Sarah Guo’s Introduction:
Sarah Guo, former Greylock General Partner and founder of Conviction, introduces the session. Generative AI has taken the world by storm, and data is crucial for enabling it. Frank Slitman and Jensen Huang, CEOs of Snowflake and NVIDIA, will discuss the potential of generative AI and data.

Jensen Huang’s Perspective:
Generative AI represents the biggest computing revolution. It’s the first time computers can write software for themselves and other computers. It’s easy to use, understands user intent, and exceeds expectations. It’s a universal translator with vast implications.

Frank Slootman’s Perspective:
Enterprises have a long, challenging history with data. COBOL was not truly common or business-oriented. Structured data in the ’80s helped, but it was still limited. Generative AI represents a significant step forward, offering natural, productive, and insightful interactions with data.

NVIDIA and Snowflake Partnership Expansion:
NVIDIA and Snowflake announce a major partnership expansion. Customers can now leverage the NVIDIA AI platform on Snowflake. This enables the development of custom generative AI models using their own proprietary data.

Historic Collaboration:
Snowflake and NVIDIA’s alliance is hailed as a significant event due to the companies’ shared values and respective strengths in data, AI algorithms, and compute engines.

New Frontier of Large Language Models:
The combined effort focuses on integrating large language models with Snowflake, enabling users to directly converse with their data and obtain insights through natural language queries.

Seamless AI Integration:
The partnership aims to empower customers to seamlessly develop AI applications using their proprietary data.

AI’s Transformative Impact:
Frank Slootman emphasizes the revolutionary potential of AI in helping businesses redefine their economics, leading to more agile and efficient operations.

Opportunities for Cost Restructuring:
By harnessing AI, companies can reevaluate and restructure their cost structures, particularly in areas like customer support and sales organizations.

Software Evolution:
Software is transitioning from deterministic code (Software 1.0) to optimizing neural networks (Software 2.0), and now to leveraging foundation models with enterprise data (Software 3.0).

Foundation Models and Enterprise Data:
Foundation models are incredibly capable but need to work with enterprise data and custom datasets for real-world applications. Foundation models offer a cost-effective solution for developing applications, acting as powerful query engines on top of important data.

General vs. Custom Models:
Some believe foundation models can replace custom models due to their versatility, but this approach might be computationally expensive. In a business context, enterprises seek AI copilot applications with deep insights into specific, complex datasets, such as business models and dynamics.

Example of Enterprise-Specific AI Applications:
Sarah Guo emphasizes the need for more examples to illustrate how enterprises can leverage AI applications beyond the general knowledge of the Internet.

Churn analysis: Complex and crucial for businesses. Understanding customer churn reasons requires analyzing extensive data to identify patterns and implement effective solutions. Deployability of AI services: NVIDIA NEMO stack running inside Snowflake demonstrates the practical application of AI integration. This highlights the feasibility and benefits of AI deployment within Snowflake. Data and AI factories: Enterprises can leverage Snowflake as a data repository and bank, enabling them to operate AI factories within the platform. This allows for efficient data processing, model training, and AI operation, transforming data into valuable intelligence. GPU computing costs: While NVIDIA products offer exceptional value, GPU computing involves costs associated with usage. Organizations need to adopt consumption models and establish disciplined practices to optimize resource utilization and justify expenses. Pre-trained models: Utilizing pre-trained models from NVIDIA’s AI factories reduces training costs significantly. These models offer advanced capabilities and serve as a foundation for further customization and fine-tuning, allowing enterprises to adapt AI applications to their specific needs. Rapid AI application development: The integration of pre-trained models enables faster development of AI applications, reducing the timeline from months to days. This ускоряет innovation and productivity, empowering organizations to quickly create customized AI solutions. Foundation models in software 3.0: The emergence of foundation models in software 3.0 signifies a paradigm shift, where a significant portion of training costs is borne by others. This presents a unique opportunity for enterprises to leverage pre-trained models and build AI applications with reduced investment, accelerating innovation and business value realization.

Introduction to Evolving Compute Paradigms
Technological advancements are reshaping the compute landscape, enabling new and innovative models. Various organizations are exploring different architectures for deploying large language models (LLMs) and integrating them into their systems. Microsoft offers an open AI version running on Azure, enabling interaction with Azure customers.

Exploring the Future of Compute Architectures
There’s no clear consensus on the dominant architecture for deploying LLMs, leading to exploration of multiple paths. The market will eventually determine trade-offs between usability, cost, and practicality. Current applications may only scratch the surface of LLM potential, and more development is needed for end-state production-scale deployments. Security considerations, including copyright and data usage, will become increasingly important as LLM applications grow.

