4 takeaways from Snowflake and Databricks Data + AI Summits 2024

Leaders across the organization want to use different systems for different reasons at different times. Snowflake’s customers have been asking Snowflake to provide data portability and they delivered. At Snowflake’s  data + AI summit, they announced plans to open source their Polaris data catalog. Now, IT teams can more easily share their data, making Snowflake more on par with Databricks, which always took an approach of relying on external data storage.

The shift from traditional licensing models, such as per-seat or per-core, to consumption-based models has been ongoing for years. However, the challenges of understanding and managing consumption-based spending persist for IT leaders. Run an inefficient query, misconfigure a node size or type, or make a coding error and risk a staggering and surprising bill the next month. Billing, budgeting, and cost governance tools exist in these platforms, but they are complex to use and understand. In many cases, teams are unaware of the level of risk around cost engineering and the tools available to mitigate that uncertainty.

Snowflake is trying to counter the prevailing narrative that they are expensive and prone to cost overruns. The platform uses customer-controlled computing resources on cloud providers and only charges additionally per data unit (not charging a markup on compute resources). They also don’t have a separate internal storage layer. Relying entirely on cloud-provider object stores seems to persuade cost conscious customers.

IT teams are feeling the heat to get started on GenAI and deliver on proof-of-concepts (POCs). However, the legal questions of liability, data leakage, copyright, and indemnification–around GenAI generally and LLMs in particular–are causing legal departments to slow down or halt internal initiatives. Data and AI platforms and cloud providers are negotiating with clients about fundamental legal and technological issues. However, it might take years for them to resolve. Now, if a data leak happens or a copyright law is violated, enterprises will bear the brunt of the legal repercussions. To sidestep risk, executive leaders are pressuring data teams to use sample data in POCs. Yet, it’s often impractical to create sample datasets that are representative of the real world at the volume needed. The effect is organizations caught in a logjam between different internal and external groups with varying degrees of understanding about the underlying problems and how to resolve them.

Jensen Huang, the CEO of NVIDIA, made high-profile appearances at both Snowflake and Databricks’ Data + AI Summits. Huang’s prominence, as well as the company’s massive 80%+ market share in the computational hardware space, punctuate the point that data and AI platforms want to help their customers optimize GenAI and LLM creation, training, and processing. Their ultimate aim is to keep customers from using external services for GenAI and LLM computation, helping them increase efficiency and reduce the risk of data leakage. Given NVIDIA’s dominance, all cloud providers and platforms will need to maintain partnerships with NVIDIA if they want to give customers an AI edge.