HUANG LAW and SUMMARY OF HIS KEYNOTE ADDRESS
Here are some detailed notes
and conceptual analysis of Jensen Huang's keynote address at the NVIDIA GTC in
Washington, D.C., on October 28, 2025.
The central business announcement, which
subsequently propelled NVIDIA to a $5 trillion market capitalization, was
Huang's statement that the company now has "visibility into half a
trillion dollars" ($500 billion) in cumulative orders for its
Blackwell and upcoming Rubin platforms through 2026.
Oh, wait… GTC stands
for: GPU Technology Conference in the next Industrial Revolution. Jensen Huang is the High Priest of this
conference. The conference is a major platform for unveiling new Nvidia
technologies and setting the direction for future advancements in AI and
computing.
Detailed Summary of GTC
Keynote (October 28, 2025)
The keynote, themed
around "America's AI Infrastructure" and the "Next Industrial
Revolution." It was a strategic presentation focused on NVIDIA's role in
building national AI capabilities.
The central business
announcement, which subsequently propelled NVIDIA to a $5 trillion market
capitalization, was Huang's statement that the company now has "visibility
into half a trillion dollars" ($500 billion) in cumulative orders for
its Blackwell and upcoming Rubin platforms through 2026.
Here are the key
announcements by category:
1. Next-Generation
Platform Roadmap (The "One-Year Rhythm")
Huang reaffirmed
NVIDIA's aggressive one-year release beat, moving beyond the chip to
full-platform co-design.
- Blackwell ( - Current Generation):
The Grace Blackwell platform (e.g., GB200 NVL72) is in full production at
facilities in the U.S. (Arizona), reinforcing the "Made in
America" theme. The Blackwell Ultra chip will be released later in 2025. (Not much time
left).
- Vera Rubin ( - Next Generation):
The next major platform, named after the astrophysicist Dr.
Vera Rubin, is scheduled for 2026. This is not just a GPU but it
is an entire system architecture.
(Vera Rubin was an American astronomer
whose work provided convincing evidence for the existence of dark matter.)
- The Full Roadmap:
- 2025 (2H):
Blackwell Ultra
- 2026 (2H):
Vera Rubin (including the Vera Rubin Superchip, CPX Compute Tray, and
BlueField-4 DPU)
- 2027 (2H):
Rubin Ultra
2. U. S. National AI
Infrastructure & Supercomputing
This is a core theme,
positioning NVIDIA as a national strategic asset.
- U.S. Department of Energy (DOE)
Partnership: NVIDIA announced it is powering
seven new AI supercomputers for the DOE.
- "Solstice" Supercomputer:
The largest of these, built in partnership with Oracle, will be one
of the world's most powerful AI systems, featuring 100,000
NVIDIA Blackwell GPUs to support national security, energy, and
science applications.
- Los Alamos National Lab (LANL):
The LANL's next-generation systems
will be among the first to be built on the upcoming Vera Rubin platform.
3. New Frontiers: 6G
and Quantum Computing
Huang detailed NVIDIA's
expansion into two new, highly complex compute domains.
- 6G Telecommunications:
- Hello!
- Nokia Partnership: A $1 billion strategic
partnership with Nokia to develop an "AI-native 6G" platform.
- NVIDIA Arc Aerial RAN Computer:
A new, 6G-ready computing platform designed to infuse AI services
directly into the mobile network. (For me and you…)
- "All-American AI-RAN
Stack": A collaboration with T-Mobile,
Cisco, and MITRE to build a U.S.-based 6G development stack.
- Quantum Computing:
- NVIDIA NVQLink:
A new interconnect architecture designed to bridge the gap between
classical and quantum computing. It allows NVIDIA GPUs to be directly
linked to Quantum Processing Units (QPUs), enabling a hybrid
quantum-classical system for complex simulations.
4. Physical AI
(Robotics & Autonomous Systems)
Huang declared
"Physical AI" as the next major wave, where AI agents perceive and interact
with the physical world.
- NVIDIA "Groot N1"
Foundation Model: A new, general-purpose
foundation model for humanoid robots.
- Newton Simulation Platform:
A new high-fidelity physics engine (an evolution of Omniverse) designed to
simulate robots and their environments for training.
- "Project Blue":
A collaborative humanoid robot project demonstrated with partners Google
DeepMind and Disney Research.
- Autonomous Vehicles:
- Uber Partnership:
A major partnership to deploy 100,000 autonomous robotaxis, powered by
NVIDIA DRIVE, starting in 2027.
- DRIVE Hyperion Platform:
Expanded adoption by automakers including GM, Stellantis, Lucid, and Mercedes-Benz.
5. Geopolitical
and Market Context
Hey, there is money to be here…
- $5 Trillion Valuation:
The keynote's $500B order visibility statement was the primary catalyst
for NVIDIA's stock surge, making it the first company to achieve this
valuation.
- China Market:
NVIDIA's market share in China has fallen from 95% to "effectively
zero" due to U.S. export controls and Beijing's policies.
