What’s Snowflake’s AI Model? Here’s Everything To Know

Snowflake (NYSE: SNOW) has had a rough few months, with its stock price dropping around 15%. The reasons? Projected slowing revenue growth, CEO transition uncertainties, and concerns about its valuation.

In response to this sharp decline, Snowflake’s new CEO, Sridhar Ramaswamy, is steering the company towards becoming an “AI data cloud” leader. The company is doubling down on AI with strategic moves like acquiring Neeva, launching its own large language model (LLM) called Snowflake Arctic, and forming partnerships with Nvidia.

This article will explore Snowflake’s AI strategy to reignite growth, its latest AI products, and the best alternatives like Voiceflow.

What is Snowflake?

Snowflake is a cloud data platform company that enables companies to store, process, analyze, and share data across multiple clouds like AWS, Azure, and GCP. Its mission is to provide an integrated platform that eliminates data silos to help organizations get more value from their data.

Founded in 2012, Snowflake went public in 2020 with a valuation of $33 billion. Since then, it has gained over 7,500 customers and processes over 1 billion queries per day on average.

Snowflake Database

Rather than traditional database hosting services like AWS RDS or Azure SQL, or transactional database systems like Oracle, Snowflake is optimized for large-scale data analytics, data engineering, and data science workloads. Snowflake’s key features include separation of storage/compute, massively parallel processing, semi-structured data support, data cloning, time travel for data, and third-party tools integration.

Snowflake’s Generative AI Strategy

Snowflake’s new CEO, Ramaswamy, told Fortune that AI now pervades “everything that is happening in Snowflake.” Indeed, the company is making significant investments in AI through acquisitions, product launches like Arctic LLM, AI-powered data management tools, and strategic partnerships — all aimed at positioning the Snowflake AI Data Cloud as a platform for enterprise AI solutions.

• Snowflake’s Acquisitions: Snowflake acquired Neeva, a search company that uses generative AI to enhance data search. It also acquired Streamlit, a platform for building AI/LLM-powered apps, and Applica, which uses deep learning for information sorting across data types.
• Launch of Snowflake Arctic LLM: Snowflake launched its own enterprise-grade large language model integrated into the Snowflake platform to provide chat capabilities with data.
• Launch of Cortex Analyst and Cortex Search: As enhancements to Snowflake Cortex AI, Cortex Analyst, and Cortex Search allow businesses to query analytical data and build AI-powered agents.
• Partnerships and Integrations: Snowflake is partnering with AI chip makers like Nvidia to accelerate enterprise AI initiatives.

{{button}}

What’s Snowflake’s Arctic LLM?

Ramaswamy said that Arctic is a “foundational investment” and Snowflake’s “first big step in generative AI”.

Arctic is an open-source large language model developed by Snowflake for enterprise AI use cases, designed to compete with existing LLMs like OpenAI’s GPT-4 and Google’s Gemini. It’s currently available on Hugging Face, NVIDIA AI Catalogue, Replicate, model hubs and cloud platforms like AWS and Azure, and its own Snowflake Cortex platform for enterprise customers.

Arctic uses a Dense-MoE (Mixture of Experts) Hybrid transformer architecture, combining a 10B parameter dense transformer model with a 128x3.66B parameter Mixture of Experts (MoE) Multi-Layer Perceptron (MLP). This results in a total of 480B parameters, with 17B active parameters chosen using top-2 gating during inference.

So, how does Arctic compare to its competitors like LLaMA? On benchmarks like Spider (SQL), HumanEval+ and MBPP+ (coding), and IFEval (instruction following), Arctic outperforms or matches other open-source LLMs like LLaMA while using significantly less compute for training. Indeed, it achieves top performance on “enterprise intelligence” metrics while being cost-effective, using under $2 million in training compute.

What’s Snowflake’s Cortex AI?

Snowflake Cortex AI is a fully managed suite of AI features and serves, aimed to democratize AI by enabling users of all skill levels to use AI models to build AI applications. Here are some key features:

• LLM Functions: Cortex AI provides access to LLMs like Snowflake’s own Arctic model, as well as other task-specific models, foundation models, and fine-tuned models. In June 2024, Snowflake announced that you can now fine-tune models for specific use cases via a no-code interactive UI using the Snowflake AI & ML Studio for LLMs.
• ML Functions: Provide time-series forecasting, anomaly detection, clustering, and other machine learning tasks.
• Snowflake Document AI: Extract insights from documents using LLMs and hybrid search (semantic search and keyword search).
• Universal Search: AI-powered enterprise search across structured and unstructured data sources.
• Snowflake Copilot: AI-assisted SQL query building and data exploration.
• Cortex Analyst (public preview soon): Build chatbots on analytical data using Meta LLaMA and other models.
• Cortex Search (public preview soon): Chatbots on documents/text using hybrid search and Snowflake Arctic embed.
• Snowflake Cortex Guard (generally available soon): Use an LLM-based safeguard to filter harmful content.

‍

Snowflake’s Pricing Model

Snowflake’s pricing is based on actual usage across three layers: storage, virtual warehouses (compute), and cloud services. Storage is charged monthly per terabyte, with costs varying by region. Compute costs are based on the size of the virtual warehouse, billed per second with a one-minute minimum. Cloud services are typically free up to 10% of daily compute usage. Snowflake offers both on-demand and pre-purchased capacity plans, with credits used to measure resource consumption.

Snowflake also offers a free 30-day $400 credit trial for customers to test out the platform before purchasing.

Snowflake’s Competitors

Snowflake’s competitors offer similar data warehousing and analytics solutions, here’s an overview:

{{black-cta}}

Snowflake vs. Databricks

• Choose Snowflake if your primary needs are SQL-based analytics and seamless cross-cloud integration without the need for deep data science or machine learning capabilities.
• Choose Databricks if your organization needs powerful data processing capabilities and collaborative tools for data scientists and engineers working on complex machine learning projects.

Snowflake vs. AWS

• Choose Snowflake if you require a multi-cloud strategy and value ease of use and automated performance optimization.
• Choose Amazon Redshift if your organization is already heavily invested in AWS services and you need a customizable, high-performance data warehouse.

Snowflake vs. BigQuery

• Choose Snowflake if you prioritize multi-cloud deployment and automated performance optimization.
• Choose BigQuery if your organization is embedded in the Google Cloud ecosystem and requires a cost-efficient, serverless solution for processing and analyzing vast amounts of data.

How to Build an Enterprise-Grade AI Agent the Easy Way

The new Cortex Analyst lets you interact with your data conversationally, while Cortex Search helps create RAG-based chatbots. However, these tools aren’t publicly available yet.

For a free alternative to build RAG- and knowledge-base-powered AI agents, try Voiceflow! Trusted by over 250,000 businesses, including LVMH and Home Depot, Voiceflow is the leading platform for teams to design, prototype, and launch AI assistants quickly and easily. Get started today!

{{button}}