The news
Hewlett Packard Enterprise (HPE) said on September 30, 2026, that cloud provider Vultr placed a $1.2 billion order for AMD Helios AI racks, HPE's first order for the system. HPE disclosed it at its Networking Investor Day in a release filed with the SEC.
HPE describes Vultr as the world's largest privately held cloud infrastructure company. Vultr is one of the so-called neoclouds, newer cloud providers that rent out AI computing capacity. It plans to use the systems for AI model training and inference in US data centers, serving managed service providers and end users, Techzine reported.
Each Helios rack combines 72 AMD (AMD) Instinct MI455X graphics processors, AMD EPYC "Venice" central processors and AMD Pensando Vulcano AI network cards, according to Techzine and Constellation Research. Six HPE Juniper Networking QFX5252 switch trays tie all 72 GPUs together over Ethernet inside the rack, a design known as scale-up networking. Techzine reported the liquid-cooled racks deliver up to 2.9 exaflops of FP4 computing power, a low-precision format used for AI, and 31 TB of HBM4 memory each.
HPE did not disclose how many racks the order covers or when they will be delivered. HPE CEO Antonio Neri said Helios shows HPE can design a complete AI system spanning rack-scale compute, scale-up networking and direct liquid cooling, Techzine reported. Vultr CEO J.J. Kardwell said the company is running at capacity, with demand exceeding supply, according to the same report.
At the investor day, HPE set fiscal 2027 networking revenue growth of high teens to low 20s percent and an operating margin in the mid-to-high 20s percent. It now targets $800 million in annual run-rate cost savings from its Juniper Networks acquisition by the end of fiscal 2028, up from $600 million, according to the filing.
The numbers
- Vultr order
- $1.2 billion
- AMD Instinct MI455X GPUs per rack
- 72
- HPE Juniper switch trays per rack
- 6
- FP4 compute per rack (Techzine)
- Up to 2.9 exaflops
- HPE fiscal 2027 networking revenue growth target
- High teens to low 20s percent
- Juniper cost savings target, end of fiscal 2028
- $800 million run rate (was $600 million)
Why CEOs should care
For companies renting AI compute, a $1.2 billion AMD rack order at a neocloud means more AMD-based capacity is heading to market. CIOs and CTOs whose AI workloads run on Nvidia (NVDA) hardware should ask their cloud providers when AMD MI455X capacity will be available, what it costs per hour of training or per million tokens of inference, and how much work it takes to move their models and software over. Even if you stay put, a credible second quote is leverage in GPU capacity negotiations.
For infrastructure and network teams, the design matters as much as the chips. Until now, networking inside the AI rack has largely been a proprietary, single-vendor choice, analyst Zeus Kerravala of ZK Research wrote in SiliconANGLE; Helios uses Ethernet and UALink over Ethernet, an open-standard approach. Ask vendors which parts of a rack can be swapped or sourced from someone else, and what that does to support contracts.
For CFOs and boards, delivery risk is the caveat. Rami Rahim, who leads HPE Networking, described HPE as still supply-limited, just less so, and Kerravala wrote that its fiscal 2027 outlook depends on how quickly it secures components. Any contract for Helios capacity should carry firm delivery dates and remedies if they slip.
The bigger picture
The order is a test of HPE's Juniper strategy a little over a year after it closed the acquisition. Kerravala argued that Juniper on its own could not have won the Vultr deal, but HPE could. Rahim said Helios represents more than a $1 billion networking opportunity over the next two years, with orders for the networking trays already above $200 million, Kerravala reported.
HPE's longer-range targets point the same way. According to the filing, HPE aims for high-teens percent compound annual networking revenue growth from fiscal 2026 to 2029, led by data center networking at low-to-high 50s percent a year, while campus, branch and security grow in the high single digits.
What’s next
Watch for the number of racks and the delivery schedule, which HPE has not disclosed, and for a second named Helios customer. HPE executive Rami Rahim, who leads HPE Networking, said HPE underestimated how fast the market is growing, SiliconANGLE reported, so component supply will decide how quickly orders like Vultr's turn into revenue.
What “Fact-checked” means
Fact-checking means testing a story’s facts against the evidence before it is published. This story went through at least two separate checks before this version was published.
- What we checked
- Its names, figures, dates, job titles, quotes and who said what were checked against the story’s sources, including its main source where it could be opened. The headline was checked for accuracy and overstatement.
- How
- A first check reviewed the whole story. If it passed, a second, skeptical check went back to the sources to look for mistakes in the most important facts. If a check flagged the story, it was edited to fix the problems found, and a separate re-check then reviewed the whole story again.
- Who
- The checks are made by our newsroom, as steps kept separate from the writing, under rules set by our editor, Hussein Mukhtar. A story the checks still flag is held for the editor, who decides whether it is fixed, published or dropped.
- If something is wrong
- “Fact-checked” does not mean error-free. If a material error is found after publication, we correct the story and add a note saying what changed. Report an error
Companies in this story
