• March 17, 2026

Yesterday I appeared on CNBC’s Closing Bell alongside analyst Dylan Patel to discuss the announcements that NVIDIA CEO Jensen Huang made during his keynote at the company’s annual GTC conference in California. You can watch an edited video of this conversation via the link below, or read on for an AI-generated, human-edited overview and summary of the discussion.

Link to edited segment.

Overview

At the NVIDIA GTC Keynote, Jensen Huang announced a $1 trillion AI forecast, up from a previous $500 billion. Patrick Moorhead noted that most observers are optimistic or neutral about Huang’s projection, with not many naysayers. Dylan Patel emphasized the broad AI ecosystem beyond the five major hyperscalers. Moorhead further noted Huang’s focus on heterogeneous computing and integrated software like Groq and Dynamo Inference. Patel also discussed NVIDIA’s strategy to lock up market capacity, controlling over 60% of long-term contracts and securing $250 billion in wafer, memory, and networking equipment deals, making it challenging for competitors to scale production.

Outline

NVIDIA’s AI Market Forecast and Ecosystem Expansion

• Melissa Lee of CNBC asks for the analyst’s perspective on Jensen Huang’s $1 trillion forecast, noting that while it seems enormous, it appears less impressive when compared to the previous $500 billion forecast through 2026.
• Patrick Moorhead argues that the $1 trillion figure is safe due to the massive volume of tokens required by consumers, enterprises, and startups, though he cautions that power generation will be a critical factor by 2028.
• Dylan Patel explains that NVIDIA is moving beyond the five major hyperscalers to build a foundation of hundreds of customer companies, reaching into sovereign AI and regular enterprises.

NVIDIA’s Technical Strategy and Competitive Response

• Michael Santoli of CNBC asks whether Huang successfully persuaded the industry that NVIDIA can dominate the next phase of compute against non-GPU solutions and custom chips.
• Moorhead states that Huang delivered by pitching heterogeneous computing and a contiguous software layer that allows different chips to work together seamlessly.
• Moorhead highlights the integration of Groq and a new piece of software called Dynamo Inference that functions across various silicon architectures including GPUs, LPUs, and CPUs.

NVIDIA’s Supply Chain Dominance and Market Control

• Lee asks whether NVIDIA is actually gaining traction given the multitude of competitors, such as Amazon with Cerebras, developing their own chips.
• Patel explains that NVIDIA is attacking competitors by both innovating new architectures and simultaneously locking up global manufacturing capacity.
• Patel notes that NVIDIA has secured over 60% of market capacity through long-term contracts and is signing $250 billion of deals for wafers, memory, and networking.
• Patel asserts that even if a startup develops a better chip, it will likely be unable to manufacture them in significant volumes because NVIDIA has laid claim to so much of the upstream supply chain.
• Patel describes how NVIDIA has mapped out the entire chain from TSMC silicon production to the specific datacenters where the GPUs will eventually land.

Keyword Summary

NVIDIA, Jensen Huang, GTC keynote, Groq, hyperscalers, heterogeneous computing, Dynamo Inference, GPUs, inferencing, data centers

Recent Posts