The global silicon shortage has been having an impact on supply chains for some time, but now there are warnings that this is starting to affect the supply of critical datacentre components, hindering enterprise IT projects as organisations look to bounce back from the pandemic.
Naturally, the pandemic was a major factor in the supply chain problems. For example, a slump in car sales as people stayed at home caused vehicle makers to reduce their orders for chips, which in turn caused chipmakers to switch to manufacturing parts for consumer electronics, laptops and tablets, as demand for these soared.
However, shortages have now spread to many other sectors, while other factors have included droughts in Taiwan affecting chipmaker TSMC, and US sanctions against China might also have contributed.
According to a report from the datacentre resiliency think tank Uptime Institute, the pandemic, extreme weather and political instability have all disrupted global supply chains of late. This is forecast to continue, with restricted supplies of certain types of key components, including chips, power electronics and even electrical equipment.
In a survey conducted for the report, the Uptime Institute found that many suppliers expect to see problems with the supply chain for critical datacentre products and services continue over the next two years.
The prediction is that this will affect either capital expenditure (capex) projects or general availability of IT equipment –possibly both. Only one in four suppliers indicated that they expect to see no delays or impacts.
Reports of supply chain shortfalls
This matches what many in the supply chain are reporting. In September, UK-based reseller Computacenter, a major enterprise supplier across Europe, detailed the problems it foresees in the supply chain. During a half-yearly earnings call, chief executive Mike Norris said that many of its customers are now returning to business as normal following the pandemic.
However, “the ongoing supply shortages in the industry have risen to the top of our challenges”, he says, adding: “While we look forward to the supply chain issues being behind us, we are not expecting this until well into 2022, but as you can see by the performance in the first half, we are rising to this challenge.”
Meanwhile, analyst firm Omdia warns that a shortage of critical components such as power management integrated circuits (PMIC) could constrain the supplies of these for servers, especially during the final quarter of 2021. Server manufacturers have apparently been taking steps to increase their component inventory levels to mitigate the shortage and limit the impact of long lead times, but Omdia warns this may only be a short-term solution.
Part of the problem is that components such as PMICs are typically manufactured on an 8in wafer to make them competitively priced, but there has been little investment in 8in fab capacity for several years, as it was considered a low-margin business. This will make it difficult to solve supply shortages in the short term, since it takes at least a year for any expansion in capacity to come on stream. So the PMIC shortage is likely to continue until next year.
According to Omdia’s principal analyst for datacentre compute and networking, Manoj Sukumaran, the big brand server suppliers such as HPE, Dell and Lenovo are confident of managing the crisis in the short term and have not reduced their annual revenue projections.
“These vendors are the main suppliers when it comes to the enterprise market, hence there may not be a big impact on the supply situation in that market segment. Obviously there could be delays in delivery, but I don’t think it would be a big bottleneck for enterprise customers,” he tells Computer Weekly.
Things may be different in the longer term and for the hyperscale cloud service providers, because they typically source kit from whitebox suppliers that tend to operate with low inventory levels. Thus, it may be the hyperscalers that will feel the heat more than the rest of the market, if Omdia is correct.
Network switches hit
But that doesn’t necessarily mean that enterprise CIOs can relax, since datacentre infrastructure requires more than just lots of servers, and supply chain issues are having an impact for many other components and equipment as well.
Networking equipment is one example. According to Dell’Oro Group’s Ethernet switch datacenter quarterly report for Q2 2021, demand has overtaken supply this year, to the extent that sales would have been higher had it not been for the constraints imposed by restricted silicon supplies, and it expects this situation to continue through the rest of 2021.
Earlier in the year, chipmaker Broadcom warned in its first quarter 2021 earnings conference that it had already allocated 90% of its chip supply for this year, because the network and communications equipment vendors that it supplies were booking components far ahead of normal.
This was confirmed by Arista Networks senior vice-president John McCool, who said on its second quarter of 2021 earnings call that supply chain issues were the worst the company had ever seen.
“Component lead times are the highest we’ve seen and have roughly doubled from pre-pandemic norms. Most notable are semiconductor lead times, which have extended in the range of 40-60 weeks,” he says.
Component lead times are the highest we’ve seen and have roughly doubled from pre-pandemic norms John McCool, Arista Networks
McCool warns that “factories are operating near full capacity, limiting flexibility for changes in demand”, and Arista therefore expects extended lead times and escalating product costs during the rest of 2021 and 2022.
According to Omdia, this situation is again particularly bad for the whitebox vendors that supply the hyperscale customers.
“There is a severe shortage of high-speed Ethernet switch silicon, especially for whitebox vendors, and this could affect server shipments from those vendors,” Sukumaran says.
This is because many whitebox suppliers deliver up to level 11 of system integration for their hyperscale customers, which means they will integrate the servers into a rack, install rack-level networking and load the infrastructure software. The fully assembled and integrated racks are then shipped to the hyperscale customer’s datacentre for deployment.