- "America First"
Alignment! Huang explicitly praised the Trump
administration's "America First" policies for incentivizing and
revitalizing U.S. manufacturing, which he stated enabled NVIDIA's new
U.S.-based production.
Here are PhD-Level
Lecture Notes and Conceptual Analysis of All That Said
Below are the core
theoretical theses presented by Huang, abstracted from the product
announcements.
Thesis 1: The End of
Classical Scaling Paradigms
- Concept:
The death of Moore's Law and, more importantly, Dennard Scaling
(which stated power density remains constant as transistors get smaller)
is now an accepted industry fact.
- Argument:
Sequential processing (CPU-centric) can no longer deliver the performance
gains required. The only path forward is Accelerated Computing, a
hybrid model where parallel processors (GPUs) work in tandem with
sequential processors (CPUs).
- Evidence:
The entire keynote was a demonstration of this thesis. The core software
foundation, CUDA-X,
is the "operating system" for this new computing model, and
every new hardware platform is designed to accelerate this specific
paradigm. (you need to follow the link if you want to get an idea what is
CUDA)
Thesis 2: The New
Computing Model: "Generative" vs. "Retrieval"
- Concept:
Huang articulated a fundamental shift in the purpose of computing.
- Retrieval Computing ( - The Past):
The old internet and all prior computing were based on retrieval.
A user requests information, and the system fetches a pre-written,
pre-stored piece of data (a webpage, a document, a video).
- Generative Computing ( - The
Future): The novel AI models do not
retrieve. They receive a prompt, understand
the context and meaning,
and generate a novel, never-before-seen answer (a
"token"). (Hello
Agentic…)
- Financial Implication:
This new model is computationally far more expensive and requires a new
infrastructure. The basic unit of this new infrastructure is the "AI Factory."
Thesis 3: The New
Scaling Law: "Extreme Co-Design"
- A new Concept
is born: If single-chip performance
(Moore's Law) is no longer the primary driver, gains must come from a new
scaling law.
Huang's Law
is: "Extreme Co-Design."
- Argument:
Performance "X-factors" (multiplicative gains) are now achieved
by co-designing the entire stack as a single product. This
includes:
1. Silicon:
The GPU and CPU (Grace Blackwell).
2. Interconnects:
High-speed chip-to-chip links (NVLink).
(NVLink
means providing high speed connectivity between two GPUs to increase
performance).
3. Networking:
The data center fabric (Spectrum-X Ethernet).
4. Power
& Cooling: Liquid-cooling and power delivery
systems.
5. Software:
Optimized libraries (CUDA) and inference engines (NVIDIA Dynamo).
- Evidence:
The Grace Blackwell NVL72 is the canonical example. It's not sold
as 72 separate GPUs but as a single, liquid-cooled "thinking
machine," a single computational unit. The Vera Rubin platform
continues this by integrating the BlueField-4 DPU directly into the system
design.
Thesis 4: The Next
Wave: From Digital AI to "Agentic & Physical AI"
- Concept:
Huang defined the next evolution of AI.
- Agentic AI:
Ha… It’s the AI that possesses all
that ai agency.
Can do. BUT it can perceive its environment,
understand the context, reason, create a plan, and act to
accomplish a goal. You can find some more on Agentic here.
- Physical AI:
The application of Agentic AI to the physical world, which is robotics.
- Argument:
To create Physical AI, models must be trained to understand physics, 3D
space, and cause-and-effect.
- Evidence:
This thesis justifies the new product stack:
- Groot N1:
The "brain" or foundation model for the robot.
- Newton:
The "gym" or virtual world where the brain is trained (via
simulation and reinforcement learning) before being deployed in the
physical world.
Thesis 5: The
Software-Defined Physical Stack (6G & Quantum)
- Concept:
NVIDIA's strategy is to turn
specialized, hardware-defined industries into software-defined ones
running on NVIDIA GPUs.
- Argument (6G):
A 5G/6G base station is currently a complex box of fixed-function hardware
(ASICs, FPGAs). The NVIDIA Arc
platform turns it into a software-defined radio (SDR) running on a GPU.
This allows telcos to push AI services (like AI-RAN) to the network edge as
a software update.
- Argument (Quantum):
Quantum computers (QPUs) are brilliant at certain problems but useless at
others. The NVQLink interconnect treats the QPU as a "quantum
accelerator" in the same way a GPU is a "parallel
accelerator," allowing developers to write hybrid algorithms (within
CUDA) that pass workloads between the CPU, GPU, and QPU.
Subscribe
to This Blog. It’s Free.
A video from NVIDIA's
YouTube channel provides the full keynote address from GTC in Washington, D.C.,
where these notes were made.
My
Take:
If you understand Huang's Law of "Extreme
Co-Design," and you understand where Agentic AI is going, you get a
sense on how close we are to AGI.
www.mandylender.net www.attractome.com www.lendercombinations.com www.mandylender.com
Tags: #AI #GTC #NVIDIA #HuangLaw
#extremecodesign #agentic #JensenHuang #Blackwell #VeraRubin #CUDA
No comments:
Post a Comment