Every computing era has a default mode for enterprises.
In the beginning of the cloud era, the rallying cry was to move everything to the cloud. Now at the end of the cloud era and at the beginning of the AI era, there is change again. The mantra of ‘everything in the cloud’ is now yielding to business reality in the AI era.
The costs of cloud and the needs of businesses clearly dictate a more flexible and resilient approach for enterprises. This is particularly true for production AI. Enterprises should look at all options, including on-premises data centres (co-location with enterprise-owned equipment is counted here too), as well as cloud and hybrid solutions. On-premises installations can give the enterprise greater control in the areas of security, data sovereignty, and costs.
Every enterprise is looking at AI as a driver for future business growth, agility, and cost savings, including decision making on where and how to locate its production AI. This one of the first questions enterprises need to be answered to achieve their AI aspirations. Careful thought into the decision to go on-premises or in the cloud needs to be made for every AI workload. But this decision is not a one-time choice, enterprises should be flexible, allowing for a hybrid approach.
Business factors and IT factors both play into the decision-making process. One such key factor is the roadmap for AI. The AI roadmap, stretched out to five to seven years, is the north star for the decision-making process around where to run the production AI workloads. It helps to understand that most IT decisions and business decisions are about managing risk. Just a few of the risk considerations on IT side are: risks around security, data sovereignty, startup/operating costs, resiliency/uptime, disaster recovery and business continuity. For some of the business risks, it’s about time-to-value, how AI can increase business efficiency, and the long-term costs of AI.
Historically, most IT professionals have not considered the idea of using an on-premises data centre for AI workloads. The immediate assumption is that upgrading an existing older data centre or building a new one is simply cost-prohibitive. But in reality, that’s not always the case or the only option, as modern co-location centers can handle the load.
Additionally, there are places such as France, where AI data centres are being built at the direction of the government that will serve as a place for small companies to share AI infrastructure, and for larger firms to use for co-location, where all of the companies in a given data center share the costs.
Story Continues
Further, we are witnessing that cloud computing, as good as it is, still involves aspects of risk. Running on shared equipment can be a data security issue. Further, customers have no insight into who has physical or virtual access to the infrastructure, and cloud costs have never really come down. Yes, more functionality is available, but the pricing has still done nothing but increase.
Even more importantly, all three of the major cloud providers, Google, Microsoft, and AWS, are all based in the United States. For American companies, that’s not much of a problem. But for international companies, the issues around data sovereignty are quite real. Keeping data not only under direct enterprise control but also covered under local data regulations is a strong motivation to explore alternatives. Geopolitical instability, climate change, and the Covid-19 pandemic have shown that supply lines, data rules, and supplies of needed technology can be compromised very quickly.
An enterprise that owns its own infrastructure can exert total control of who has access to that infrastructure, ensure all local laws and regulations are respected with no worry about interference from a foreign court, and can reassure its own customers that responsibility for customer data lies with them, not with a third party.
Enterprises can also right-size their investment in AI infrastructure based on their AI roadmap and intended use. In times where money may be tight, enterprises can also stretch their investment in AI infrastructure by delaying upgrades and keeping equipment and systems a bit longer. Whereas in a cloud deployment….the bills never stop coming.
There are more factors that need examined in evaluating the AI workload deployments, including the considerations of a cloud production AI installation, the merits of a hybrid installation, and other business and IT risk factors to consider.
Take a look at this sponsored paper via the link below, where GlobalData takes a much longer and intensive look at why enterprises should adjust their thinking around where to deploy their AI production workloads.
“Navigating risk and reward: Where should enterprises run their AI workloads?” was originally created and published by Verdict, a GlobalData owned brand.
The information on this site has been included in good faith for general informational purposes only. It is not intended to amount to advice on which you should rely, and we give no representation, warranty or guarantee, whether express or implied as to its accuracy or completeness. You must obtain professional or specialist advice before taking, or refraining from, any action on the basis of the content on our site.
