Digital ocean research:
DigitalOcean: The Cloud We Actually Use, Now Renting Out AI Too
- 1. Why This Showed Up On Our Screen
- 2. The Megatrend: Somebody Has to Run All This AI
- 3. The Third Way
- 4. What Actually Happened Last Quarter
- 5. Why Cheaper Models Are Good News, Not Bad News
- 6. How the Business Is Actually Built
- 7. The Flywheel: How a Customer Turns Into a Bigger Customer
- 8. The Open-Weight Model Wave
- 9. The Biggest Customers Are Growing the Fastest
- 10. Why the Margins Actually Hold Up
- 11. The Paperspace Deal, Revisited
- 12. Building the Physical Plant
- 13. Wall Street Is Catching On
- 14. Where Guidance Is Headed
- 15. The Price Tag: Then, Now, and Relative to Itself
- 16. What Could Go Wrong
- 17. Muffett's Take
1 Why This Showed Up On Our Screen
Most of the names in this research library get found the normal way: a scanner flags something, we pull the filings, we build a model. DigitalOcean is different, and we want to be upfront about that. We already use DigitalOcean. Muffett runs infrastructure on their platform, the same Droplets and cloud services regular developers and small businesses use every day. So when DigitalOcean started showing up on our screens as one of the fastest-growing, most profitable names in cloud computing, we paid extra attention, because we already know the product works. That's not a scientific reason to buy a stock on its own. But it's a real reason to go look closer, and when we looked closer, we liked what we found.
Here is the simplest way to describe what DigitalOcean does. For over a decade, it has been the cloud provider that individual developers, startups, and small businesses actually enjoy using, instead of the massive, complicated cloud platforms built for giant corporations. Simple pricing. No confusing bill. No army of consultants needed to set it up. That's the whole reason it built a loyal following in the first place. What's changed in the last two years is that this same simple, developer-friendly approach has been pointed at artificial intelligence, and it's working better than almost anyone expected.
2 The Megatrend: Somebody Has to Run All This AI
We keep coming back to one idea across a lot of our technology research: the world is racing to build AI applications, and every single one of those applications has to run somewhere. That's the megatrend. It doesn't matter how good a language model is if nobody can afford to run it, or if it takes a team of specialized engineers to plug it in. Somebody has to own the computers, rent out the compute, and make it simple enough that a small team can actually ship a product. That's an infrastructure business, and infrastructure businesses are exactly the kind of thing we like to own — a real, physical, hard-to-copy asset sitting underneath a demand curve that just keeps climbing.
The obvious way to invest in "AI infrastructure" is to buy the giant cloud companies — Amazon, Microsoft, Google — or the specialized GPU rental companies that have popped up over the last few years. DigitalOcean is a third option, and it's the one we think the market has been slower to appreciate. It's not trying to be everything to everyone. It's building the AI cloud for the same customers who already loved it: startups, independent developers, and growing digital businesses who don't want to hire a cloud infrastructure team just to use AI.
3 The Third Way
DigitalOcean's own materials describe this as "The Third Way," and honestly, it's a fair description. On one side you have the hyperscalers — Amazon Web Services, Microsoft Azure, Google Cloud — built for huge enterprises and the labs building the biggest AI models, with thousands of products that take real expertise to configure. On the other side you have the "neoclouds" — companies that mostly just rent out raw GPU chips, leaving customers to do all the technical work of stitching everything together themselves. DigitalOcean sits in between: one simple, fully connected platform, with an AI layer that's already set up and optimized, sold to customers who just want it to work.
The business reason this matters is margin. Renting out raw computer chips is a commodity business — everyone's chip does the same thing, so the only way to compete is on price, and prices keep getting cut. DigitalOcean instead bundles the compute together with software: managed databases, storage, developer tools, and now AI-specific tools like model hosting and inference. Customers pay for the whole package, not just the hardware, and that's a much better business to be in.
4 What Actually Happened Last Quarter
Numbers first, because that's really the whole story. DigitalOcean reported second-quarter 2026 results on August 4. Revenue came in at $281 million, up 29% from a year earlier — more than double the 14% growth rate the company was posting just one year ago. That's an acceleration, not a slowdown, which is unusual for a company of this size and is exactly the kind of thing that gets our attention.
