As Gen AI gains traction, would be curious to learn how your org is preparing its cloud budgets to support specific Gen AI initiatives. Beyond the potential increase in cloud spending, are there other cost considerations specific to Gen AI, such as specialized hardware, unique training needs, or talent acquisition that you're actively setting aside budget for?

We are hiring a position with the specific focus of getting us ready for and implementing our AI initiatives. We are currently focused on data governance and quality. Once we feel we are in a good place with our data, we will engage in the internal AI development process.

It's an all of the above, really. I think we will need new or upskilled talent, some unique hardware and training to accommodate.

It's likely going to be either a little "net new" or figuring out where to allocate existing budget to Gen AI in lieu of other things.

For us, it has given rise to a number of nuanced discussions. From the most "basic" - how do you treat investment in generative AI, as a capital cost or an operating cost; to the more involved costs. 

Given the direction most commercial AI/LLM models are going with it being a subscription base, I think that gives a degree of certainty on what is most likely the largest or one of the largest cost drivers. Unless building your own (which we are also doing) the commercial offering means costs like hardware and storage are not there, your enterprise subscription and entitlement shifts the accountability to the vendor.

So if subscription, I think your other costs are attracting, training and retaining professionals to help govern, manage, secure, run and support your platform (from an end user support).

If building your own, I think you would want to chart and track a number of key metrics for at least a few months to get a sense of growth. For example, I'd expect exponential storage growth if you want your model to continue to consume large data volumes (and also as you look at performance and throughput based on disk type, SAN and LAN connectivity, etc). As users become more familiar and reliant on the tool, horizontal scalability (processors, compute capability)

So to come at it differently, I might suggest you wrestle with the fundamental question on how you plan to deliver generative AI in your org? Buy or build?