Jensen Huang, Michael Dell on Memory Demand & China

Category: Memory & HBM · Duration: 20 min · ▶ Watch

Switch language → 中文

Segments (8)

00:00:00 · Introduction: Dell’s AI Factory Growth
- The discussion begins with Dell’s recent acquisition of 1,000 new clients for its AI server/AI factory solutions, highlighting the shift from testing to production.
01:10:00 · The Shift to On-Prem and Agentic AI
- Jensen Huang explains that intelligence must be produced at the ‘point of context,’ driving the need for on-prem AI, and introduces the concept of ‘Agentic AI’ that performs work, not just creates content.
02:32:00 · GPU Supply and Demand Dynamics
- The conversation addresses the high demand for GPUs, with Michael Dell explaining the supply chain is scaling up but demand still exceeds supply, and Jensen Huang detailing how Dell is a key channel to the enterprise market.
04:20:00 · Nvidia’s Full-Stack AI Platform
- Jensen Huang describes Nvidia’s role as a technology company providing the full stack for Agentic AI, including the ‘brain’ (GPU), the ‘harness’ (CPU), long-term memory (storage), and networking.
08:00:00 · Supply Chain Bottlenecks and Memory
- Michael Dell identifies memory and advanced-node semiconductors as key supply constraints, while Jensen Huang explains Nvidia’s long-term supply chain planning to integrate all components.
09:04:00 · The Future of AI and the Memory Market
- The speakers discuss the long-term nature of the AI buildout, dismissing the idea of a cyclical boom-and-bust for memory, and predicting a decade-long growth phase.
12:52:00 · US-China Tech Relations and Market Access
- Jensen Huang discusses his trip to China, the licensing of H200 chips, and the broader geopolitical context of the US-China tech competition and market access.
17:01:00 · The Role of the AI PC
- The conversation concludes by touching on the evolution of the PC into an ‘AI PC’ capable of running local models, driving a new upgrade cycle for more powerful devices.

Memory Facts (3)

[08:02:00] Memory is a supply chain challenge.
[08:44:00] Nvidia’s supply chain planning aligns HBM with CoWoS, Grace Blackwell, and CPUs.
[09:07:00] The memory market is historically cyclical (boom and bust).

Bottleneck Claims (4)

[02:56:00] There’s more demand than supply for AI hardware.
- Evidence: Direct statement by Michael Dell.
[08:02:00] Memory is a supply constraint.
- Evidence: Direct statement by Michael Dell.
[08:05:00] Advanced node semiconductors are still challenging.
- Evidence: Direct statement by Michael Dell.
[08:57:00] The overall capacity of the world’s supply chain is smaller than the demand.
- Evidence: Direct statement by Jensen Huang.

Predictions (4)

[12:07:00, 10+ years] The AI buildout will last for a decade, maybe more.
[11:04:00, Future] We will have hundreds of billions of AI agents.
[12:17:00, Long-term] After digital agents, the next wave will be physical agents (physical AI/robotics).
[13:50:00, Long-term] Over time, the market in China will open.

Key Technologies (10)

AI Factory: Dell’s term for a full-stack, on-premise AI infrastructure solution for enterprises.
Agentic AI: AI that can perform tasks, reason, plan, and use tools, going beyond content generation.
Grace Blackwell (GB200): Nvidia’s superchip that acts as the ‘brain’ for large language models.
NVLink 72: High-speed interconnect for linking GPUs and CPUs.
Vera CPU: Nvidia’s CPU designed to run the ‘harness’ for Agentic AI.
Dell AI Data Platform: A new type of long-term memory (storage) for AI agents, built on Nvidia technology.
Nemo Claw: Nvidia’s agent runtime software.
HBM (High Bandwidth Memory): A key memory component for AI accelerators, currently a supply constraint.
CoWoS (Chip-on-Wafer-on-Substrate): Advanced packaging technology used to integrate chips like GPUs and HBM.
Silicon Photonics: Technology for high-speed optical interconnects in data centers.

Companies Mentioned (7)

Nvidia · Dell Technologies · Eli Lilly · Samsung · Micron · SK Hynix · TSMC

Notable Quotes (5)

Intelligence has to be performed, produced, at the point of context. — Jensen Huang @ 01:33:00

Making content is very important, but doing work is really valuable. And now we’re doing productive work incredibly well. That’s why they’re called Agentic AI. — Jensen Huang @ 02:22:00

They don’t get 10 or 20 or 30% improvement, they get 10 times or 20 times or 100 times. And that is really the speed that matters to make a business successful. — Michael Dell @ 03:25:00

Dell will do for the world’s enterprises what the clouds do for the clouds. — Jensen Huang @ 06:28:00

We’re in the beginning of the AI buildout. This is literally the very beginning of the Agentic AI buildout. We’re going to be building this out for a decade, maybe more. — Jensen Huang @ 12:07:00

Key Topics

Agentic AI · AI Factories · On-Premise AI · GPU Supply and Demand · Semiconductor Supply Chain · Memory Bottlenecks · CPU Role in AI · US-China Tech Relations · AI PC · Accelerated Computing

Takeaways

The AI boom is rapidly expanding from cloud data centers to on-premise ‘AI Factories’ within enterprises, driven by the need to process proprietary data locally.
The next major wave in AI is ‘Agentic AI,’ which focuses on performing complex tasks and workflows, a significant step beyond the content generation capabilities of current models.
Demand for AI hardware, particularly GPUs and high-bandwidth memory (HBM), is far outpacing the global supply chain’s capacity, with memory identified as a primary bottleneck.
This AI infrastructure buildout is not a short-term trend; both CEOs predict it will be a sustained, decade-long cycle of investment and growth.
The partnership between Nvidia (technology provider) and Dell (solutions integrator) is crucial for delivering full-stack AI solutions to the vast enterprise market.
The role of the CPU is being redefined in the AI era to act as a ‘harness’ for the GPU ‘brain,’ orchestrating tasks for AI agents.
The future of AI will involve trillions of digital and physical agents, creating an unprecedented demand for computing power across all industries.