AI-Ready Data Centers Are Poised for Fast Growth Buoyed by rising demand for enterprise AI tools, CoreWeave secures a $2.3 billion debt facility to build data centers that can handle more intensive computer applications
By Angus Loten (WSJ)
Aug. 3, 2023 9:00 am ET
Spending in the global AI infrastructure market is expected to grow at a compound annual rate of 44% over the next six years, according to Data Bridge Market Research.
Companies’ insatiable appetite for artificial intelligence is driving a fast-emerging market for data centers purpose built to handle compute-intensive AI applications.
As veteran enterprise cloud vendors race to retrofit existing data centers with advanced chips and other upgrades aimed at meeting demand for AI software, some upstart builders see an opening to develop new facilities from scratch, analysts say.
Data centers are the warehouse-like buildings equipped with multiple racks of servers, routers and other information-technology hardware that store and process data. Conventional data centers are typically outfitted with servers that run workloads with standard, general-purpose chips, said Manju Naglapur, senior vice president and general manager for cloud, applications and infrastructure solutions at IT services and consulting firm Unisys.
A purpose-built AI data center, he said, houses servers leveraging AI chips, such as Nvidia’s graphics processing units, that can run multiple computations at once as AI applications sift through enormous stores of data. They also furnish optical networking and more efficient storage to support AI models at scale, he said.
“This requires a massive investment in capital and time to overhaul an existing center or to create these new purpose-built data centers,” Naglapur said.
Spending in the global AI infrastructure market—including data centers, as well as networks and other hardware that supports the use of AI applications—is expected to reach $422.55 billion by 2029, growing at a compound annual rate of 44% over the next six years, according to research firm Data Bridge Market Research.
CoreWeave, a data-center startup based in Roseland, N.J., on Thursday said it had secured a $2.3 billion debt facility, with backing by hedge fund Magnetar Capital and private-equity firm Blackstone, which will be used to accelerate the construction of AI-ready data centers. The latest financing follows the company’s $221 million Series B funding round in April and a $200 million Series B extension round in May.
Launched six years ago, CoreWeave currently has seven AI data centers online and expects to double that number by the end of the year, said Brian Venturo, CoreWeave’s chief technology officer. Its latest data center, a 450,000-square-foot facility in Plano, Texas, valued at roughly $1.6 billion, began commercial operations this week, Venturo said.
“About this time last year we started to understand what was coming,” Venturo said, citing the August 2022 release of Stability AI’s Stable Diffusion, an AI-powered image generator. Three months later, he said, the public launch of OpenAI’s online chatbot ChatGPT put the AI market into overdrive. “That’s when we thought ‘this just got real,’” Venturo said.
Among other differences, AI data centers have far more servers running on high-performance chips than conventional data centers. As a result, the average power draw of a full stack of servers in an AI data center can reach 50 kilowatts or more per rack, compared with roughly 7 kilowatts per rack in a conventional data center, analysts say.
That means AI data centers need to be built with added infrastructure capable of supplying a much higher amount of power. Since the extra power usage creates more heat, AI data centers also need alternative cooling methods, such as liquid cooling systems, to prevent equipment from overheating. Lease rates for data centers are typically based on power consumption.
Other companies building AI data centers include firms like Turner Construction, Holder Construction and DPR Construction, as well as data-center operators like DataBank.
Dallas-based DataBank in April broke ground on a 200,000-square-foot data center in Atlanta, specifically designed for high-performance computing, including generative AI, the company said. It currently operates more than 65 facilities across 27 markets. DataBank Chief Executive Raul Martynek said the pace of AI deployments, on top of already robust demand for traditional information-technology workloads, will likely lead to a shortage in data-center capacity within the next 12 to 24 months.
Facebook parent Meta Platforms late last year paused construction of a proposed $800 million AI data center in Temple, Texas, as it considers a redesign of the roughly 900,000-square-foot facility. Two similar projects, in Idaho and Alabama, are also on hold, with work on all three facilities expected to resume by the end of the year, a Meta spokesperson said.
With businesses set to deploy AI tools at a quicker pace in the year ahead, conventional data centers may not have the power to keep up. Sean Graham, a research director at International Data Corp. covering cloud-to-edge data-center trends, estimates that about 20% of total data-center capacity is being used for AI.
In Northern Virginia, a data-center market leader with more than 275 facilities, the amount of power available for lease this year shrank to 38.4 megawatts, from 46.6 megawatts a year ago, according to estimates by commercial real-estate services firm CBRE Group. Over several days, a single AI model can consume tens of thousands of kilowatt-hours, analysts say.
Existing data centers already running on near-peak loads will struggle to cope when businesses shift from testing AI apps to full-blown deployments, said Naveen Chhabra, principal analyst at IT research firm Forrester Research.
“It’s like saying you need an eight-lane road, not a two-lane road, to drive your AI apps,” Chhabra said.
Write to Angus Loten at Angus.Loten@wsj.com
AI
