DigitalOcean has signed several nine-figure annual deals for AI inference and cloud services, and the more telling number isn’t the size of those contracts but how long they run.
The company says its remaining performance obligations, essentially contracted business it hasn’t recognized as revenue yet, will jump more than tenfold from the second quarter of fiscal 2025 to over $800 million. Average contract length will grow from 1.6 years to more than three. In an industry that spent the past two years scrambling for GPUs, power and floor space, that shift toward multi-year commitments matters more than it might sound.
Investors have grown wary of infrastructure spending that lacks real, lasting demand behind it. Short pilot projects don’t prove much. Longer contracts do, though they also lock providers into delivering capacity years out, in a market where pricing and model performance keep shifting underneath them.
DigitalOcean seems to want a spot between the hyperscalers and the bare-metal GPU rental shops. CEO Paddy Srinivasan argues customers aren’t just buying compute, they’re buying a platform, pointing to tools like the company’s Inference Router, which routes workloads across different AI models based on cost and performance rather than leaving that decision to the customer.
That’s a bet on software mattering as much as hardware. Plenty of competitors still sell on GPU availability and price alone. DigitalOcean is trying to convince buyers that developer experience and orchestration deserve a bigger share of the decision, a case the market will test as more companies move AI projects out of pilot mode and into production, where reliability and cost start to matter a lot more.
None of this makes physical capacity less important. DigitalOcean has added another 20 megawatts of data center capacity, due online in late 2027 and early 2028, bringing its committed footprint to roughly 155 megawatts. Notably, even a company touting strong demand today is talking about infrastructure that won’t exist for another two or three years, a reminder that construction timelines, not chip supply, have become the real bottleneck.
The company also expects second-quarter revenue to grow around 29% year over year, with results likely landing at or above its own guidance. Contracted demand reduces uncertainty, but delivering on it over several years still depends on power availability and supply chains staying cooperative.
