Q3 2026 earnings call, November 3, 2026: revenue growth reaches 40%, the highest in a decade, and management pitches a differentiated AI cloud model

DigitalOcean reported third quarter revenue of $321 million, up 40% year-over-year and more than $14 million above the high end of its guidance range. This is the fifth consecutive quarter of acceleration, from 16% growth a year ago. The more consequential news was the 2027 outlook. Management raised its growth target to 60% or more, from "50% plus" previously and from 30% when the company first framed 2027 in February. CFO Matt Steinfort said the increase adds "over $0.5 billion at the midpoint" to projected 2027 revenue, and it is applied to a 2026 base that has itself been raised by $120 million since February. Detailed 2027 margin and cash flow guidance will wait for the fourth quarter call.

Yield per megawatt becomes the central bull argument

The call's core thesis was that DigitalOcean's growth is driven more by yield per megawatt than by the number of megawatts activated. Management put the fleet-wide figure at $18 million of ARR per average active megawatt, which CEO Paddy Srinivasan described as "not a vanity outlier metric" and as the return on the entire fleet, current-generation hardware and still-ramping new data centers included. Steinfort noted the figure was depressed slightly in the quarter because Memphis was counted for the portion of the period it was live while contributing little revenue. Management said it has closed several individual deals "well north of $20 million" of ARR per megawatt.

The company pointed to four levers: repricing through a consumption and short-term contract model, newer and more token-efficient GPU generations, higher utilization of its elastic compute fleet, and attach of higher-margin compute, storage and data products. The comparison investors will make is with bare-metal neoclouds that sign multi-year contracts with a few large customers. Management was explicit that it does not want that model, and Steinfort said demand "is well in excess of our supply," which lets the company be "super picky" about customers on funding, economics and viability.

There is a trade-off. Short-term and consumption-based pricing captures upside when capacity is scarce, but it carries the opposite exposure when the market loosens. Goldman Sachs analyst Gabriela Borges pressed on what happens when industry supply comes online, model efficiency improves and spot prices normalize. Srinivasan's answer leaned on platform breadth rather than pricing resilience: the use-case pie is growing faster than supply-demand equilibrium effects, and the company has "very few GPU-only customers." That is a reasonable argument but not a quantified one. Management offered no sensitivity to a decline in spot pricing, and the Q3 beat was partly attributed to "higher prices on short-term capacity."

Quality of the beat: software and platform, not just new capacity

Management stressed that the upside came mostly from software and platform rather than new data centers. Drivers included inference services, core cloud attached to AI workloads, elastic fleet utilization and early agentic adoption. Only a small part came from Memphis, which launched a few weeks early. Record net new ARR of $160 million, more than 3.5x the prior-year figure, took total ARR to about $1.28 billion. AI customer ARR reached $376 million, up 288%, and more than 90% of it comes from inference services and core cloud, with inference services up nearly 600%.

Token volume on the Inference Engine rose roughly 10x quarter-over-quarter. The company has launched 37 models since April, and more than 4,000 customers use its inference router. On OpenRouter, which management described as a spillover channel for idle cycles, it averages well over 0.5 trillion tokens per day. Customers on four or more products generated roughly 60% of revenue, and 90% of AI customer revenue comes from customers that also use core cloud primitives.

The agent platform and a disclosure on agent-driven demand

DigitalOcean spent considerable time on its agentic stack, which it shipped in the quarter: Managed Agents (CPU-based secure sandboxes and a harness runtime), Agent Learning Services for context and memory, and Agent Data Services covering high-performance storage and databases. The company said it released more than 70 updates in 65 days and claims to be the only provider offering recursive self-improvement for agents and a zero-ETL, near-real-time data substrate for them. These are management claims we cannot verify, and the commercial scale of the agent products was not disclosed beyond customer anecdotes.

One unusual datapoint: roughly half of unique visitors to digitalocean.com are now agents, and more than 37,000 unique customers have deployed on DigitalOcean through an agent this year. Management framed agents as buyers as well as workloads, saying that agents "increasingly select it, sign into it and deploy on it." Amplitude was cited as a reference customer, running an open model on DigitalOcean that costs about 67% less than a frontier model for one core workload.

On differentiation in routing, Srinivasan said the company is "3 to 4 months ahead" of competitors, adding that "3 to 4 months in AI is like 3 to 4 years in the rest of software." That is a narrow lead by his own measure, and he returned repeatedly to platform breadth, noting that competing routers are proliferating.

