(Dec 2): Cloud-computing provider Vultr is building a 50-megawatt (MW) cluster of Advanced Micro Devices Inc (AMD) artificial intelligence (AI) processors at a data centre in Ohio, a move aimed at offering AI infrastructure at a lower cost.
The AMD-backed company announced plans on Tuesday to spend more than US$1 billion on the facility, which will allow a future customer to train or run AI models. It’s slated to come online in the first quarter of 2026.
Vultr is part of a cadre of cloud providers trying to benefit from booming AI demand on a more modest budget. Its new chip cluster will operate at a much smaller scale than the massive facilities run by Microsoft Corp, Meta Platforms Inc and Alphabet Inc’s Google — but will offer computing power at lower rates, the company said.
Vultr’s cloud services are “typically half the price” of what the biggest data centre operators charge, chief executive officer JJ Kardwell said in an interview. Its 50MW, 24,000-chip facility compares with giant AI data centres measured in gigawatts of power. One gigawatt is roughly the amount of energy generated by a nuclear power plant.
Vultr, based in West Palm Beach, Florida, buys chips and computer gear and then runs the systems from leased data centre space. The new cluster — located in Springfield, Ohio — will rely on AMD’s Instinct MI355X AI accelerators, one of the most advanced processors that the chipmaker produces in volume.
“This is an unprecedented cluster of its size for an independent cloud compute company,” Kardwell said.
See also: Bank of England warns debt-fuelled AI boom could unravel
AMD, meanwhile, is trying to prove it can compete with Nvidia Corp in the market for AI accelerators, the powerful chips used to develop AI models. As part of that push, AMD has said that its processors will be more affordable than Nvidia’s. Though Vultr was an early user of the MI355X processors — and will adopt the next-generation MI450s when they’re ready — it also plans to keep buying chips from Nvidia.
There’s no customer lined up for the AMD cluster in Ohio yet, but the company said it’s in active discussions with potential clients. Several of its customers — including AI companies Clarifai Inc and LiquidMetal AI, as well as biotechnology firm MindWalk Holdings Corp — currently use its AMD-powered services.
In total, Vultr serves hundreds of thousands of customers in 185 countries, the company said.
See also: Taiwan charges Tokyo Electron in TSMC trade secrets theft case
“The commitment we made to build this cluster was without a specific customer in mind,” Kardwell said. “I would be very surprised if it’s not sold long before it’s live, but we’re not waiting for a customer contract to build it.”
Vultr was founded in 2014 and spent years offering access to central processing units, or CPUs, a chip category then dominated by Intel Corp. In 2021, it began buying the graphics processing units, or GPUs, that underpin AI computing.
Within the last two years, AI infrastructure has become the fastest-growing part of its business and now accounts for the majority of revenue. That AI concentration will only strengthen in 2026, Kardwell said.
Vultr bootstrapped itself for a decade by relying on its operating cash flow. It then raised money from outside investors last year at a valuation of US$3.5 billion. Vultr took in US$333 million from the round, which was led by investing firm LuminArx Capital Management and AMD.
In June, Vultr raised an additional US$329 million in credit financing provided primarily by JPMorgan Chase & Co, Bank of America Corp and Wells Fargo & Co. That included some US$74 million asset-backed financing in which GPUs served as the collateral. Vultr significantly expanded the credit line to help fund the AMD cluster, Kardwell said.
Companies building AI infrastructure are facing growing concerns that the industry is in an investment bubble. There also are fears that GPUs will depreciate quickly, contributing to a potential bust.
But Kardwell said that AI computing infrastructure remains “massively underbuilt”, even if some overextended companies in the industry will likely fail.
To stay ahead of Singapore and the region’s corporate and economic trends, click here for Latest Section
When it comes to how long chips maintain their value — a question raised by Michael Burry of The Big Short fame — Kardwell thinks that tech companies have actually been conservative in assessing the technology’s depreciation.
With GPUs, six years is a “reasonable, conservative estimate” of useful life, he said.
Uploaded by Felyx Teoh