“The scarcity in Ethernet switch silicon and top of rack switches is creating a bottleneck for shipping populated racks, and this is one of the most challenging issues affecting whitebox suppliers these days,” Sukumaran says.
Even if organisations can be sure of getting hold of all the hardware and components they require for any IT infrastructure expansion plans, it does not necessarily mean that they will have the available space in which to install them. Demand for datacentre capacity has remained strong across Europe, Middle East and Africa (EMEA) during 2021, and datacentre operators have been prioritising hyperscale customers, perhaps at the expense of making colocation space available.
Be prepared
So what actions can enterprise CIOs take to avoid having to postpone upgrades to their datacentre and IT infrastructure, which might ultimately lead to projects being put on hold and, with them, their organisation’s post-pandemic recovery plans?
There are a number of options available for organisations to call on, from bringing forward purchase plans to making greater use of cloud resources and even broadening their supplier base to consider sources beyond their preferred reseller or partner.
“The first piece of advice I can offer is that every organisation should look further ahead to identify and plan for future capacity needs,” Uptime Institute chief technical officer Chris Brown tells Computer Weekly.
“Understand that the days of near-term decisions and speedy turnarounds are gone for now, that suppliers will need time and flexibility, and that you might not be able to use your preferred provider in every circumstance. Additionally, enterprise clients should assess their spare parts inventory, along with that of local distributors, and consider keeping a wider array of equipment on hand to hedge against parts shortages and delivery delays,” Brown adds.
In other words, organisations should consider buying more than they currently need and stockpile parts and equipment in case of urgent requirements for additional infrastructure or to refresh or upgrade systems that are in service. This may mean bringing forward purchases of servers and other hardware that is at risk of shortages, such as network switches. It may also mean swallowing higher prices in the short to medium term, as some distributors warn that next year will likely see price rises.
Enterprises will also likely work harder to ensure their existing infrastructure, such as servers and storage, is being used as efficiently as possible, which ought to buy time to compensate for supply chain delays.
It has even been suggested that enterprises might turn to purchasing pre-owned hardware as a strategy to deal with hardware delays. This will obviously not appeal to every organisation for various reasons, but at least one IT services firm in the US is pushing this solution, stating that a large variety of the most high-demand current and previous generation equipment is readily available on the secondary market for immediate delivery, while delays of three to nine months are commonplace for new hardware due to persistent component shortages and supply chain issues.
Cloud to the rescue
Perhaps the most obvious way of avoiding the supply chain issues with IT infrastructure is to use cloud-based infrastructure, either as a stopgap measure or by bringing forward any plans an organisation may already have had for migrating workloads to the public cloud.
“One thing that we saw from the CIOs response to the pandemic was to accelerate the use of cloud, in all its different forms,” said Omdia chief analyst Roy Illsley.
According to Illsley, Omdia’s ICT enterprise insights survey for this year found that 46% of respondents had workloads that were running in some form of cloud – private, hybrid, public or software as a service – and this is an increase from 25% in the 2019 survey.
“While we expect some of this to drop back slightly as we exit the pandemic, as a correction from those organisations using cloud as a stopgap, the overall trend is to adopt more cloud,” he says.
The move to cloud is now well-established and the pandemic made its use even more widespread, so if enterprises have supply chain problems then the cloud will be an option as most organisations have tested it out in some form by now, he added.
Brown agrees with this, but adds that the cloud may prove even more of a necessity in light of datacentre operators favouring hyperscale customers.
”It’s important to note that this situation may force some organisations to reassess their cloud footprint. As datacentre providers continue prioritising hyperscale customers, it might become more challenging to get additional colocation space, which could change the calculus for determining which workloads and the overall amount of the business you put in the cloud,” Brown says.
For organisations that have adopted hyperconverged infrastructure (HCI) in the datacentre, expanding into the cloud may be a relatively easy move to make. VMware and Nutanix both offer versions of their HCI software platform that can be deployed into some public clouds, effectively using cloud resources as the underlying infrastructure layer instead of physical server hardware. In this scenario, the cloud infrastructure can be managed from the same console as the HCI systems in the organisation’s own datacentre.
The upshot of all this is that the pandemic, combined with other factors, is having a very complex and wide-ranging effect on global supply chains for certain key components, and datacentre components are just some of them. Enterprises should be prepared to stockpile a wider range of parts and equipment to hedge against shortages and delivery delays.
If the experts are right, the big enterprise suppliers are confident that they will not be affected for the short term, which means customers should not go short of servers. However, if datacentre operators are prioritising hyperscale customers rather than colocation, there may not be available space to deploy them. The simple answer for many CIOs may thus be to turn to cloud infrastructure, even if only for the short term.
