Nvidia CEO Jensen Huang on AI Demand, Supply Chain, and China Market

Nvidia's Strong Performance and Forecast

• 🚀 Nvidia reported third-quarter results exceeding estimates and provided a robust revenue forecast of about $65 billion for the current period, signaling sustained demand for its AI accelerators.
• 💡 The company's gross margins are projected to reach the mid-70% range, underscoring the profitability of its high-demand chips.
• 📈 CEO Jensen Huang stated that sales for Blackwell GPUs in the cloud are sold out, with ample supply planned for both Blackwell and the upcoming Vera Rubin.

Vera Rubin and Rack Architecture

• 🧠 The silicon for Vera Rubin is in labs, with bring-up underway by approximately 20,000 engineers, on track for a Q3 next year delivery.
• 🛠️ The rack architecture, including the MVLink 72 switch, is described as revolutionary and will seamlessly transition from Grace Blackwell Ultra to Vera Rubin.
• ⚡ This integrated architecture ensures smooth ramp-up and efficient supply chain utilization.

China Market and US Regulations

• 🇨🇳 Nvidia's forecast for the China market is currently zero, despite its importance.
• 🤝 The company aims to re-engage the Chinese market with competitive products and is actively engaging with both US and Chinese governments to allow participation in the open market.
• ⚠️ Regarding US government requirements, Nvidia is committed to preventing tech transfer to China through the Middle East, implementing rigorous measures to ensure no diversion of chips.

Energy and Compute Constraints

• ⚡ While chip supply is strong, energy is identified as a significant constraint for AI expansion, with power generation and land/shell providers needing to keep pace with Nvidia's rapid growth.
• 🌍 Nvidia leverages its vast network of cloud service providers and OEMs to find available power sources globally, a key advantage in deploying its technology.
• 💰 The company's architecture is essential for all AI models, including those from OpenAI, Anthropic, and Google's Gemini, ensuring high off-take regardless of the cloud provider.

Investment and Depreciation

• 💡 Investors may be underestimating the longevity of Nvidia's chips due to the versatility of its architecture and the continuous value added through CUDA software updates.
• 📈 Even older chips like the A100, shipped six years ago, remain significantly faster than CPUs for many cloud workloads.
• 📊 For a 1-gigawatt data center, Nvidia's contribution for Blackwell is around $30 billion, and for Vera Rubin, it's approximately $35 billion out of a $50-55 billion total.