DigitalOcean borrows $725M to buy its AI cloud — the machines have to pay for themselves

For years, DigitalOcean was the cloud investors scrolled past. The New York company sold cheap virtual servers to solo developers and two-person startups — a fine little business, a dull one, and for a long time a no-growth one. That version of the company is over, and this week it signed the document that shows how completely. On September 10, DigitalOcean announced a $725 million equipment financing facility to buy the GPU and CPU hardware its cloud now runs on, weeks after guiding to revenue growth above 50% next year and with its shares up more than 170% year to date.

The loan matters not for its size but for what it reveals. Equipment financing is how GPU clouds buy their machines: lenders take the hardware itself as collateral, because a Nvidia accelerator holds resale value a bank can underwrite. The facility matures in 2030, can grow by another $300 million through an accordion option, and is led by MUFG as administrative and collateral agent with Wells Fargo, BMO and Axos as joint lead arrangers. DigitalOcean didn't just ask for growth capital — it bet the physical machines are worth enough to back the loans.

Borrowing against demand that's already on paper

No lender signs that kind of facility without a believable reason the hardware will generate revenue, and here the story stops being about the loan and becomes about the demand behind it. DigitalOcean's remaining performance obligations — the value of services it has contracted but not yet delivered — reached $894 million, roughly twelve times the $71 million of a year earlier. It signed its first nine-figure annual customer commitments, extending average contract length from under two years to more than three. Its AI customer annual recurring revenue hit $234 million, up 212% year over year, with 85% of that from inference and core cloud rather than bare-metal server rental.

That is the difference between a keynote and a commitment. Twelve times contracted backlog, locked in over multi-year terms, is the kind of evidence that turns a marketing claim into something a CFO can borrow against. This is why the "AI Native Cloud" story deserves to be taken seriously rather than treated as another CEO's slide deck.

The lean company is deliberately re-leveraging

What the deal quietly converts is DigitalOcean's financial personality. For most of its life this was a low-capital business — first-half 2026 spending on equipment was about $82 million, a trivial amount for a company its size. Over the past year it did the opposite of what it's doing now: it retired $1.19 billion of early convertible debt in 2025 and another $472 million of convertibles in mid-2026, funding the latter with a share sale, in each case leaning the company out. Net borrowings were roughly $154 million against $767 million of cash heading into this deal.

Now it pivots and borrows to buy machines. The accordion could push net borrowings toward $1.2 billion. In absolute terms that is still modest against a roughly $15 billion market cap and a 40% adjusted EBITDA margin — the CFO's "low leverage" and "attractive cost of capital" talk is fair in the narrow sense, and the 2030 maturity is sensibly matched to the useful life of the equipment. The point isn't that the company is over-levered. The point is that a business which spent a year making itself lean is now deliberately loading up secured debt, and it did so because it and its bankers believe the demand is real enough to secure against silicon.

The call the stock has already priced in

That belief is the whole question for anyone looking at this stock today, because the market has moved well ahead of the financing. DigitalOcean trades at roughly 15 times trailing revenue and about 48 times trailing EBITDA on a $15.4 billion market cap, with the shares up more than 170% year to date and about a quarter in the past week alone. Some of that rally is company-specific, but part is the tide lifting the sector: analysts turned more constructive on the whole AI-infrastructure group, and neighbors like CoreWeave and Nebius rode it too.

The company's standing within that group is genuinely different, and it's worth stating plainly. CoreWeave and Nebius carry enormous backlogs but lose money; DigitalOcean is profitable, generating adjusted free cash flow margins near 22% in the second quarter. "The profitable AI infrastructure play" is a real and defensible position. But it is a position built on a specific assumption: that the inference workloads DigitalOcean is now contracted to serve remain durable and monetize at margins that cover fast GPU depreciation plus the new debt service. Palo Alto Networks' CEO put the industry's version of the worry bluntly this summer, warning that as compute scarcity eases, some neocloud valuations could compress sharply.

The financing is solved; the demand is the bet

So the $725 million facility is, on its own terms, a fair deal: cheap collateralized money, matched to the life of the assets, funding demand that is already under contract, with no dilution of shareholders. It is the answer to a financing question, and it is a reasonable answer.

What it is not is the answer to the durable question. The loan doesn't create the growth — it funds a bet on it. A company that was written off as a no-growth developer cloud is now borrowing against GPUs it must turn into profitable, recurring inference revenue for years, in a sector where the CEO of the industry's largest security vendor is warning that the scarcity premium could deflate. The deal tells you that DigitalOcean and its lenders believe the AI-native wave will hold. That's a real signal, not a proof. For a stock already up more than 170% this year, the machines have to earn more than they cost to own and finance — and that, not the loan, is what has to be true for the story to work.