Core cloud is accelerating, though AI classification is blurring

Canaccord's Kingsley Crane noted that ARR excluding AI appears to be growing only around 10% to 11%. Steinfort responded that it is "getting harder to tell who an AI customer is versus a non-AI customer," because new self-serve customers deploying agent frameworks such as OpenClaw or Hermes do not identify themselves as AI users. He said the roughly $900 million ARR core cloud business is accelerating and is "wildly profitable." The company also introduced Agent Droplets, a bundle meant to give traditional customers a simple path into agentic primitives. The practical implication is that the AI and core split is becoming less useful for modeling, and the company did not offer a replacement breakdown.

Capacity: 180 megawatts committed, with 2027 volume largely back-half weighted

All three planned 2026 data centers, Richmond, Kansas City and Memphis, are live, each on or ahead of schedule. The company has just over 75 megawatts active, about 75 megawatts signed for 2027 and another 30 for 2028, for roughly 180 megawatts committed. That includes an incremental 20 megawatts secured with Tier 1 partners in the quarter. Steinfort said 2027 capacity will arrive through the year, so "the second half will grow faster than the first half." This is a consideration for anyone assessing the 60% growth target, since delivery dates sit with colocation partners. Management said it has built conservatism into the guidance but acknowledged that timing is "a partnership with the data center providers."

On capital intensity, Steinfort offered an unguided benchmark of mid-$30 millions of capex per megawatt for a GB300-class build, and closer to $40 million for Vera Rubin. He said 2027 is likely to land in the mid-30s. Management said it underwrites newer generations to the same return thresholds, since higher token output per megawatt should offset higher capex, and it also said the returns could be better where software contributes.

Profitability strong, but gross margin is the pressure point

Adjusted EBITDA was $144 million, a 45% margin against guidance of 38% to 39%. Adjusted operating income was $74 million, a 23% margin. Non-GAAP EPS was $0.45 against guidance of $0.28 to $0.30. GAAP gross margin was about 50% and is declining as new data centers and GPU capacity add depreciation and lease costs ahead of revenue. Steinfort acknowledged that gross margin "will be pressured" but said operating leverage has more than offset it. The GAAP net loss of $43 million, or $0.37 per share, was driven by a $57 million induced conversion expense tied to a partial repurchase of 2030 convertible notes.

Guidance implies some normalization. Fourth quarter adjusted EBITDA margin is guided to 42% to 43%, below the Q3 level, and non-GAAP EPS of $0.30 to $0.32 is well below Q3's $0.45. Trailing 12-month adjusted free cash flow was $220 million, a 20% margin, down from 21% a year earlier despite the growth.

Outlook and funding

For the fourth quarter, the company guided revenue to $348 million to $351 million, or 44% to 45% growth. Full-year 2026 revenue is guided to $1.208 billion to $1.211 billion, about 34% growth, with adjusted EBITDA margin of 41% to 42%, adjusted free cash flow margin of 20% to 22% and non-GAAP EPS of $1.60 to $1.62.

Funding was a prominent concern in the Q&A, with Piper Sandler's James Fish asking why equity was being used alongside lease financing. DigitalOcean announced a $1 billion at-the-market equity program, which Steinfort described as "option value" rather than a commitment. Liquidity is about $1.2 billion, comprising $845 million in cash and securities and about $400 million of revolver capacity, and net leverage is 1.2x. The company also closed a $785 million equipment lease facility in September and plans to retire the remaining $312 million of 2026 convertible notes by maturity in December. Steinfort said management believes 2027 is "covered" and that equipment financing matched to revenue remains the primary strategy. Even so, an ATM filed alongside a sharply higher growth target signals that dilution is a live option, and the stock's reaction will likely depend on how investors weigh that against the balance sheet flexibility management emphasized.

Positioning against the neocloud cohort

Management repeatedly framed the company as "different by design" versus bare-metal AI infrastructure providers. The top 25 customers represent only 27% of revenue, and no single customer is at a high-single-digit percentage. Revenue from customers above $100,000 of ARR grew 133% to 41% of total ARR, and revenue from $1 million-plus customers grew 273% to 30% of ARR. The latter shows that growth is increasingly coming from larger accounts, even if concentration remains low. Asked about market nervousness, Srinivasan replied: "We are not nervous. We are very confident because we are different." The numbers support that DigitalOcean's profile is less exposed than most AI infrastructure peers to single-customer and heavy-leverage risks. The open questions are how durable the pricing environment is and whether the company can deliver roughly 75 megawatts of new capacity on schedule while funding it.