Subvolts take Software and hardware co-design, not just faster chips, unlocks AI's 100x potential. Jun 30, 2026 - Developers
- Learn how to optimize AI models and software for specific hardware to achieve peak performance.
- Business
- Understand the critical role of hardware-software integration in driving AI efficiency and market growth.
This interview provides expert analysis on a critical factor driving AI performance and efficiency, directly relevant to major AI labs.
Why Hardware-Software Co-Design Is AI's Real 100x: Dylan Patel of SemiAnalysis
Audio streams from Sequoia Capital. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Dylan Patel of SemiAnalysis joins Sequoia Capital's Training Data podcast to discuss how software-hardware co-design drives AI's biggest gains. He explains how optimizing models, kernels, and silicon together yields massive performance improvements. Patel also details his InferenceX benchmark, the growing inference market, and the compute crunch.
This episode explains that the biggest AI improvements come from making software and computer parts work together perfectly. This smart teamwork makes AI much faster and cheaper, which is important for everyone using AI tools.
This episode offers deep insights into the engineering behind AI's rapid progress, moving beyond just faster chips. Listeners will learn how specialized hardware and software integration create significant efficiency leaps. It's for AI engineers, hardware designers, and strategists at frontier labs like Google Gemini, OpenAI, and Anthropic who are grappling with compute limitations and seeking performance breakthroughs. Patel's perspective on co-design directly impacts how these labs approach model development and hardware investment.
- HostSequoia Capital
- GuestDylan Patelfounder, SemiAnalysis
- Guest: Dylan Patel, founder of SemiAnalysis
- Episode Title: Why Hardware-Software Co-Design Is AI's Real 100x
- Show: Training Data (an official AI show)
- Length: 70 minutes
- Co-design impact: Optimizing model, kernels, and silicon together turns 2x gains into 100x.
- InferenceX benchmark: Tracks roughly 60x annual drop in cost per unit of quality.
Dylan Patel, founder of SemiAnalysis, argues the biggest gains in AI don't come from faster chips, they come from software-hardware co-design. Optimizing the model, the kernels, and the silicon together turns a 2x here and a 2x there into 100x. He explains why DeepSeek's experts were shaped for Nvidia's Hopper (and why TPUs struggle to run it), why OpenAI's sparser models and Anthropic's denser ones pull them toward different hardware, and why the so-called CUDA moat was never really about CUDA. Dylan breaks down InferenceX, his living benchmark that runs the latest models on over $50M of donated hardware daily, tracking a roughly 60x annual drop in cost per unit of quality. He makes the case that inference will be a bigger market than oil, that the compute crunch persists because models expand the value of useful work faster than compute grows, and why Jensen Huang is bankrolling neoclouds to engineer a multipolar world.
Hosted by Shaun Maguire and Sonya Huang, Sequoia Capital
1 h 08David Silver and Lasse Espeholt of Ineffable Intelligence: A Superlearner for the Age of Experience
WithDavid Silver, Lasse Espeholt
1 h 041 h 0552 minMaking Cities Awesome: Peregrine’s Nick Noone & Ben Rudolph
WithNick Noone, Ben Rudolph
Same guests, other shows
1 h 02Can AI Keep Growing Exponentially? Let’s Ask SemiAnalysis — With Dylan Patel and Jordan Nanos
WithDylan Patel, Jordan Nanos
1 h 172 h 31
1 h 17
1 h 01
Questions people ask
- What is software-hardware co-design?
- It is the practice of optimizing AI models, software kernels, and the underlying silicon together to achieve significant performance gains, potentially reaching 100x improvements.
- What is InferenceX?
- InferenceX is Dylan Patel's living benchmark that tracks the cost per unit of quality for running the latest AI models on donated hardware.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Sequoia Capital.