Dylan Patel - Inside the Trillion-Dollar AI Buildout - [Invest Like the Best, EP.442]
Episode
118 min
Read time
2 min
Topics
Relationships, Investing, Startups
AI-Generated Summary
Key Takeaways
- ✓Infrastructure Economics: One gigawatt of data center capacity costs $50-75 billion over five years in rental payments, with $10-15 billion annual operating costs. NVIDIA captures roughly $35 billion of initial $50 billion CapEx per gigawatt, maintaining 75% gross margins while effectively lowering prices through equity investments in customers like OpenAI.
- ✓Scaling Law Reality: Model improvement follows log-log scaling where 10x more compute yields one tier of capability increase. This resembles progression from six-year-old to thirteen-year-old intelligence levels. Pre-training on text data reaches late innings, but multimodal pre-training and reinforcement learning remain in second inning, with vast unexplored territory in environment-based learning.
- ✓Tokenomics Trade-offs: Companies face critical decisions between serving larger, slower models with higher intelligence versus smaller, faster models with broader adoption. OpenAI chose GPT-5 at similar size to GPT-4 rather than scaling up because user experience degrades with latency, limiting revenue despite superior capabilities in larger models like Claude Opus.
- ✓Reinforcement Learning Paradigm: Post-training through synthetic environments enables models to learn tasks absent from internet data, like spreadsheet manipulation or physical object recognition. This approach generates training data through iterative trial-and-error in simulated environments, teaching models to reason through problems rather than memorize answers, fundamentally changing capability development.
- ✓Value Capture Dynamics: Gross profit currently flows to hardware layer (NVIDIA, Broadcom) while application companies like Cursor send most revenue to model providers (Anthropic), who reinvest in training compute. Power shifts as application companies accumulate proprietary user interaction data and can train specialized models, creating frenemy relationships throughout the stack.
What It Covers
Dylan Patel maps the trillion-dollar AI infrastructure buildout, explaining OpenAI's strategic partnerships with NVIDIA and Oracle, the economics of gigawatt-scale data centers costing $50 billion each, reinforcement learning's early innings, and why America's competitive position depends on AI success.
Key Questions Answered
- •Infrastructure Economics: One gigawatt of data center capacity costs $50-75 billion over five years in rental payments, with $10-15 billion annual operating costs. NVIDIA captures roughly $35 billion of initial $50 billion CapEx per gigawatt, maintaining 75% gross margins while effectively lowering prices through equity investments in customers like OpenAI.
- •Scaling Law Reality: Model improvement follows log-log scaling where 10x more compute yields one tier of capability increase. This resembles progression from six-year-old to thirteen-year-old intelligence levels. Pre-training on text data reaches late innings, but multimodal pre-training and reinforcement learning remain in second inning, with vast unexplored territory in environment-based learning.
- •Tokenomics Trade-offs: Companies face critical decisions between serving larger, slower models with higher intelligence versus smaller, faster models with broader adoption. OpenAI chose GPT-5 at similar size to GPT-4 rather than scaling up because user experience degrades with latency, limiting revenue despite superior capabilities in larger models like Claude Opus.
- •Reinforcement Learning Paradigm: Post-training through synthetic environments enables models to learn tasks absent from internet data, like spreadsheet manipulation or physical object recognition. This approach generates training data through iterative trial-and-error in simulated environments, teaching models to reason through problems rather than memorize answers, fundamentally changing capability development.
- •Value Capture Dynamics: Gross profit currently flows to hardware layer (NVIDIA, Broadcom) while application companies like Cursor send most revenue to model providers (Anthropic), who reinvest in training compute. Power shifts as application companies accumulate proprietary user interaction data and can train specialized models, creating frenemy relationships throughout the stack.
Notable Moment
Patel reveals three-month-old infants calibrate finger sensitivity by placing hands in mouths, using tongues as reference sensors. He argues AI models need equivalent embodied learning experiences to achieve human-level intelligence, suggesting current approaches miss fundamental aspects of how biological intelligence develops through physical world interaction and sensory feedback loops.
