AI is rapidly driving up data center prices: Silicon Valley vacancy rates are near historic lows, and cloud computing costs are skyrocketing

Image source: Generated by Unbounded AI

With the popularity of ChatGPT supported by Microsoft at the beginning of the year, and the rapid launch of large language model (LLM) and generative artificial intelligence competing products by competitors such as Google and Meta, AI seems to be rapidly advancing into more aspects of human life, and the related demand is also rising. .

What is little known is that the AI fire has also led to a soaring cost of data centers. The latest report points out: "The energy use associated with running artificial intelligence number calculations is rapidly becoming a key driver of rising data center costs."

Soaring demand for artificial intelligence has prompted some data center operators to raise commercial lease prices to account for the extra cost of powering and cooling stacks of computer servers running increasingly energy-intensive workloads.

According to CBRE Group, one of the world's largest commercial real estate services companies, data center customers, ranging from small businesses to large cloud service providers, are currently consuming power faster than data center operators can expand capacity. Increasing supply constraints due to increased use cases for artificial intelligence are putting upward pressure on prices charged by data centers.

For example, in Northern Virginia, the United States, the world's largest data center market with more than 275 facilities, the electricity available for lease this year has decreased from 46.6 MW a year ago to 38.4 MW, a drop of up to 17.6%, while the overall electricity inventory Actual year-over-year growth of 19.5% to 2,132 MW was mainly due to the large and rapidly increasing power consumption of GPUs used to train generative AI models.

Additionally, the extra power associated with AI needs will need to be complemented by more advanced hardware cooling systems that are not only energy-intensive, but also tend to be more expensive and have a larger footprint than traditional air coolers, all of which are part of the AI-powered cooling system. One of the drivers of high data center prices.

At the same time, when the demand for AI is strong, the electricity bills paid by data center customers are also higher. John Dinsdale, chief analyst at Synergy Research Group, a market research firm, admitted that global data center operators are passing the extra costs of running AI applications directly on to customers. **

According to CB Richard Ellis statistics, in the first three months of this year, data center customers in Northern Virginia in the United States paid as much as $140 per kilowatt of electricity per month, an increase of 7.7% from $130 a year ago. In Silicon Valley, where data center vacancy rates are now at a near-record low of 2.9 percent, the maximum price per kilowatt per month for customers has climbed to $250 a month, a jump of 43 percent from $175 last year.

According to reports, artificial intelligence applications consume more energy than traditional software because they are designed to read larger amounts of data. While a single AI model can consume tens of thousands of kilowatt-hours of electricity in a matter of days, a generative AI model can be 100 times larger than standard AI tools. Market research firm Enterprise Technology Research surveyed about 500 enterprise IT decision makers this year, and more than half of them said they planned to evaluate, deploy or invest more resources in generative AI technologies like ChatGPT.

Another analysis pointed out that since many cloud providers also lease data center space, as more and more companies adopt generative artificial intelligence, rising data center costs may lead to higher cloud computing fees. Given that generative AI workloads require more computing, it is bound to affect the energy efficiency and cooling system of the data center more broadly, that is, the impact will be across all aspects of the industry.

According to the popular science article of Lanyang Technology, the industry generally believes that data centers have high energy consumption and fast growth, and the proportion of my country's energy consumption is increasing year by year. The energy consumption of the data center is relatively concentrated. In addition to IT equipment, auxiliary facilities such as refrigeration systems have a high proportion of energy consumption:

“Electricity used by IT equipment is ultimately converted to heat, which must be removed by data centers using cooling equipment that also consumes energy. On average, cooling systems and servers account for the majority of data center energy consumption, followed by networking equipment and storage driver."

Energy consumption is also directly related to operating costs. According to the modeling forecast of consulting firm Tirias Research, the power consumption of data centers will be close to 4,250 megawatts by 2028, an increase of 212 times compared to 2023, and the total data center infrastructure plus operating costs may exceed 76 billion US dollars.

This could be called “generative AI disrupting the data center” as its growth challenges the business models and profitability of emerging services such as search, content creation, and business automation incorporating generative AI, at higher costs It is more than double the annual operating cost of Amazon AWS.

The innovative capabilities enabled by generative AI come at a high cost in terms of processing performance and power consumption, the agency said. So while the potential of AI may be limitless, physics and cost may ultimately be the boundaries.

In order to reduce costs, the agency suggested that highly optimized, even simpler and more specialized small neural network models can be used to reduce data center costs by reducing the model size in the cloud and using massive parameter networks for fast training of smaller neural network models. models and move workloads completely out of the cloud, making it more cost-effective to distribute generative AI applications to run on distributed platforms such as smartphones, PCs, vehicles, and mobile XR products:

“Five years ago, companies started sounding the alarm about data center power consumption at the annual Hot Chips semiconductor technology conference, when it was predicted that global computing demand could outstrip global electricity generation within a decade. That was before the rapid adoption of generative AI , while generative AI has the potential to increase computational demands at an even faster rate. There is no such thing as a free lunch - consumers will demand better generative AI output, which will offset gains in efficiency and performance. Costs will inevitably climb as consumer usage increases. Offloading computing to the edge and distributing it to clients such as PCs, smartphones and XR devices is a key way to reduce capital and operational costs. "