Summary

  • DigitalOcean said its top 25 customers represented approximately 20% of Q2 2026 revenue, up from approximately 9% a year earlier. Applying those rounded shares to reported revenue produces an illustrative increase of about US$36.6 million—roughly 59% of the quarter's year-on-year revenue growth—but it is not a matched-customer disclosure.
  • “AI Customer ARR” reached US$234 million and includes all IaaS and PaaS/SaaS revenue from an account once that customer uses at least one AI or machine-learning product. It therefore measures the full cloud relationship of an AI-using customer, not AI-product revenue.
  • DigitalOcean disclosed more customers above several spending thresholds and said it was beginning to land nine-figure annual commitments. It did not disclose whether those cohorts overlap, which customers entered or left them, or how much revenue, capacity, margin and cash each cohort produced.

DigitalOcean's second-quarter filings contain a sharp change in customer shape. Revenue rose to US$281.2 million from US$218.7 million, an increase of US$62.5 million or 29%. At the same time, the Form 10-Q says the 25 largest customers accounted for approximately 20% of revenue, against approximately 9% in the prior-year quarter.

Those percentages are rounded, and the company does not identify a fixed cohort. Still, they show the scale of the change. Multiplying the reported quarterly revenue by each stated share gives about US$56.2 million for the current top-25 envelope and US$19.7 million for the prior-year envelope. The difference, US$36.6 million, is about 58.5% of total year-on-year growth. This is a BTW illustration, not an issuer-reported figure: the identities may have changed, the percentages are approximate, and no customer-level reconciliation is available.

A rank is not a cohort

“Top 25” sounds stable but is a ranking recalculated at each reporting date. A customer can enter because its spending grew, because another customer's spending fell, or because the ordering changed at the margin. Without identities, entry and exit records, or prior-period revenue for the same accounts, the two percentages cannot establish that one constant group more than doubled.

The same limitation applies to DigitalOcean's other customer measures. Customers spending more than US$100,000 annually rose to 632 from 582; those above US$500,000 rose to 92 from 68; and those above US$1 million rose to 45 from 26. The thresholds are nested. The 45 customers above US$1 million are already inside the US$500,000 and US$100,000 totals, so the groups must not be added.

The earnings presentation gives a corresponding nested ARR view: US$395 million from customers above US$100,000, US$291 million from those above US$500,000, and US$259 million from those above US$1 million. These values reveal an increasingly heavy upper tail, but they do not reveal the distribution inside each band or its overlap with the top 25.

AI Customer ARR includes ordinary cloud workloads

DigitalOcean reported AI Customer ARR of US$234 million, up from US$75 million and equal to 21% of total ARR, compared with 9% a year earlier. The label's definition is decisive. AI Customer ARR annualizes the entire revenue of customers using at least one AI or machine-learning product, including their infrastructure-as-a-service and platform/software-as-a-service revenue.

An account can therefore qualify because it uses inference or accelerator compute while also buying conventional virtual machines, networking, storage, databases or application services. All of that account revenue enters the metric. US$234 million should not be described as revenue earned from AI products.

ARR itself is another constructed measure: DigitalOcean generally takes revenue from the latest quarter and multiplies it by four. AI Customer ARR mechanically implies about US$58.5 million for the latest-quarter account envelope, but it is neither a contract balance nor a guarantee of the next four quarters.

The composition data are useful within that boundary. DigitalOcean said 85% of AI Customer ARR came from offerings other than bare metal. It reported growth of 158% in core cloud revenue within the measure, 762% for inference, and a 20% decline for bare metal. Those rates show that the labelled accounts are broad cloud relationships. They still do not allocate the US$234 million among AI-specific and non-AI workloads.

Large accounts bring both operating leverage and dependence

Management said its highest-spending customers and sophisticated AI-native companies were driving acceleration, and that DigitalOcean was beginning to land nine-figure annual commitments. These statements point toward longer and larger commercial relationships, but the public material gives no customer count, names, start dates, consumption schedules, utilisation, cancellation terms or margin for those commitments.

No disclosed bridge connects the nine-figure commitments, the AI Customer ARR population, the 45 customers above US$1 million and the current top 25. They may overlap substantially or only partly. Treating them as one group would turn four different disclosures into a fictional customer ledger.

The economic evidence is mixed enough to make that ledger important. Cost of revenue rose 44% to US$126.5 million, faster than revenue, while GAAP gross margin fell to 55% from 60%. DigitalOcean attributed the pressure largely to data-centre expansion costs incurred ahead of revenue. Six-month operating cash flow was nearly flat at US$156.9 million versus US$156.5 million.

Capacity installed ahead of demand can depress margin before utilisation rises. Large commitments can help fill it. But a commitment, billed revenue, cash collection and gross profit are separate states. Concentration can improve sales efficiency and capacity planning while also increasing the consequence of one large customer's delay, optimisation or departure.

The missing bridge is account-level movement

DigitalOcean also reported roughly 22,000 customers in its Digital Native Enterprise group, up from about 20,000, and said that group represented 67% of revenue versus approximately 59%. Customer counts are averages of month-end observations, and customers can move between categories during a quarter. They do not provide a clean opening-to-closing cohort roll-forward.

A useful disclosure would preserve customer identity privately while publishing a stable reconciliation: opening revenue or ARR, new accounts, expansion, contraction, exits and closing balance for each material cohort. For AI-labelled accounts it would separate AI-specific products from ordinary cloud services. For commitments it would distinguish signed value, activated capacity, recognised revenue, gross profit, cash collected and renewal status.

Until then, the strongest conclusion is narrower than the headline. DigitalOcean is gaining larger accounts, and customers using AI products are buying a broad set of cloud services. The filings do not yet show how much of the growth is durable AI-product demand, how concentrated it is in the same named customers, or whether the capacity built for them is converting into margin and cash.

Sources