Nvidia CEO Jensen Huang on Earnings, Demand and Blackwell Chip (Full Interview)

Summary

In this video script, the discussion revolves around NVIDIA's Blackwell, emphasizing it's a production issue rather than a design flaw. The company has made mass changes to improve yield and is gearing up for volume production in Q4, expecting billions in revenue. The demand for Blackwell is high, with supply expected to ramp up in the coming quarters. The conversation also touches on NVIDIA's diversified data center business, the growth of sovereign cloud sales, and the company's focus on energy efficiency and performance in next-gen models. There's no plan to become a cloud compute provider, with NVIDIA focusing on optimizing its cloud presence within existing providers.

Takeaways

• 📈 The market is seeking more specifics on Blackwell, indicating a keen interest in the product's details and potential impact.
• 🔍 The issue with Blackwell was identified as a production problem rather than a fundamental design flaw, suggesting that the product's underlying concept is sound.
• 🚀 Blackwell has started volume production and is expected to ship in Q4, with the company anticipating billions of dollars in Blackwell revenues, indicating a strong market demand and financial outlook.
• 🌐 Blackwell systems are being showcased worldwide, with images available online, demonstrating the company's confidence in the product and its readiness for market deployment.
• 📊 Demand for Blackwell is currently outstripping supply, a common occurrence for innovative products in their initial stages, but the company plans to ramp up supply starting Q4.
• 💡 Accelerated computing is in high demand beyond hyperscalers, with a diversified customer base including ISPs, sovereign entities, and various industries, reflecting the broad applications of the technology.
• 🌐 Sovereign cloud initiatives are gaining traction globally, with multiple countries investing in their digital infrastructure to capitalize on their digital data as a national resource.
• 💡 Energy efficiency is a key focus for NVIDIA, with Blackwell offering superior performance at the same power level as its predecessor, Hopper, and the adoption of liquid cooling to further enhance efficiency.
• 🌟 NVIDIA's GPU Cloud is designed to integrate with existing cloud providers to offer the best version of NVIDIA's cloud services, rather than becoming a cloud compute provider themselves.
• 📈 NVIDIA's supply chain is robust and expected to improve sequentially, with more supply anticipated in upcoming quarters, reflecting the company's commitment to meeting growing demand.
• 🌟 The generative AI market is diversifying, with enterprises and different countries joining the space, and NVIDIA is well-positioned to support this expanding ecosystem with its technology.

Q & A

What was the market's reaction to the information provided about Blackwell?

The market seemed to want more specifics on Blackwell, particularly regarding its tangible deployment in the real world and any potential delays in its timeline that might affect revenue.

Was the issue with Blackwell related to its design or production?

The issue with Blackwell was a production issue, not a fundamental design issue, indicating that the product's concept is sound but there are challenges in manufacturing or yield improvements.

What is the current status of Blackwell's deployment and when is it expected to ship?

Blackwell has started volume production and is expected to ship in Q4, with the company anticipating billions of dollars in Blackwell revenues from there.

What is the demand situation for Blackwell?

The demand for Blackwell far exceeds its supply, especially in the beginning, due to the high demand for the product.

How does NVIDIA plan to ramp up supply for Blackwell?

NVIDIA plans to ramp up supply starting in Q4 and continuing into Q1 and Q2 of the following year, with the aim of meeting the high demand for Blackwell.

What is the significance of 'sovereign cloud' in the context of NVIDIA's business?

Sovereign cloud refers to deals with specific governments or regional service providers funded by governments, indicating a growing trend where countries are investing in their own data infrastructure to manage and leverage their digital data as a national resource.

Which countries are currently investing in sovereign cloud infrastructure?

Countries such as Japan, India, Canada, the U.K., France, Italy, Singapore, and Malaysia are among those investing in their regional data centers to build out their sovereign cloud infrastructure.

How does NVIDIA address the increasing energy requirements for next-generation models?

NVIDIA focuses on improving the performance efficiency of their next-generation models, such as Blackwell, which is more performant than its predecessor Hopper at the same power level. They also advocate for the use of liquid cooling, which is more energy efficient.

What is NVIDIA's strategy regarding cloud computing and whether they plan to become a cloud compute provider?

NVIDIA does not plan to become a cloud compute provider. Instead, they focus on building the best version of NVIDIA Cloud within existing cloud providers like GCP, Azure, AWS, and OCI, ensuring high performance and cost-effectiveness.

How does NVIDIA use its own GPU Cloud?

NVIDIA uses its GPU Cloud for internal purposes such as designing their chips with AI, developing software, working on self-driving cars, robotics, and the Omniverse platform. They also offer AI model creation as a foundry service for other companies.

What is the current supply situation for NVIDIA's products and how do they expect it to evolve?

NVIDIA expects an improvement in supply sequentially, with each quarter expected to have more supply than the previous one. They anticipate a significant improvement in supply conditions by the next fiscal year.

How is the demand for accelerated computing beyond hyperscalers diversified?

The demand for accelerated computing is diversified across hyperscalers, internet service providers, sovereign cloud initiatives, and various industries and enterprises, all of which require accelerated computing for tasks from AI to scientific simulations.