Total annual recurring revenue reached $1.13 billion, and the company added $93 million of new recurring revenue in the quarter alone — a record, and nearly three times what it added in the same quarter a year ago. The AI piece of the business is the real headline: AI customer annual recurring revenue hit $234 million, up 212% year over year, powered by an 800% jump in inference services, which is the part of the business that actually runs AI models for customers. And it did all of this while staying solidly profitable — a 40% adjusted EBITDA margin and $175 million of trailing twelve-month adjusted free cash flow, a 17% margin. Growing fast and making real money at the same time is rare. Most AI-related companies right now are doing one or the other.
5 Why Cheaper Models Are Good News, Not Bad News
There's a shift happening in how companies use AI that's worth explaining simply, because it explains a lot of what's driving DigitalOcean's growth. Early on, most companies just threw the biggest, most expensive AI model at every problem, because it was new and nobody had figured out the cost math yet. DigitalOcean calls this "tokenmaxxing." The problem is that as AI shifts from a person typing a question to AI systems running on their own, in loops, handling tasks automatically, the amount of AI usage explodes, and using the most expensive model for everything becomes unaffordable fast.
What's replacing it is "valuemaxxing" — using the right-sized model for each specific job, and only paying for the expensive one when the task actually needs it. This is exactly what smaller, openly available AI models are good for, and it's exactly what DigitalOcean has built its inference business around: giving customers an easy way to pick the right model for the right price, instead of just defaulting to the most expensive option. As this shift happens across the industry, more of the value moves away from the companies that build the giant AI models and toward the companies, like DigitalOcean, that actually serve those models to customers reliably and cheaply. That's a structural tailwind, not a one-quarter trend.
6 How the Business Is Actually Built
It helps to see the whole platform laid out, because it shows this isn't just one product bolted onto the old business — it's a full stack, built in layers, with each layer feeding the next.
At the bottom is owned hardware: DigitalOcean actually owns the computers, including NVIDIA's high-end H100 chips, spread across 20 data centers worldwide, rather than just renting time on someone else's machines. On top of that sits the original core cloud business — the Droplets and virtual machines that made DigitalOcean's name. Above that is the inference engine, which is the fastest-growing layer and the one doing the heavy lifting of actually running AI models efficiently. Then comes a data layer — managed databases and storage that give AI applications a memory. And at the very top sit managed AI agents — secure, ready-made environments where AI systems can actually go do tasks on their own. The company has shipped more than 80 product releases across these five layers since April 2026 alone. That's a fast pace of building, and it tells us management is treating this as the main event, not a side project.
7 The Flywheel: How a Customer Turns Into a Bigger Customer
The part we find most convincing isn't any single product. It's how customers move through the platform once they arrive, because it turns a cheap first purchase into a much bigger one over time.
A customer typically shows up first for cheap, fast access to open AI models — DigitalOcean says a company called OpenRouter alone is running more than 20 billion AI "tokens" a day through DigitalOcean's systems, which gives a sense of scale. From there, usage tends to graduate into AI agents doing more autonomous work, which need memory and secure places to operate — another company, OpenCode, uses DigitalOcean to serve 7.5 million developers this way. That agent activity generates data, which needs to be stored and managed, pulling in DigitalOcean's database and storage products — the company Vercel is cited as an example, plugging DigitalOcean's inference engine directly into their own AI product. And all of that ongoing activity drives demand for the core computing power underneath everything. The result the company points to: 70% of AI customers spending more than $100,000 a year have also attached at least one of DigitalOcean's core cloud products. In plain terms — customers don't just buy one thing and leave. They keep buying more, and each purchase makes them stickier.
8 The Open-Weight Model Wave
One of the more remarkable numbers in this whole story is how fast the market has shifted toward using openly available AI models instead of the closed, proprietary ones from the big AI labs.
In April 2026, open-weight models made up just 15% of the total AI usage running through DigitalOcean's systems. Sixty days later, that number was 75%. At the same time, total usage — the actual volume of AI activity — increased 30 times over. This is a real-world example of something economists call the Jevons paradox: when something gets cheaper and more efficient, people don't use less of it, they use dramatically more. Cheaper AI models didn't shrink the market. They set off an explosion in AI usage. DigitalOcean got a head start here too — when a new model called Kimi K3 launched, DigitalOcean was the only cloud offering full support for it from day one, and it picked up 400 new customers in the first week alone. The company's catalog now includes more than 75 different AI models, each one automatically tuned for cost and performance. Being first and being useful, in a fast-moving market like this, tends to compound.