Episode Transcript
Here's an interesting question to think about. If your finance team suddenly had an extra week every month, what would you have them work on? Most CFOs don't know because their finance teams are grinding it out on lost expense reports, invoice coding, and tracking down receipts until the last possible minute. That's exactly the problem that Ramp set out to solve. Looking at the parts of finance everyone quietly hates and asking why are humans doing any of this? Turns out they don't need to. Ramp's AI handles 85% of expense reviews automatically with 99% accuracy, which means your finance team stops being the department that processes stuff and starts being the team that thinks about stuff. Here's the real shift. Companies using Ramp aren't just saving time, they're reallocating it. While competitors spend two weeks closing their books, you're already planning next quarter. While they're cleaning up spreadsheets, you're thinking about new pricing strategy, new markets, and where the next dollar of ROI comes from. That difference compounds. Go to ramp.com/invest to try Ramp and see how much leverage your team gains when the work you have to do stops getting in the way of the work that you want to do. To me, Ridgeline isn't just a software provider. It's a true partner in innovation. They're redefining what's possible in asset management technology, helping firms scale faster, operate smarter, and stay ahead of the curve. I wanna share a real world example of how they're making a difference. Let me introduce you to Brian. Brian, please introduce yourself and tell us a bit about your role. My name is Brian Strang. I'm the technical operations lead, and I work at Congress Asset Management. How would you describe your experience working with Ridgeline? Ridgeline is a technology partner, not a software vendor, and the people really care. I get sales calls all the time, and I ignore them. Ridgeline sold me very quickly. We went from 7,000,000,000 to 23,000,000,000, and the goal is 50,000,000,000. Ridgeline was the clear front runner to help us scale. In your view, what most distinguishes Ridgeline? They reimagined how this industry should work because obviously they were operating on another level. It's worth reaching out to Ridgeline to see what the unlock can be for your firm. Visit ridgelineapps.com to schedule a demo. One of the hardest parts of investing is seeing what's shifting before everyone else does. AlphaSense is helping investors do exactly that. You may already know AlphaSense as the market intelligence platform trusted by 75% of the world's top hedge funds, providing access to over 500,000,000 premium sources from company filings and broker research to news, trade journals, and over 200,000 expert transcript calls. What you might not know is that they've recently launched something game changing, AI powered channel checks. Channel checks give you a real time expert driven perspective on public companies, weeks before they show in earnings or consensus revisions. AlphaSense uses an AI interviewer to run …
Get the full transcript (27,589 words) + summary by email — free
One-time email with the complete transcript and AI summary of this episode. No account needed.
One email, no spam. We’ll also show you what SignalCast does.
Browse all Invest Like the Best with Patrick O'Shaughnessy transcripts →
You just read a 3-minute summary of a 115-minute episode.
Get Invest Like the Best with Patrick O'Shaughnessy summarized like this every Monday — plus up to 2 more podcasts, free.
Pick Your Podcasts — FreeKeep Reading
More from Invest Like the Best with Patrick O'Shaughnessy
Walter Russell Mead - How America Keeps Winning - [Invest Like the Best, EP.490]
Sep 8 · 78 min
a16z Podcast
Can Anyone Catch NVIDIA? | The Future of Chips and Infrastructure
Jul 15
More from Invest Like the Best with Patrick O'Shaughnessy
Sarah Guo - What the 250 People Building AI Believe - [Invest Like the Best, EP.489]
Sep 1 · 59 min
Dwarkesh Podcast
Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute
Mar 13
Books, tools, and gear mentioned in this episode
SignalCast may earn commission on purchases via these links.
Tools
“application companies like Cursor send most revenue to model providers (Anthropic), who reinvest in training compute.”
More from Invest Like the Best with Patrick O'Shaughnessy
We summarize every new episode. Want them in your inbox?
Walter Russell Mead - How America Keeps Winning - [Invest Like the Best, EP.490]
Sarah Guo - What the 250 People Building AI Believe - [Invest Like the Best, EP.489]
Neil Movva - Making AI 10x Cheaper - [Invest Like the Best, EP.488]
Ben Thompson on Big Tech, China, and the AI Boom Running Out of Money - [Invest Like the Best, EP.487]
Eric Vishria - A Decade of Lessons Investing in Software & Hardware - [Invest Like the Best, EP.486]
Similar Episodes
Related episodes from other podcasts
a16z Podcast
Jul 15
Can Anyone Catch NVIDIA? | The Future of Chips and Infrastructure
Dwarkesh Podcast
Mar 13
Dylan Patel — Deep dive on the 3 big bottlenecks to scaling AI compute
Moonshots with Peter Diamandis
Jan 27
Davos 2026: The US-China AI Race, GPU Diplomacy, and Robots Walking the Streets | #225
The Diary of a CEO
Aug 27
The Man Who Calls BS On AI: AI Is The World’s Greatest SCAM, And They All Know It! | Ed Zitron
Dwarkesh Podcast
Aug 25
Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028
Explore Related Topics
This podcast is featured in Best Investing Podcasts (2026) — ranked and reviewed with AI summaries.
Read this week's Investing & Markets Podcast Insights — cross-podcast analysis updated weekly.
You're clearly into Invest Like the Best with Patrick O'Shaughnessy.
Every Monday, we deliver AI summaries of the latest episodes from Invest Like the Best with Patrick O'Shaughnessy and 192+ other podcasts. Free for one show.
Start My Monday DigestNo credit card · Unsubscribe anytime