Date: October 13, 2025
By Thorsten Meyer | ThorstenMeyerAI.com

OpenAI and Broadcom have unveiled a historic collaboration to deploy 10 gigawatts (GW) of OpenAI-designed AI accelerators, signaling a decisive shift toward vertical integration in AI infrastructure. The multi-year project will roll out between 2026 and 2029, positioning Broadcom as OpenAI’s strategic manufacturing and connectivity partner.

The End of Dependence on Nvidia

This collaboration represents OpenAI’s boldest step yet to diversify away from Nvidia’s dominant GPU ecosystem. By co-designing accelerators optimized for its own models, OpenAI gains both cost control and architectural freedom.
The accelerators will embed neural sparsity, memory optimization, and distributed inference orchestration at the hardware level—features designed to maximize efficiency for GPT-6 and future multimodal models.

10 GW of Compute Power

At 10 GW, this infrastructure rivals the power consumption of a mid-sized nation. OpenAI plans to deploy the systems in data centers across North America and Europe, in line with its ongoing energy partnerships with Sur Energy and ExaGrid Renewables.

Broadcom’s Strategic Role

Broadcom will not only fabricate custom silicon but also integrate advanced interconnects (PCIe Gen7, 1.6Tb optical links) and rack-level cooling systems to handle unprecedented thermal loads. This could make Broadcom the connective tissue of OpenAI’s compute future.

A Shift in the AI Supply Chain

If successful, this collaboration could mark the start of a post-GPU era—where large AI labs design, optimize, and deploy their own chips at scale. Nvidia may remain a leader, but OpenAI is clearly aiming to own the stack, from architecture to execution.

AI Hardware Engineering: Designing GPUs, TPUs, and Neural Processing Units for High-Throughput Machine Learning Workloads (AI Infrastructure, Hardware & Compiler Engineering Series)

AI Hardware Engineering: Designing GPUs, TPUs, and Neural Processing Units for High-Throughput Machine Learning Workloads (AI Infrastructure, Hardware & Compiler Engineering Series)

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Supermicro 4U Active CPU Heat Sink Socket OLGA4094 -3800 rpm - Ball Bearing

Supermicro 4U Active CPU Heat Sink Socket OLGA4094 -3800 rpm – Ball Bearing

  • Product Model: Supermicro SNK-P0064AP4 4U

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Computer PCIe Sound Card with Front Panel Plugs Supports SPDIF PCM 1000Mbps Data Transmission Through Opticals Coaxial PCIe Sound Card

Computer PCIe Sound Card with Front Panel Plugs Supports SPDIF PCM 1000Mbps Data Transmission Through Opticals Coaxial PCIe Sound Card

  • High-Resolution Sound: Supports up to 96kHz, 24-bit audio
  • Multiple Digital Interfaces: Includes optical and coaxial outputs
  • Front Panel Connectivity: Headphone and microphone ports

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

Broadcom and the New AI Chip Race: How Custom ASICs, Google’s TPUs, Meta’s AI Plans, and the Demand for Powerful Silicon Could Shape the Future of Artificial Intelligence

Broadcom and the New AI Chip Race: How Custom ASICs, Google’s TPUs, Meta’s AI Plans, and the Demand for Powerful Silicon Could Shape the Future of Artificial Intelligence

As an affiliate, we earn on qualifying purchases.

As an affiliate, we earn on qualifying purchases.

You May Also Like

Grimfaste: Operations for a Fleet

For eight days this series has been about two things: making stuff…

OpenAI’s New Power Stack: How AMD, NVIDIA, CoreWeave, and “Stargate” Rewire AI’s Supply Chain—and Why the ChatGPT “App OS” Matters

Executive summary OpenAI is building a vertically integrated power stack across compute,…

OpenAI’s Move into Cloud Hosting: Impact on Partnerships and Market Landscape

OpenAI’s Cloud Ambitions and Infrastructure Shift OpenAI is transitioning from being solely…

Outcome-First Decisions: Keep, Change, or Kill

The hardest decision in any portfolio isn’t what to start. It’s what…