9 The Biggest Customers Are Growing the Fastest
A common worry with a company built around small developers and startups is that it will always be stuck serving small customers who don't spend very much. DigitalOcean's numbers say the opposite is happening.
Customers spending more than $1 million a year with DigitalOcean grew their combined spending 214% year over year — up sharply from 92% growth the year before — and that group now makes up 23% of total company revenue. The $500,000-plus group grew 160%, up from 64% the year before, and now represents 26% of revenue. Even the smaller $100,000-plus group nearly doubled, growing 98%, up from 37% a year earlier. So the whole customer base is growing faster, and the biggest spenders are growing fastest of all. What we like just as much: even with these much larger customers signing on, the top 25 customers overall still only make up 20% of total recurring revenue. That means DigitalOcean isn't becoming dependent on a handful of giant accounts. It's landing bigger customers while staying broadly diversified — which is exactly the combination you want to see.
10 Why the Margins Actually Hold Up
Renting out AI computing power sounds like a business that should have thin, fragile margins, given how competitive AI infrastructure pricing has become. DigitalOcean's numbers show it's avoided that trap so far, and it's worth understanding why.
Only 15% of AI customer revenue today comes from plain, bare-metal computing power — just renting out a raw chip. The other 85% comes from software and managed services layered on top, up sharply from 43% just a year ago. That shift matters because bare-metal pricing is where the brutal, race-to-the-bottom competition happens. Software and managed services command real pricing power. Average revenue per user has crossed $100, and DigitalOcean recently pushed through a 30% list price increase on its GPU capacity with, by the company's own account, minimal customer churn — a good sign that customers value the product enough to accept a price increase rather than walk. All of this supports the 40% adjusted EBITDA margin and 24% adjusted operating income margin the company is now running at scale.
11 The Paperspace Deal, Revisited
Every AI story needs an origin point, and DigitalOcean's traces back to a relatively small acquisition that's aged extremely well.
In July 2023, DigitalOcean paid $111 million to acquire a company called Paperspace, which specialized in GPU computing for AI workloads. At the time, it was a modest deal, easy to miss. Over 2024 and 2025, the company quietly wove that acquired technology into its core platform. By April 2026, that work had turned into the full AI Native Cloud and Inference Engine we've described in this note — including direct access to NVIDIA's H100 chips. The payoff: that original $111 million purchase has grown into a $234 million and still-climbing annual run-rate of AI revenue in about three years. That's the kind of capital allocation we like to see — a relatively small, well-timed bet that management then followed through on and actually built into a real business, rather than letting it sit on the shelf.
12 Building the Physical Plant
Since this is ultimately an infrastructure business, it's worth looking at the actual physical buildout, and at how the company is paying for it.
DigitalOcean now has 155 megawatts of committed data center capacity worldwide, having added 20 megawatts in the second quarter alone, with new facilities in Richmond, Virginia and Kansas City, Missouri both coming online ahead of schedule. On the financial side, the company proactively bought back $472 million of its 2030 convertible debt, cutting its debt load without meaningfully diluting shareholders in the process. The result is a pro forma net leverage ratio of just 0.7 times — a genuinely strong balance sheet for a company in the middle of an expensive infrastructure buildout. That gives management real flexibility to keep expanding capacity and to negotiate better terms on equipment financing, instead of being forced into a corner by debt covenants during a critical growth phase.
13 Wall Street Is Catching On
We don't put a lot of weight on what other investors think, but it's a useful sanity check when a large, credible institutional investor reaches a similar conclusion independently.
Goldman Sachs Asset Management's Small Cap Growth Fund specifically called out DigitalOcean in its first-quarter 2026 commentary, describing it as well positioned to capture AI-driven growth while holding onto its core small-business customer base, and noting the company was a top contributor to the fund's returns over that period. This lines up with our own read of the business: the AI growth story is real, but it isn't replacing DigitalOcean's original customer base, it's being built on top of it.
14 Where Guidance Is Headed
Management's own forecasts have been getting more confident with each passing quarter, not less, which is generally a good sign when it happens alongside actual results beating expectations.
Full-year 2026 revenue guidance was raised to a range of $1.17 to $1.18 billion, which works out to roughly 30.5% growth for the year, alongside raised guidance for adjusted free cash flow margin, now expected at 11% to 13%. Management is projecting the growth rate will keep accelerating into year-end, targeting an exit rate of 35% or higher in the fourth quarter. Perhaps the most telling number here: remaining performance obligations — essentially contracted future revenue that hasn't been recognized yet — reached $894 million, twelve times higher than a year ago. That's a real backlog of signed business, not just optimistic talk, and it's the main reason management says it has real conviction in reaching 50% or higher revenue growth for the full year 2027.
