AI companies spend differently: compute, model APIs and cloud commitments can outrun revenue fast. We keep those costs visible, classify them the way investors expect, and show you gross margin, burn and runway every month.
Book your free consultation
Leave your details and we’ll get in touch to set it up. You get a fixed fee in writing before any work starts.
What we handle
- GPU, cloud and model-API costs split between cost of revenue and research and development
- Gross margin by product or model, so you know what each customer really costs to serve
- Cloud credits and prepaid compute commitments tracked separately from cash spend
- Usage-based and hybrid pricing recognized correctly under ASC 606
- Contractor and international team payments recorded and reconciled
- Research and development costs tagged for your tax CPA
- Monthly investor reporting: revenue, gross margin, burn and runway
- Books kept in QuickBooks Online, Xero, NetSuite, Sage Intacct or an AI-native platform such as Rillet or Digits
Where AI company books usually go wrong
Compute in one bucket
Training runs and customer-facing inference booked to the same account hide your true gross margin.
Credits treated like cash
Runway that counts on cloud credits looks longer than it is once the credits run out.
Usage billed, not earned
Usage-based revenue booked when it’s invoiced, not when it’s delivered, swings your numbers month to month.
How do you calculate gross margin for an AI company?
Gross margin is revenue minus the cost of serving customers, divided by revenue. For an AI product, cost of revenue usually includes inference compute and hosting, the model-API calls behind each customer request, third-party data or tools used in delivery, customer support and onboarding staff, and payment processing. Training runs, research experiments and internal tools sit below the line in research and development. Investors compare AI margins with software margins, so getting this split right every month matters more than almost any other number in your books.
- Cloud and API invoices split by project, environment or tag, not booked as one line
- Gross margin by product, model or customer tier where the data allows
- Compute cost per unit of usage tracked over time, so you see whether margins improve as you scale
How should you account for prepaid compute and cloud commitments?
A prepayment for compute is an asset that is expensed as you use it, not an expense on the day you pay. A committed-spend contract with a cloud provider is a promise to pay, so we track how much of the commitment you have used and what is left, and flag any shortfall you would owe if usage falls short. Both flow into your runway, which should show when credits end and commitments come due.
What changed for AI research and development costs?
The One Big Beautiful Bill Act restored immediate deduction of domestic research and experimental costs, including software and model development, for tax years beginning after December 31, 2024, while foreign research costs are still amortized over 15 years. Where your engineers and contractors are located now matters for tax, so we tag research costs by type and location in the books, and your tax CPA applies the rules and any research credit.
Our startup accounting page shows what we do at each funding stage, our SaaS accounting page covers subscription revenue and deferred revenue, and our AI accounting software comparison covers the ledgers many AI startups now use. Monthly support is quoted on your volume and stage; see pricing.
Frequently asked questions
Should compute costs be cost of revenue or R&D?
It depends on what the compute is for. Compute that serves paying customers (inference, hosting and the model-API calls behind your product) generally belongs in cost of revenue. Training runs and experiments generally belong in research and development. We set up your chart of accounts and tagging so the split happens every month, not at audit time.
How do you handle cloud credits?
We track credits separately from cash spend, so your reports show both what compute really costs and what you’ll pay once the credits run out. Your runway should be based on the second number.
Do you work with AI startups before revenue?
Yes. Pre-revenue companies need clean books, burn and runway reporting and R&D cost tracking from day one. It makes fundraising and due diligence much easier later.
How do you recognize usage-based revenue?
Under ASC 606, usage fees are generally recognized in the period the customer uses the service, even if you invoice later. Minimum commitments, prepaid credits and overages each need their own treatment, which we set up from your billing data.
Are customer prepaid credits revenue?
Not when you receive the cash. Credits a customer buys in advance are deferred revenue until they are used, and unused credits that expire are recognized under rules your auditor will want to see applied consistently.
What is a good gross margin for an AI company?
It varies widely with how much inference each customer uses and whether you run your own models or pay for APIs. The useful question is whether your margin is improving as you grow, which only shows up if compute is split correctly every month.
Do you work with AI companies that have international teams?
Yes. We record contractor and employer-of-record payments, reconcile multi-currency accounts, and tag costs by location, which now matters for how research costs are treated for tax.
Which accounting software suits an AI startup?
Many start on QuickBooks Online or Xero; others choose AI-native ledgers such as Rillet or Digits, or move to NetSuite as entities and audits arrive. We work in all of them.
How soon after month-end are the books closed?
Typically by the 7th to 10th business day of the following month, with gross margin, burn and runway ready for your investor update.
Do you file our taxes or research credit claims?
No. We keep the books tax-ready and research costs tagged, and work alongside your tax CPA, who prepares the returns and any research credit claim.
Know your real margins and runway
Tell us about your stack and your next milestone, and we’ll recommend the right level of support.