The four pieces
For most early AI companies, the stack looks like this:
| Piece | What it typically pays for | What it costs you | Where to read more |
|---|---|---|---|
| Compute and cloud credits | GPU, cloud and API bills, for a limited time, on named services | Usually no equity. Credits expire and may not cover every service. | Our Guides hub, then the compute-credits guide and list |
| Grants | A defined research or innovation project, on the funder's terms | No equity, but a fixed scope, reporting, and often a delayed start. Some UK grants fund only a percentage of costs. | Our grants guide, UK grants for AI, and NSF SBIR and STTR |
| R&D tax relief | A credit or payable amount on qualifying research you have already done, or are doing | No equity, but a compliance burden, and not all software work qualifies. | R&D tax credits for AI startups in the UK and the US |
| Equity, including SEIS, EIS and accelerators | Runway, salaries and work a grant will not fund, in exchange for a share of the company | Dilution, and often a change of company structure or location. | The Guides hub for accelerators, and our grants guide if you are still looking at non-dilutive money first |
You can often combine them. A company may use cloud credits, claim R&D relief on its own qualifying spend, submit a grant, and raise a seed round in the same year. The interactions matter: grant money can affect what expenditure counts for tax relief, and some credit programmes have rules about how you have already been funded. Tell whoever prepares your tax claim and your grant budget about the other money coming in.
A sensible order, not a rigid sequence
There is no official order you must follow. A practical sequence, for a company that is still early, is:
- Credits first, when you qualify. They are usually the fastest way to cut a GPU or cloud bill, and they do not dilute you. They also expire, so they are a poor way to fund salaries or a two-year research plan.
- Then look at grants that actually fit the current work. UK companies often start with Innovate UK, UKRI or ARIA, depending on who would hold the award. U.S. small businesses doing high-risk R&D should look at NSF SBIR and STTR, remembering that a Project Pitch and an invitation have to come before a full proposal, and that the next published NSF deadline after 18 November 2026 is 4 March 2027 [source]. Grants are slow. They fund a defined project, not "whatever the company needs".
- Claim tax relief on qualifying R&D, in parallel. In the UK, the merged RDEC credit is 20% (and is taxable) for accounting periods beginning on or after 1 April 2024; loss-making R&D-intensive SMEs may instead be able to use ERIS [source]. In the U.S., the section 41 research credit is claimed on Form 6765, and some qualified small businesses may elect to apply up to $500,000 of it against payroll taxes [irs-6765]. Relief is calculated on qualifying expenditure; it will not usually fund the next month's rent on its own.
- Use equity when you need runway that the first three cannot provide, or when a specific programme is worth the dilution. UK companies raising from individual investors often look at SEIS and then EIS, which we cover in a separate article under accelerators on the Guides hub. Accelerator offers, including Y Combinator and Entrepreneur First, are also listed there. Equity is the flexible part of the stack, and the expensive part.
That order is not a rule. If a strong accelerator offer is on the table, waiting a year for a grant window can be the more expensive choice. If you are a researcher at an eligible UK organisation, a UKRI grant may be the main route, and "startup equity" may not apply at all.
Match the instrument to the cost
A simple test is to name the cost first, then the instrument:
- GPU and API bills this quarter: credits, then paid GPU cloud if you still have a gap.
- A defined technical project with a public-funder scope: a grant, written to that scope. See how to write a UK AI grant application or how to write an NSF SBIR proposal.
- Qualifying research you are already doing: R&D tax relief, with an adviser, not a guess.
- Salaries, runway, and work that no open grant will fund: equity, including SEIS or EIS if you qualify, or an accelerator.
Do not stretch a grant application to cover a cost the call will not fund. Do not treat a tax credit as if it were a loan approved in advance. Do not assume an accelerator is "free money" because the programme also includes credits; read the equity terms.
Where to go next
If you already know which piece you need, go straight to the specialist article: UK grants, NSF SBIR and STTR, R&D tax credits, or the Guides hub for compute credits and accelerators. Open grant calls are on the grants list, soonest deadline first.
If you are writing a bid and wondering whether to hire help, see grant-writing help: when to hire a bid writer.