15 The Price Tag: Then, Now, and Relative to Itself
Here's where we have to be honest about the stock's recent history, because it has been genuinely wild, and it matters for how we think about price. DigitalOcean stock started 2025 around $75 and finished the year at roughly $25 — a loss of more than 66% for the year, at a time when the underlying AI business was already starting to inflect. That's the kind of disconnect between price and business reality that we find interesting rather than scary. From there, as the AI story became impossible to ignore, the stock ripped higher, eventually touching a 52-week high of $187.50 in early June 2026. Since then it's pulled back to today's $136.00, down about 27% from that June high, even as the company just raised guidance and posted record incremental ARR.
On traditional measures, the stock isn't cheap. It trades at roughly 65 times trailing earnings, which sounds expensive by old-fashioned value-investing standards. But that number alone misses the point for a company whose revenue growth just doubled and whose AI segment is growing at over 200% a year. The more useful comparison is the stock against its own recent history: even after this year's rally, DOCN is still trading well below where it was at its 2026 peak, despite the fact that every fundamental metric — revenue growth, ARR, margins, guidance — has gotten better, not worse, since that peak. The analyst community, for what it's worth, broadly agrees with this read: the consensus rating across 16 Wall Street analysts is "Buy," with an average 12-month price target of $175.21, roughly 29% above where the stock trades today. Recent moves have mostly been upgrades — Citi raised its target to $190, Barclays to $161 following the Q2 report, and Stifel upgraded the stock to Buy after the summer selloff, citing valuation. A handful of analysts, including UBS, have trimmed targets, largely on concerns about pricing pressure from smaller "neocloud" rivals rather than anything wrong with DigitalOcean's own execution.
16 What Could Go Wrong
- Price wars at the bottom of the market: smaller GPU rental competitors, named in the company's own materials as Linode and Vultr, are cutting prices aggressively on raw bare-metal computing. DigitalOcean's defense is that 85% of its AI revenue now comes from software and managed services rather than raw hardware rental, which insulates it from the worst of that price competition — but it's a real and ongoing pressure to watch.
- Getting enough of the right chips: next-generation AI chips like NVIDIA's H100s are in constrained supply industry-wide. DigitalOcean's strong balance sheet (0.7 times leverage) gives it room to secure priority access, and its software layer is increasingly built to work across different hardware, but a prolonged shortage would still slow expansion plans.
- The technology keeps moving: workloads are shifting quickly from simple AI question-and-answer requests toward autonomous AI agents doing multi-step tasks, which need more sophisticated infrastructure than basic inference. DigitalOcean is investing heavily to keep up — new tools for autonomous operations and secure agent environments — but this is a fast-moving target and staying ahead of it isn't guaranteed.
- The stock's own volatility: a name that fell 66% in one year and then rose more than 600% in the following months is not a low-volatility holding. Investors who can't stomach large swings in either direction should size any position accordingly.
- Valuation still matters: a 65 times trailing earnings multiple leaves little room for error. If growth decelerates meaningfully from here, or if AI infrastructure spending broadly cools off, the stock could re-rate lower regardless of how well DigitalOcean itself is executing.
17 Muffett's Take
Set aside the fact that we're existing customers for a moment, and the case for DigitalOcean still holds up on the numbers alone. Revenue growth just doubled year over year. The AI side of the business is growing at triple-digit rates while the original small-business cloud franchise keeps humming along underneath it. Margins are expanding, not shrinking, as the business scales. The balance sheet is clean. Guidance keeps going up, not down. And the stock, despite all of that, is still trading meaningfully below where it was earlier this year.
We think DigitalOcean has found something genuinely hard to copy: a way to sell AI computing power to the exact customers — startups, independent developers, small and mid-sized digital businesses — who are least equipped to build it themselves, using the same simple, trusted platform those customers already knew and liked. That's not a flashy story, and it's a smaller company than the giants that usually dominate AI headlines. But it's a real business, with real earnings, growing into a real and expanding market. We'd start with a modest position here and add on further confirmation — a few more quarters of the current trajectory holding up, and continued discipline on margins as the AI buildout continues. Given how sharply this stock can move in both directions, we're comfortable if it gets more attractive before it gets more expensive.
