Off-Prem

PaaS + IaaS

Server shipments fall ... just as AI drives demand for costlier kit

Conflicting feedback makes 2023 an especially difficult year to predict


Changing priorities among enterprise and cloud providers are driving apparently conflicting trends, according to research outfit Omdia, with demand for compute resources remaining high, especially as ChatGPT has spurred interest in AI training, yet at the same time server shipments are declining.

Omdia’s Cloud and Data Center Market Update predicts a contraction in server shipments for the first time since the global financial crisis of 2007-2008. However, it warns that conflicting feedback and data points are making 2023 a particularly difficult year to predict.

A change in priorities for both enterprise buyers and large cloud service providers is at the root of this, it believes. One trend appears to be that more costly servers that are configured for AI model training are being prioritized over the usual server refresh cycle, driven by the recent surge of interest in Generative AI models caused by ChatGPT.

Omdia said that 2.8 million servers were sold during the first quarter of 2023, and that this was significantly below its expectations, not to mention those of the server OEMs themselves. The decline in shipments from 4Q22 to 1Q23 is the largest on record, it claims.

Those OEMs have apparently told Omdia that many enterprises are now reluctant to commit to large projects, with IT spend pull back seen particularly among financial services and manufacturing companies in North America.

Meanwhile, server shipments to cloud operators were said to be healthy during the first quarter, but a sharp decline seen during the second quarter has now forced Omdia to revise its full year forecast so that it now expects to see server shipments going to the big cloud providers to decline by 2 percent for this year.

Microsoft, Google and Meta were all investing in new servers in Q1, while AWS pulled back somewhat on new server installations, but by Q2 both AWS and Meta are holding back on spending, Omdia said.

Sweating the assets

Another contributing factor in the mix is lengthening server lifecycles. The normal lifespan for server kit at all the big cloud providers is now 6 years, while a step down a level to the tier 2 providers shows that some are reporting life spans of nine or 10 years, according to Omdia.

It foresees that a two-tier system for cloud services could soon become the norm, with cloud providers offering a premium tier using the latest server hardware alongside a value tier based on servers older that are more than five years old. This is not new, Omdia concedes, but the model appears to be spreading.

Euro cloud operator OVHcloud told The Register recently that it operates just such a scheme, because many customers do not require high performance and are “very happy” with servers that are 6, 7, 8 or even 10 years old, for a discounted price.

Omdia indicated that such deployments of refurbished second hand servers have surged since the pandemic, where they were chiefly used to offset the long lead times caused by semiconductor shortages.

OVH also operates a “reverse supply chain”, where old servers are returned to the factory (it builds its own servers), and disassembled and components individually tested to see if they can be repurposed or sold on to the grey market via brokers.

This also chimes with what Omdia is seeing; it notes that the useful life of some components will be longer than that of the servers they are part of, and a movement to recycle and reuse components is gaining traction at cloud operators. It cites reuse of memory modules leading the way, followed by power supplies and Ethernet adapters.

Looking to the future, Omdia said it expects the growth in the server installed base to slow down while shipments of new servers will accelerate, driven by a new server refresh cycle.

However, the datacenter industry in general is in good health, according to Omdia, with capex levels at the hyperscale companies still at high levels and datacenter buildout, as measured in square feet/meters, continuing to grow by about 10 percent each year on average. ®

Send us news
3 Comments

Tech world forms AI Alliance to promote open and responsible AI

Everyone from Linux Foundation to NASA and Intel ... but some big names in AI are MIA

HPE targets enterprises with Nvidia-powered platform for tuning AI

'We feel like enterprises are either going to become AI powered, or they're going to become obsolete'

Creating a single AI-generated image needs as much power as charging your smartphone

PLUS: Microsoft to invest £2.5B in UK datacenters to power AI, and more

Thirty-nine weeks: That's how long you'll be waiting for an AI server from Dell

Revenue and net income down, server market flickers, PCs fail to ignite

Server shipments to fall 20% this year, but AI means vendors still raking it in

Less is more, as hyper heterogeneous computing heats up

Don't be fooled: Google faked its Gemini AI voice demo

PLUS: The AI companies that will use AMD's latest GPUs, and more

The AI everything show continues at AWS: Generate SQL from text, vector search, and more

Invisible watermarks on AI-generated images? Sure. But major tools in the stack matter most

Mere minority of orgs put GenAI in production after year of hype

Folks are dipping their toes in without a full commitment

EU running in circles trying to get AI Act out the door

Bloc risks missing out on first-to-legislate status if timetable slips

Google launches Gemini AI systems, claims it's beating OpenAI and others - mostly

Gemini accepts text, images, audio, and video and comes in three flavors

UK and US lead international efforts to raise AI security standards

17 countries agree to adopt vision for artificial intelligence security as fears mount over pace of development

AMD slaps together a silicon sandwich with MI300-series APUs, GPUs to challenge Nvidia’s AI empire

Chips boast 1.3x lead in AI, 1.8x in HPC over Nv's H100