Benoît Dageville
Expect to see more innovators build AI-enabled apps on the technology company’s cloud-based data platform during the next 12 months. That’s the message from Snowflake co-founder Benoît Dageville, who says:
The next year will be about what our partners will build in our AI Data Cloud. Our goal is to make it simple to build fully-fledged applications. As an enterprise, you don’t have expertise in everything. You’ll need expertise from outside. I see Snowflake being like an iPhone that provides this expertise to customers as apps.
Dageville used that iPhone analogy when he spoke with diginomica at last year’s Summit, where he outlined nascent plans for Snowflake’s approach. It’s interesting to see how the company’s approach to what it calls Native Apps has evolved during the past 12 months. Dageville is now even more sure than he was last year that the direction of travel for Snowflake’s platform is a collaboration between the tech firm and its third-party partners:
The future of computing is that apps will be built at targeted subjects, with AI at the centre of these developments. These apps will be put on the Snowflake Marketplace and our customers will install the apps in their accounts. They will run these apps with all the security and governance as if they had written the apps themselves. And these apps will do amazing things because of AI.
Dageville gives the example of Maxa, which is an AI-enabled Native App that provides financial and operational insights to companies with ERP systems – and he expects many more similar applications to be available soon:
With Maxa, you can grab data from all these different ERP systems and consolidate this information on Snowflake to run analytics. Maxa already has big customers who trust this startup because they know the app is set within Snowflake’s governance framework. Lots of other people will be building apps like this in our cloud.
Building models
It’s been a busy year for Snowflake. As well as ongoing attempts to refine its AI Data Cloud, the technology firm welcomed a new CEO in February, Sridhar Ramaswamy, who previously held the position of Senior Vice-President of AI at Snowflake. Dageville says Ramaswamy’s strengths appealed to the board and the product announcements at Summit showed how the new CEO is already making a mark:
We’re releasing all these different AI services. Some of them are just about executing a large language model. But then Cortex Search runs Neeva, a world-class search engine, inside Snowflake. This deep approach to technology creates a full search capability inside Snowflake.
Snowflake bought the search company Neeva last year to boost its search capabilities through generative AI. Product demonstrations at Summit illustrated the potential power of Cortex AI, the company’s large language model (LLM) and vector search service, and two new chat capabilities, Snowflake Cortex Analyst and Snowflake Cortex Search, which allow users to build chatbots that work with structured and unstructured data. Dageville says he’s excited by the innovations that are likely to spring from these technological foundations:
Neeva was an interesting company because their team used AI-enabled technology to rank their answers. Having the power of search inside Snowflake through Cortex is amazing and the possibilities are endless.
As well as its work around Cortex AI, Snowflake recently launched Arctic, an enterprise-focused LLM. The company is certainly investing heavily in AI. However, generative AI is a competitive and fast-moving market. Dageville recognizes the challenge, especially when it’s almost impossible to know which models will dominate in the longer term:
This is a $1 billion question but I’ll try to answer it. I don’t think it’s a situation where one model will win. It’s not a one-horse race where only one animal crosses the finish line first. The best model will be a different answer, depending on what you want to do.
Developing relationships
Dageville recognizes Open AI’s ChatGPT has inherent advantages in generative AI because of the size and power of the model. Yet he says it can be expensive to run high-performance GPT models, especially if you’re working for a cash-sensitive enterprise experimenting with AI. For these executives, there are other options on the table – and that’s where he hopes models like Arctic will come to the fore:
The idea that smaller models are not as good as bigger models is not true. I use Arctic and it’s amazing. Developers can fine-tune models to perform a specific task. You might start by training a model in a very general way. But then you can take this model and push it to do something specific, like SQL code generation. You must make a trade-off between where you spend your energy and money to train models.
For companies already exploring AI use cases, Snowflake announced a new collaboration with computing giant NVIDIA at this week’s Snowflake Summit that customers and partners can harness to build customized AI applications. Arctic is supported by NVIDIA TensorRT-LLM software and Dageville expects the relationship to pay dividends:
We have a strong partnership with Nvidia. Where it makes a big impact for us is they have interesting technology we can leverage. They have GPUs that we use to create our models, such as Arctic. So, there are many benefits for our customers.
Dageville says the challenge for Snowflake going forward is to ensure it brings all these disparate capabilities together for customers in a serverless manner. He says simplicity is the key success factor for a tech company that wants to help CIOs embrace AI:
Using AI can be a daunting task. People wonder where they will find and how they will fund the GPUs. But now you can use technology from a partnership, like ours with NVIDIA, to build amazing things in minutes. And that’s because we try to provide technology for our customers that is ready to use.