A Paradigm Shift in Computing
The industry is experiencing a fundamental computing platform shift, the first in 60 years. Central processing units (CPUs) have reached their scaling limits, leading to a paradigm shift toward accelerated computing. Software has also undergone significant changes in its operation and capabilities, contributing to this shift.

The Rise of Accelerated Computing
NVIDIA has pioneered acceleration, recognizing the need for a different approach beyond CPUs. Accelerated computing offers significant performance advantages, enabling faster throughput and cost-effectiveness. Cloud providers are adopting NVIDIA GPUs as the most expensive compute instances due to their unbeatable speed and efficiency.

Reimagining the Compute Stack
The entire compute stack is being reinvented, including processors, operating systems, and LLMs. Software 3.0 enables new application development models where users can connect prompts, commands, and data to LLMs. This paradigm shift empowers individuals to become application developers, while data remains under their control and fine-tuning is essential.

Dismantling Conventional Wisdom
In the context of cloud computing, faster computing can also be more cost-effective, challenging traditional cost-saving strategies. The shift from data moving to the work to work moving to the data presents an opportunity to break down silos and facilitate AI factories. Snowflake and NVIDIA are addressing these challenges by bringing work to the data, enabling seamless integration and efficient AI operations.

Most Quickly Realized Value of AI for Customers:
Augmented queries will yield the most immediate benefits, democratizing data accessibility and understanding, enabling users to seamlessly interact with data using simple questions, transforming dashboards into dynamic tools for data exploration.

Phase Two: Tackling Hard Questions with AI
Once the low-hanging fruit is addressed, AI will move on to more complex challenges, such as analyzing a combination of structured and unstructured proprietary enterprise data to address specific business issues.

Example of a Hard Question in Phase Two: Supply Chain Management
Optimizing complex supply chains in the face of disruptions like consumer churning requires AI’s ability to ask and answer intricate questions, considering multiple entities and factors to reconfigure and rejigger the supply chain effectively.

Impact of AI in the Next Five Years:
CEOs of large institutions recognize the true potential of AI in redefining business models, optimizing pricing, and making data-driven decisions across various industries.

Identifying the Most Valuable Database for AI Application:
Enterprises should identify their most valuable database and consider how a “super smart person” interacting with it could yield valuable insights, leading to transformative applications of AI.

NVIDIA’s Experience with AI for Internal Data
NVIDIA leverages AI for its own data, including its customer database, optimizing compilers, and bug database, enabling rapid innovation and efficiency gains.

AI’s Role in Maintaining Software Complexity:
AI plays a crucial role in maintaining software complexity, ensuring security patching, maintaining compatibility, and supporting backward compatibility, addressing security concerns, and ensuring smooth functioning of complex software ecosystems.

AI for Answering Complex Questions:
AI, especially large language models, can answer complex questions about data, revealing exposures that engineers can then fix. AI can recommend fixes, which humans can then evaluate and implement.

Improving Data Understanding Across the Organization:
AI enables more people to ask questions and understand data, leading to better decision-making.

Changing Business Margins and Expanding Customer Scope:
AI can help businesses change their margin structure and expand their offerings to customers.

AI for Complex Engineering Problems:
NVIDIA uses AI to solve complex engineering problems that are beyond human capabilities. AI’s ability to traverse multiple layers of data makes it valuable in these situations.

Identifying the Most Valuable Data:
Businesses should identify their most valuable data and use AI to understand and leverage it effectively.

Examples from Auto Loans and Auto Insurance:
In auto loans, AI helps price loans accurately for individual customers, leading to increased profits. In auto insurance, AI uses telemetry data to price risk, determining claims and driving profits.

Snowflake and NVIDIA’s Superiority:
Frank Slootman, Snowflake’s CEO, confidently stated that their combined offering is the best for ML workloads, emphasizing their commitment to delivering exceptional outcomes for their customers.

A New Era of Data Warehousing:
Frank Slootman highlighted the remarkable evolution of data warehousing, from begging for 2.30 a.m. time slots to the current era where intelligent applications can be shipped within days or weeks.

Unlocking the Potential of Data:
Jensen Huang, NVIDIA’s CEO, emphasized the vast potential of data to provide valuable insights and drive meaningful impact, which Snowflake and NVIDIA aim to unlock.

Snowflake as the Safest Data Storage:
Jensen Huang emphasized Snowflake’s position as the safest and best place to store and process data, making it an ideal choice for businesses.

Transforming Data into Intelligence:
NVIDIA’s AI engine, when connected to Snowflake’s data warehousing capabilities, will enable customers to transform their data into intelligence continuously, even while they sleep.

A Valuable Partnership for Snowflake Customers:
Jensen Huang assured Snowflake customers that the partnership with NVIDIA will help them unlock the full potential of their data by turning it into valuable intelligence.