Subvolts take Understand the core hardware limits slowing AI progress and shaping its future economics. Mar 13, 2026 - Developers
- Learn about the hardware limitations that shape the AI models you can build and deploy.
- Business
- Understand the economic factors and supply chain challenges affecting AI infrastructure costs and availability.
This is a solid interview with an expert on the critical hardware constraints impacting AI development.
Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute
Audio streams from Dwarkesh Patel. Your place is remembered on this device.
From Subvolts
Summary and analysis by Subvolts editorial, AI-assisted.
Dwarkesh Patel interviews Dylan Patel, founder of SemiAnalysis, on the three main bottlenecks to scaling AI compute: logic, memory, and power. The discussion covers the economics of AI labs, hyperscalers, foundries, and fab equipment manufacturers, offering a deep dive into the entire technology stack.
This episode explains the biggest problems in building enough computer power for AI: the chips themselves, their memory, and the electricity they use. Knowing these limits helps us understand why AI development is happening at its current speed.
This episode is for anyone tracking the physical constraints of AI development, from engineers to investors. It explains the economic realities behind AI hardware, which directly impacts the pace of innovation at frontier labs like OpenAI, Google Gemini, and Anthropic. Understanding these bottlenecks is crucial for anticipating future AI capabilities and the supply chain challenges that accompany them. The conversation highlights how hardware limitations shape the AI landscape.
- HostDwarkesh Patel
- GuestDylan Patelfounder, SemiAnalysis
- Guest: Dylan Patel, founder of SemiAnalysis
- Title: Deep dive on the 3 big bottlenecks to scaling AI compute
- Show: Dwarkesh Podcast
- Length: 151 minutes
- Bottlenecks: logic, memory, and power
Dylan Patel, founder of SemiAnalysis, provides a deep dive into the 3 big bottlenecks to scaling AI compute: logic, memory, and power.
And walks through the economics of labs, hyperscalers, foundries, and fab equipment manufacturers.
Learned a ton about every single level of the stack. Enjoy!
Watch on YouTube; read the transcript.
Sponsors
* Mercury has already saved me a bunch of time this tax season. Last year, I used Mercury to request W-9s from all the contractors I worked with. Then, when it came time to issue 1099s this year, I literally just clicked a button and Mercury sent them out. Learn more at mercury.com.
* Labelbox noticed that even when voice models appear to take interruptions in stride, their performance degrades. To figure out why, they built a new evaluation pipeline called EchoChain. EchoChain diagnoses voice models’ specific failure modes, letting you understand what your model needs to truly handle interruptions. Check it out at labelbox.com/dwarkesh.
* Jane Street is basically a research lab with a trading desk attached – and their infrastructure backs this up. They’ve got tens of thousands of GPUs, hundreds of thousands of CPU cores, and exabytes of storage. This is what it takes to find subtle signals hidden deep within noisy market data. If this sounds interesting, you can explore open positions at janestreet.com/dwarkesh.
Timestamps
(00:00:00) – Why an H100 is worth more today than 3 years ago
(00:24:52) – Nvidia secured TSMC allocation early; Google is getting squeezed
(00:34:34) – ASML will be the #1 constraint for AI compute scaling by 2030
(00:55:47) – Can't we just use TSMC's older fabs?
(01:05:37) – When will China outscale the West in semis?
(01:16:01) – The enormous incoming memory crunch
(01:42:34) – Scaling power in the US will not be a problem
(01:54:44) – Space GPUs aren't happening this decade
(02:14:07) – Why aren't more hedge funds making the AGI trade?
(02:18:30) – Will TSMC kick Apple out from N2?
(02:24:16) – Robots and Taiwan risk
This is a public episode. If you would like to discuss this with other subscribers or get access to bonus episodes, visit www.dwarkesh.com
- AI compute
- Dylan Patel
- The processing power needed to run AI models.
- hyperscalers
- Dylan Patel
- Large cloud computing providers like Amazon, Google, and Microsoft.
- foundries
- Dylan Patel
- Factories that manufacture semiconductor chips.
- fab equipment manufacturers
- Dylan Patel
- Companies that build the machines used in chip factories.
1 h 371 h 201 h 37AI researchers debate how close we are to recursive self-improvement
WithJohn Schulman, Beren Millidge, Charlie O’Neill
2 h 21Same 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 17
1 h 10
1 h 17
1 h 07Andrej Karpathy on Code Agents, AutoResearch, and the Loopy Era of AI
WithAndrej Karpathy
1 h 17Tokens Are the Next Commodity
WithVal Kilmer
1 h 01Benedict Evans: OpenAI’s Moat Problem & the Future of Software
WithBenedict Evans
47 minOpenClaw: Why the Internet Isn't Built for AI Agents
WithYoko Li, Guido Appenzeller, Joel de la Garza
Questions people ask
- What are the main bottlenecks to scaling AI compute?
- The episode identifies logic, memory, and power as the three primary bottlenecks to scaling AI compute.
- What economic aspects of AI hardware are discussed?
- The conversation delves into the economics of AI labs, hyperscalers, foundries, and fab equipment manufacturers.
Page generated Oct 19, 2026. Summaries are Subvolts' own; the audio, artwork and notes belong to Dwarkesh Patel.