Ignore the anecdotal stories you’re hearing about AI workloads driving a migration from the cloud to on-premises and private cloud environments. In 2024, only 27% of professional respondents in Anaconda’s latest “State of Data Science” report deploy most of their models to on-premises servers, which is a large drop from the 41% that were doing so when the same study was conducted in 2022.
In fact, the use of cloud environments has jumped among respondents who deploy models that primarily perform data science, machine learning or artificial intelligence jobs. 59% of those deploying these models say their models are mostly located in the cloud, which is up from 49% in 2022.
Despite obvious growth in the world of AI, only 64% of professionals surveyed actually deploy models into production, which is down from 76% in the 2022 study. The decline was driven by fewer developers and data engineers saying they deploy models. An explanation for the decline may be increased reliance on third-party managed large language models.
In 2024, 87% of the professionals surveyed are involved with model training and development, which is up from 85% in 2022. Among those who conduct training and model development, half as many are now doing so using on-premises local servers (10% in 2024 versus 20% in 2022). Even though more training is now being done in the cloud, local desktops and laptops are the most common place it occurs. We interpret this to mean that users have their computers connected directly to the cloud, and that local servers are not used.
Challenges to Deploying Models Into Production
Anaconda also asked about the roadblocks companies face when moving data science or AI models to a production environment. As stated earlier, fewer respondents are deploying models into production, but the question only focused on the problems faced by companies that already have moved models into production usage. Among this group, meeting IT/InfoSec standards (58%) is still the most cited roadblock when moving a data science or AI model to a production environment. Security and IT standards are particularly concerning for data engineers, with 88% saying this is a challenge.
Securing data connectivity (53%) is the second most commonly cited roadblock for companies that deploy models into production, followed by an organizational skill gap (50%). The skills gap has worsened since 2022, when only 34% noted this as a roadblock.
There is room for hope, as fewer respondents say they are inhibited by a necessity to re-code models from Python or R to another language: This ranks #7 on the roadblocks list, down from #2 in 2022. Interestingly, this drop only occurred among professionals who do not primary perform data science, AI or ML job functions. For this reason, we believe that improved tooling, including more low/no-code options, may have made this challenge less common among those who are not focused every day on data science.
Other Findings About Technology Usage
- Since Anaconda provides a managed Python product, it is not surprising that 67% of the professionals surveyed regularly use Python (always or frequently). That figure goes up to 82% among people who primarily perform a data science, ML or AI job function.
- Data cleaning, visualization and analysis (66%) is the most common way AI is used by the professionals surveyed. The next most likely uses of AI by professionals were automating tasks (52%), prediction or detection models (49%) and LLMs (i.e., chatbots; 44%).
- Data engineers are leading the way in building new tools with AI and building AI models to be used internally. Overall, 59% of professionals say their company is building new tools with AI, with data engineers (69%) even more likely to report this. Furthermore, 56% of professionals say their company is building AI models for internal use, with data engineers (65%) even more likely to report this.
About the Study
Anaconda received 3,096 responses to an online survey conducted from June 2024 to September 2024, with 36% of participants being located in Asia, 29% in North America and 15% in Europe. Anaconda provided The New Stack with raw data files for both the 2022 and 2024 studies.
Some of the findings in this article differ from those in Anaconda's published report. This is because Anaconda segmented its data based on a question about current job role, while we segmented the data based on current job function. Our approach allowed for accurate time series comparison with the 2022 study. Also note that because of small sample sizes for specific questions, we are not confident in Anaconda's published findings about job security and the types of AI jobs being hired for.
The post Cloud Deployment of AI Models Jumps, Says Data Science Study appeared first on The New Stack.
Use of cloud environments has jumped among those who deploy AI models for data science, machine learning or artificial intelligence jobs.