# OpenAI x Broadcom — The OpenAI Podcast Ep. 8

Source: https://www.youtube.com/watch?v=qqAbVTFnfk8
Recap page: https://rapidrecap.app/video/qqAbVTFnfk8
Generated: 2025-10-13T13:34:52.982+00:00

---
## Quick Overview

OpenAI and Broadcom announced a major partnership involving the design of a new custom chip and a full system approach, culminating in the planned deployment of 10 gigawatts of computing infrastructure starting late next year to serve global advanced intelligence needs.

**Key Points:**
- OpenAI and Broadcom are partnering to design a new custom chip specific for OpenAI's workloads, which evolved into designing a whole custom system over 18 months of collaboration.
- The partnership plans to deploy 10 gigawatts of these new systems and chips starting in late next year to serve the world's need for advanced intelligence.
- Sam Altman stated that optimizing across the entire stack, from transistors to the final token output, allows for huge efficiency gains, leading to faster and cheaper models.
- Greg Brockman noted that OpenAI applied its own models to chip design, achieving optimizations that human designers would have taken much longer to reach, accelerating the schedule.
- Hock Tan characterized the collaboration as natural, positioning the resulting infrastructure as critical utility over time for 8 billion people, like the industrial revolution of a different sort.
- The scale of the AI infrastructure build-out is described by Sam Altman as potentially the biggest joint industrial project in human history.
- Charlie Kawwas detailed the technical progression from custom AI accelerators (XPUs) to integrating networking and ultimately incorporating optics for 100 terabits of switching integrated into the same chip.

**Context:** The OpenAI podcast episode featured Andrew Mayne hosting Sam Altman and Greg Brockman from OpenAI, alongside Hock Tan and Charlie Kawwas from Broadcom, to announce a significant strategic partnership focused on scaling AI compute infrastructure. The discussion centered on the necessity of vertically integrating the design process—from transistor-level chip design to system architecture—to meet the exponentially growing demand for advanced AI models like GPT-6 and future iterations.

## Detailed Analysis

The core of the announcement is a deep collaboration between OpenAI and Broadcom, spanning 18 months, which initially focused on creating a custom chip optimized for OpenAI's specific AI workloads but expanded into designing an entire custom system stack. This vertical integration allows optimization across all layers, promising significant efficiency gains, resulting in faster and cheaper models, which historically leads to an even larger surge in demand, as seen when optimization is achieved. The immediate deployment goal is massive: starting late next year, OpenAI plans to deploy 10 gigawatts of this new computing infrastructure, which is described as a gigantic amount, though Sam Altman noted it is still a 'drop in the bucket' compared to where the world needs to go to achieve pervasive AI agents for all 10 billion humans. Greg Brockman highlighted that OpenAI used its own models to aid in chip design, pulling the schedule forward by finding optimizations human engineers might have missed or taken longer to find. Technically, Charlie Kawwas explained the evolution from custom XPUs to incorporating advanced networking and integrating optics to achieve 100 terabits of switching within the same chip, projecting that compute performance will continue doubling every six to 12 months. Both companies view this infrastructure build-out, exemplified by the 10 gigawatt scale, as analogous to historical critical infrastructure projects like railroads or the internet, aiming to create compute abundance so that intelligence becomes the fundamental driver of economic growth accessible to everyone.

### Partnership Scope

- Designing a new custom chip for specific AI workloads
- Evolved into designing a whole custom system
- Collaboration has been ongoing for about the last 18 months

### Infrastructure Deployment

- Plans to deploy 10 gigawatts of these systems starting late next year
- This is incremental capacity on top of existing partnerships
- Sam Altman views the AI infrastructure build-out as the biggest joint industrial project in human history

### Technical Optimization

- Optimizing across the entire stack yields huge efficiency gains
- Goal is to wring out more intelligence per unit of energy (watt)
- Customization allows for optimization tailored to training (TFLOPs, network) vs. inference (memory access)

### Chip Design Process

- OpenAI applied its own models to chip design, speeding up iterations
- They are moving from 2D chip layouts to stacking components in the Z dimension
- Integration of optics for 100 terabits of switching is part of the roadmap

### Historical Context and Vision

- Greg Brockman noted that scaling proved critical around 2017, shifting focus from pure ideas to compute
- Hock Tan compares the effort to the railroad or internet, making AI a critical utility for 8 billion people globally
- The ultimate goal is to realize the vision of AGI benefiting all of humanity through compute abundance