Aspiration Versus Action for Enterprise AI

Most of the large companies and institutions we cover here at The Next Platform have already been in the AI/ML trenches, particularly the hyperscale web and media companies (where much of this work began) and in high performance computing. However, for the broader enterprise world, the transition is still catching up as companies decide how to divvy up precious AI/ML talent and see clearly where it can be best used.

AI systems maker SambaNova, which has catered to those early adopters, took a look at this wider world in a recent survey. One hundred responses each from financial services, healthcare and life sciences, manufacturing and auto, retail, public sector and oil and gas industries – for a total of 600 responses from director or above positions – sheds light on what to expect for enterprise AI in 2022 and beyond, much of it directly in line with the IDC projection that AI/ML will grow 18 percent year over year.

70 percent of respondents have strategic technology budgets of over $100 million. 32 percent said more than 20 percent of that would go toward funding AI/ML. Those numbers aren’t huge, especially considering 78 percent said AI/ML is important to driving revenue.

For the companies that have put AI/ML to the test, 72 percent said they were measuring the success of such initiatives by the level of cost savings the programs created. Revenue growth (67%) was a close second with time savings (60%), new product development (56%) and time to insight (52%) trailing.

The strongest showing in the survey results are from the financial services side with 81 percent planning to significantly increase investments in AI/ML. However, so far only 31 percent of responding organizations in finance said they’re spending more than a quarter of IT budget in this area.

Across all industries, natural language processing was identified as the most important workload (81%) with computer vision (68%) and recommendation (55%) as the next most common. The computer vision numbers are driven by the public sector (at least in the survey) with 71 percent of that segment identifying it as the top AI/ML use case. 67 percent of the financial base tagged recommendation engines as their top use case, far above other industries.

We asked SambaNova’s Marshall Choy about how AI/ML for the broader base of enterprise is shaking out. He says that aside from the first wave of hyperscale (and their ilk) adopters with massive training clusters and custom-built models, many enterprises are using SaaS services for AI/ML analysis and other packaged software applications, especially in areas where high specialization is needed (recommendation systems, for instance).

The goal is to get up and running without added complexity, whether that means models or systems. Choy adds that for this wider AI/ML enterprise base that is still emerging, “conversations are around outcomes, not generally about speeds of hardware” or writing custom software or building complex models in-house. For the AI chipmaker, this means the bifurcation of their product line from pure systems for on-prem training and inference (DataScale) to the Dataflow as a Service managed hardware and software is right in line with the survey results.

Still, hardware is an issue. Choy says the most frequently cited challenges for AI/ML adoption among respondents include difficulty customizing models, lacking the right underlying infrastructure, and not unexpected, the AI/ML skills gap. On that infrastructure note, the survey found that 65 percent of respondents are running out of rack space for their needs and 53 percent agree that they’ll run out of computing power relative to their goals without a new architecture.

Full results from SambaNova’s 2021 survey here.

 

Sign up to our Newsletter

Featuring highlights, analysis, and stories from the week directly from us to your inbox with nothing in between.
Subscribe now

Be the first to comment

Leave a Reply

Your email address will not be published.


*


This site uses Akismet to reduce spam. Learn how your comment data is processed.