Subvolts take OpenAI details a new network protocol to keep massive AI training runs stable and efficient. May 6, 2026 - Developers
- Learn about a new protocol designed to optimize large-scale AI training infrastructure and improve system resilience.
- Business
- Understand the engineering challenges and solutions for building the compute backbone of frontier AI models.
This episode details a novel networking protocol from OpenAI, crucial for scaling AI training, developed with key industry partners.
Why AI needs a new kind of supercomputer network - Episode 18
Audio streams from OpenAI. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
The OpenAI Podcast episode 18 features Mark Handley and Greg Steinbrecher discussing a new supercomputer network design for training AI models. They explain how their Multipath Reliable Connection protocol, developed with industry partners, keeps massive GPU clusters synchronized and resilient to failures. OpenAI is making this protocol an open standard for the entire industry.
Training giant AI models requires thousands of computer chips working together. This episode explains a new way to connect them, like a super-smart road system, that prevents delays and keeps the whole system running smoothly even if one part breaks.
This episode offers insight into the complex hardware infrastructure powering frontier AI development. It's for anyone interested in the engineering challenges behind training massive models, particularly those in hardware and AI infrastructure. The discussion on Multipath Reliable Connection highlights OpenAI's efforts to optimize training efficiency and reliability, a critical area as labs like Google Gemini, Anthropic, and Meta AI scale their own compute resources.
- HostAndrew Mayne
- GuestMark HandleyOpenAI
- GuestGreg SteinbrecherOpenAI
- Episode Title: Why AI needs a new kind of supercomputer network - Episode 18
- Show: The OpenAI Podcast
- Length: 38 minutes
- Protocol Developed: Multipath Reliable Connection (MRC)
- Development Partners: AMD, Broadcom, Intel, Microsoft, Nvidia
- Protocol Availability: Open standard for the industry
Training frontier models isn’t as simple as adding more GPUs—one small problem and the whole coordinated dance falls apart. OpenAI’s Mark Handley and Greg Steinbrecher discuss how a new supercomputer network design, used to train some of the company’s latest models, keeps the whole system moving in lockstep, even with record numbers of GPUs. They break down Multipath Reliable Connection, a new protocol OpenAI developed with AMD, Broadcom, Intel, Microsoft, and Nvidia, and why they’re making it available for the whole industry to use.
Chapters
00:00 Intro
00:39 Greg and Mark's paths to OpenAI
04:34 Why training AI stresses networks differently
10:05 Bottlenecks, failures, and the cost of waiting
15:19 How Multipath Reliable Connection works
18:59 A protocol to route around failures
25:05 Why OpenAI is making MRC an open standard
35:09 Could AI compute move to space?
Hosted on Acast. See acast.com/privacy for more information.
- GPUs
- Greg Steinbrecher and Mark Handley
- Graphics Processing Units, specialized computer chips used for AI calculations.
- Multipath Reliable Connection
- Greg Steinbrecher and Mark Handley
- A new networking protocol designed to keep many computer chips working together smoothly during AI training.
41 minWhat racing reveals about working with AI - Episode 22
WithJoyce Ruffell, Chase Holden
44 minWhy Tejal Patwardhan stopped underestimating the models - Episode 21
WithTejal Patwardhan
41 minHow a reasoning model cracked an 80-year-old math problem - Episode 20
WithAlexander Wei, Hongxun Wu, Lijie Chen
29 minInside image generation’s Renaissance moment - Episode 19
WithAdele Li, Kenji Hata
39 min
22 minQuestions people ask
- What is Multipath Reliable Connection (MRC)?
- MRC is a new network protocol developed by OpenAI and industry partners to keep large numbers of GPUs synchronized during AI model training, routing around failures.
- Why is OpenAI making MRC an open standard?
- OpenAI is sharing the protocol with the entire industry to improve AI training efficiency and resilience across the field.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to OpenAI.