is available.
- Form 5472 with a pro-forma Form 1120 annually for a foreign-owned single-member LLC, $25,000 penalty for failure to file.
- US funds generally require a Delaware C-corp. If raising from them is the plan, the flip is coming; do it early.
- Where your engineering happens matters for Section 174A, and where your data subjects are matters for privacy law.
When to revisit
| Trigger | What to reconsider |
|---|---|
| A priced round becoming plausible | Delaware C-corp now, not at signing. |
| Fine-tuning on customer or scraped data | Provenance, licences and your own terms. |
| Selling into regulated sectors | Sector rules that already apply to automated decisions. |
| Serving EU users | The EU regime reaches you regardless of where you are. |
| Gross margin falling as usage grows | Pricing, before growth makes it worse. |
Get the corporate housekeeping right early
Founders 8 holds the entity, the filings and the deadlines — the things diligence checks and founders forget.
Build your workspaceFounders 8 does not provide tax advice. Tax residency depends on facts and rules specific to each jurisdiction — review your position with a qualified adviser.