DigitalOcean secured a $725 million equipment financing deal this week, with another $300 million potentially available on top of it, to buy GPUs and CPUs ahead of demand it’s still trying to prove will arrive on schedule. That timing detail is the whole story here. The company isn’t borrowing to catch up with existing customers. It’s borrowing to have hardware ready before the customers who need it have signed anything.
That’s a real shift from what built DigitalOcean’s reputation in the first place. The company grew by offering developers and smaller tech companies straightforward, relatively cheap infrastructure, Droplets, managed databases, simple hosting, without the scale or contractual weight that comes with the big hyperscalers. AI infrastructure does not work that way. It demands expensive accelerators and massive power allocations well before a single dollar of customer revenue shows up, and this financing targets exactly that gap.
The numbers explain why DigitalOcean is taking the risk. AI customer revenue hit $234 million in annual run rate during the second quarter, up 212 percent year over year, and remaining performance obligations climbed from $71 million to $894 million in just twelve months. The company also held roughly 155 megawatts of committed data center capacity by quarter’s end, with another 20 megawatts arriving in 2027 and 2028. Contracts are growing faster than the physical infrastructure the company needs to fulfill them, which is precisely the mismatch this debt aims to close.
The risk sitting underneath all of it is brutally simple: GPUs cost money whether anyone’s using them or not, and newer chips make older ones less valuable well before they actually break. DigitalOcean does have early evidence that demand is real. Token consumption among its early Inference Engine customers grew roughly 30 times over 60 days, and 85 percent of AI revenue is now coming from inference and core cloud services rather than bare metal rentals.
Whether the bet pays off comes down to sequencing, plain and simple. DigitalOcean needs the hardware, the power and the paying customers to show up in something close to the right order. Not perfectly. Just on time.
