Auto-translation used

How hardware develops for AI

Sammari gave several recent speeches by NVIDIA CEO Jensen Huang, who presented new products and the philosophy of the technological shift.

1. AI is rapidly moving beyond just a tool.

It is becoming the basis of digital production, a new computing infrastructure on which the next wave of the industrial revolution is being built.

2. Blackwell

Blackwell is NVIDIA's new flagship chip, created as the core infrastructure for large-scale AI systems. It provides training and operation of models with extreme computing and energy efficiency requirements. It is designed for in-depth analysis, forecasting and decision-making.In one rack with Blackwell: 600,000 components, 1.36 tons of weight, 120 kW of power consumption and 1 quintillion operations per second (!).Blackwell's performance:– Performance: 2.5 times higher than Hopper's– Energy efficiency: 25 times higher (!)– Bandwidth: up to 1.8 billion tokens per Secondblackwell is optimized for AI with a large contextual window and complex logic. It is a computing core for next-generation complex processes.

3. Agentic AI

Agentic AI models that set goals themselves, form plans, use external tools, and adapt in the process. It is an AI capable of thinking and acting as an agent.It integrates into business processes, distributed systems, and standalone modules. The new Nemo + NIMS stack allows you to develop and implement such models for specific tasks.

4. Physical AI

AI goes beyond the interface and the screen — it learns to interact with the real world.

Omniverse is a simulation platform that trains AI in digital doubles, taking into account the physics of processes, the logic of movements and the environment.

Jetson AGX is a hardware platform for AI in robots and manufacturing.

- Groot, Newton, Cosmos are elements of an ecosystem where AI can see, analyze and act in real time.This technology stack forms the basis for autonomous factories, production, logistics and service solutions.

5. Knowledge Factories

The new architecture of AI models transforms data centers into cognitive factories. They not only store and process data, they create meaning, manage knowledge, and ensure continuous decision generation, where each token is an element of reflection.

NVIDIA Dynamo is a new operating system designed to manage generative cycles and phases of thinking.

6. Evolution of hardware architecture

A roadmap for product development is presented:

2025 — Blackwell

2026 — Blackwell Ultra

2027 — Vera Rubin

2028 — Rubin Ultra

Each new generation significantly increases cognitive capacity, reduces energy consumption, and allows AI to scale deeper into reasoning, agency, and interaction with reality.

7. The transition to a new computing plane

AI does not evolve linearly, but in leaps and bounds — along with the growth of models, data volumes, and architectural complexity.We are already in the transition phase: along with following instructions, the AI is already beginning to formulate new assumptions.

And along with processing and servicing requests, he can already design response systems and is able to initiate actions.

To summarize, we are already approaching the moment of the birth of architectural platforms, in which deep reasoning becomes a simple computational unit, and each token is an element of the architecture of a new world. Along with the evolution of "intelligence", a new engineering logic is being formed: deep machine thinking is becoming a simple computational procedure — designed, scalable, reproducible. Each act of reasoning is rolled up into a token and woven into the flow of a billion of the same tokens. And when you start visualizing the scale of such a system, you get a little dizzy.

We are on the verge of a genuine revolution.

Comments 0

Login to leave a comment