Vultr Steps Up for Open Source Just as DigitalOcean Steps Back

TL;DR: Vultr launched open source credits the same week DigitalOcean dropped its own, and a week after a US$1.2 billion bet on AMD Helios racks. The credits protect that hardware bet, because Helios depends on open software like ROCm and PyTorch.

• Helios is the first non-NVIDIA rack-scale system to win a billion-dollar deal, so the single-vendor era is cracking.
• In a multi-vendor market, open source is the only software layer no vendor owns, and that’s where lock-in gets decided.
• The risks: credits are easy to cut, Helios has no public rollout date, and ROCm still trails CUDA.
• Expect other neoclouds to launch rival credits programs soon.

On October 6, Vultr launched an Open Source Credits Program offering infrastructure credits and community support to actively maintained projects, from solo-maintained libraries to mission-critical infrastructure.

The timing is pointed: days earlier, DigitalOcean quietly told projects such as Node.js and Pidgin that its own program would take no new applications or renewals, and only offered Node.js a revamped deal after maintainers discussed dropping its logo. The launch builds on Vultr joining the FinOps and Tokenomics Foundations in September and on a Cloud Alliance model that favors partners over proprietary services.

DigitalOcean's retreat says more about priorities than budgets. The company had already dropped its GitHub Student Pack credits earlier this year without fanfare, and only weeks ago it pledged US$3 million to the foundation behind Omarchy Linux. The impact is real for smaller projects: Pidgin's lead developer says he had been covering hosting costs out of his own pocket for over a year. Maintainers who are already wading through low-quality, Artificial Intelligence (AI)-generated pull requests now have an infrastructure bill to worry about as well.

Vultr's launch also arrives a week after Hewlett Packard Enterprise (HPE) announced a US$1.2 billion Vultr order for AMD Helios racks, each running 72 MI455X Graphics Processing Units (GPUs) on the open-source ROCm stack and standards-based Ethernet, with deployments reportedly due from 4Q 2026. It is HPE's first Helios order, and AMD claims up to 2.9 exaflops of FP4 compute and 31 Terabytes (TB) of HBM4 memory per rack.

Vultr has been backing AMD since it brought MI300X capacity and ROCm to its Kubernetes engine in 2024, so this is less a pivot than a doubling down. Read together, the moves are a clear statement that Vultr sees open software, not any single chip vendor, as the foundation of the next AI cloud. Here is the full run of recent announcements:

Table 1: Recent Vultr Market Activities

It would be easy to file the credits program under developer relations, but that misses the point. A Helios rack is only as good as the open software running on it, and ROCm, PyTorch, virtual Large Language Models (vLLMs), and the libraries beneath them will decide whether those 72 GPUs earn their keep.

Vultr has long pitched itself as the cloud that builds on open-source Kubernetes and AI projects instead of copying hyperscalers' proprietary services, and this is that pitch with money behind it. Funding the people who maintain that stack is the cheapest insurance Vultr can buy on a billion-dollar hardware commitment. Because Vultr also runs NVIDIA GB300 systems, this is hedging rather than a crusade, but a cloud selling both stacks needs portable, open software more than anyone.

The same logic runs through the rest of the stack, where Helios wires its GPUs together over standards-based Ethernet rather than NVIDIA's integrated, InfiniBand-first rack designs. On the software side, Vultr is working with NVIDIA on NemoClaw, an open-source stack for running always-on AI agents on open models such as Nemotron. That consistency, with openness showing up in hardware, software, and community funding alike, is what makes the credits program credible rather than cosmetic.

There is a wider backdrop as well, with policy attention converging on a small club of frontier labs, which leaves open-model builders leaning on neutral infrastructure that independent clouds like Vultr increasingly provide.

The skeptics have fair points, too: credits are easy to cut, as DigitalOcean just proved, Helios has no public rollout calendar, and ROCm still trails CUDA in ecosystem depth. The best answer is to make the program boring in the best sense, with multi-year terms, published criteria, and an annual account of who received what. Trust is the whole product here, and DigitalOcean has just shown how quickly it evaporates.

The Helios order matters well beyond Vultr, because it is the first real proof that a non-NVIDIA rack-scale system can win a billion-dollar deal. For 3 years AI infrastructure has effectively been a single-vendor market, and this deal suggests that era is ending at the rack level, not just the chip level. Neoclouds are the obvious winners, and HPE gets a story it badly needed, recasting Juniper networking as the open alternative inside the AI rack.

In a multi-vendor market, whoever controls the software layer controls switching costs, and open source is the only layer no single vendor owns. That makes the small credits program the more revealing move, and it makes DigitalOcean's retreat look shortsighted.

ABI Research's view is that the AI cloud market is splitting into integrated stacks built around one vendor and open, composable stacks assembled from several, and Vultr has picked its side. If Helios ships on time and the open ecosystem keeps maturing, independent clouds will gain genuine pricing power for the first time; if it slips, the window for challengers will narrow for years.

Expect other neoclouds to answer with credits programs of their own within a couple of quarters, and expect DigitalOcean's promised revamp to be judged against Vultr's terms. The PyTorch Conference on October 20–21 is the obvious place to watch for Vultr naming its first funded projects. Either way, open source has stopped being a line item, and Vultr now has to prove it will still be paying the bill in 2028.

View more research into Vultr from our Data Centers analyst team.