Hyperscale cloud companies like Microsoft and Google operate vast data centres containing tens of thousands of servers. These facilities represent massive upfront and ongoing capital investment.
As such, hyperscale cloud companies are seeking new ways to reduce cost, optimise upgrade cycles, and reduce environmental impact. It might seem like a strange approach for facilities that span millions of square feet and consume multiple megawatts of power, but the prevailing strategy for these companies is figuring out how to do more with less. For example, for several years, hyperscalers have been running their servers hotter and hotter to save on cooling.
The next frontier for hyperscale cloud infrastructure efficiency is the upgrade cycle. This is especially important as many hyperscalers’ large facilities are being repurposed for AI workloads, which are much more demanding than cloud storage and computing.
Stretching the servers’ lifespan
In 2020, the industry-wide understanding was that servers in hyperscale data centres would complete an upgrade cycle every three years. Then, Meta managed to push the lifespan of its servers from three years to four at the end of 2021. By the next year, it had pushed that figure to four and a half years, then to five by 2023.
Similarly, Amazon Web Services’ servers were operating with five year life spans by the start of last year. Microsoft was operating its servers for around four years at the same time.
Now, Amazon has pushed the “useful life” of its servers to six years. It achieves this through a “robust maintenance and repair program designed to increase component reuse and further reduce carbon emissions and waste across [AWS’] supply chain.”
Long server life spans mean big savings
Lengthening the upgrade cycle in data centres could lead to significant reductions in e-waste. This is especially true in hyperscale facilities that are home to tens of thousands of servers. For cloud operators like Amazon and Meta, however, the real rewards are financial.
When Amazon pushed its server lifespan to six years, it reduced its expenditures by $900 million in a single quarter. Google also hit the six year lifespan mark in 2023. As result, the company saved $3.9 billion in depreciation. Also, it increased its net income by $3.0 billion over the course of the year.
What about the suppliers?
Less frequent server swap-outs does bode ill for the third party manufacturers and suppliers in the hyperscale ecosystem. Billions fewer dollars spent on new hardware is bound to hurt the companies who design and manufacture it.
However, one thing will likely compensate for the drop in demand. Right now, virtually all hyperscalers are growing their operations just as fast as their need for replacement servers drops. The majority of this demand is expected to stem from AI. Throughout the industry, new money is pouring into building facilities capable of supporting the higher load from generative AI.
“AI infrastructure spending is propping up the revenue streams for servers and storage,” noted The Next Platform. At the same time, the article adds that “underlying spending for datacenter gear for other workloads has gotten even weaker” compared with H1 2023.
Early this year, Baron Fung, a Senior Research Director at Dell’Oro Group noted that data centre spending to support AI workloads would likely exceed $200 billion by 2028. “In order to drive long-term sustainable growth, the cloud service providers will seek to streamline general-purpose computing infrastructure costs by transitioning to next-generation server platforms and rack-scale architectures,” he added.
- Infrastructure & Cloud
