Invest Like the Best EP.451
Guest: Gavin Baker
Title: Nvidia v. Google, Scaling Laws, and the Economics of AI
Date: December 9, 2025
Host: Patrick O’Shaughnessy
Episode Overview
In this episode, Patrick O’Shaughnessy sits down with Gavin Baker—founder of Atreides Management and renowned technology investor—for a characteristically candid and wide-ranging conversation about the accelerating developments in artificial intelligence (AI), semiconductors, and market dynamics. The discussion ranges from the technical arms race between Nvidia and Google, the importance of scaling laws in AI, transformative new business models, and the implications for global economics and geopolitics. Gavin also shares personal stories and insights into cultivating investment talent, closing with a candid account of his own career journey.
Key Discussion Points & Insights
1. Keeping Up with AI’s Breakneck Pace
- Personal Workflow:
- Gavin describes how he and his team approach rapidly evolving AI developments. Rather than relying solely on summaries, he insists on hands-on use of latest top-tier models (e.g., Gemini Ultra, Super Grok).
- “I'm amazed at how many famous investors are reaching really definitive conclusions about AI based on the free tier. The free tier is like you're dealing with a 10 year old... you have to pay for the highest tier, where those are like a fully fledged 30, 35 year old.” (05:26)
- Gavin describes how he and his team approach rapidly evolving AI developments. Rather than relying solely on summaries, he insists on hands-on use of latest top-tier models (e.g., Gemini Ultra, Super Grok).
- Sources of Signal:
- Stresses the importance of following frontier researchers (OpenAI, Gemini, Anthropic, Xai) on X (Twitter) and reading papers and posts from figures like Andrej Karpathy.
- "Everything in AI is just downstream of those people. Everything Andrej Karpathy writes, you have to read it three times minimum." (07:04)
2. Scaling Laws: The Foundation for AI Progress
- Gemini 3 as a Milestone:
- Gemini 3 confirmed that “scaling laws for pre-training are intact,” a pivotal empirical development even if the underlying mechanics are not fully understood.
- “No one on planet Earth knows how or why scaling laws for pre-training work... Perfect measurement. They didn't understand orbital mechanics.” (08:21)
- Gemini 3 confirmed that “scaling laws for pre-training are intact,” a pivotal empirical development even if the underlying mechanics are not fully understood.
- Post-Training and RL with Verified Rewards:
- The field shifted rapidly in 2024-2025 due to breakthroughs in reinforcement learning with verified rewards and test time compute—dubbed "new scaling laws."
- “With software, anything you can specify you can automate. With AI, anything you can verify you can automate.” (10:10)
- The field shifted rapidly in 2024-2025 due to breakthroughs in reinforcement learning with verified rewards and test time compute—dubbed "new scaling laws."
3. Nvidia vs. Google: Hardware Wars & Strategic Economics
- Blackwell GPU Rollout:
- Explains the technical leap and challenges in deploying Nvidia’s Blackwell chips, from shifting to liquid cooling and increasing power draw, to an unprecedented product transition.
- “Imagine if to get a new iPhone you had to change all the outlets in your house... That’s the power.” (11:21)
- Explains the technical leap and challenges in deploying Nvidia’s Blackwell chips, from shifting to liquid cooling and increasing power draw, to an unprecedented product transition.
- Google’s TPU Strategy:
- Google gained a temporary advantage training Gemini 3 on advanced TPUs while Nvidia struggled to scale Blackwell.
- “Google has been sucking the economic oxygen out of the AI ecosystem, which is an extremely rational strategy.” (13:40)
- Google gained a temporary advantage training Gemini 3 on advanced TPUs while Nvidia struggled to scale Blackwell.
- Economics of Token Production:
- Unique to AI: Being the lowest cost token producer is finally a competitive advantage, unlike previous tech booms (Apple, Microsoft, Nvidia’s past hardware).
4. Frontier Labs, the Flywheel of Reasoning, & Competitive Barriers
-
New Feedback Loop:
- Reasoning and RL with verified rewards allow for the flywheel famously seen in Internet companies to also benefit AI (ex: users → data → improved models → more users).
- “Reasoning fundamentally changed the industry dynamics of Frontier Labs... That flywheel has started to spin.” (42:49)
- Reasoning and RL with verified rewards allow for the flywheel famously seen in Internet companies to also benefit AI (ex: users → data → improved models → more users).
-
Widening the Gap:
- Having internal, more advanced “checkpoints” of new models gives the four leading labs (OpenAI, Gemini, XAI, Anthropic) a sustained edge.
- “If you do not have that latest checkpoint, it's getting really hard to catch up.” (46:56)
- Having internal, more advanced “checkpoints” of new models gives the four leading labs (OpenAI, Gemini, XAI, Anthropic) a sustained edge.
-
China’s Position:
- China’s self-imposed limitations on Blackwell chip imports may widen the gap between US and Chinese AI research.
- “China thought they had the leverage. They're going to realize, oh, whoopsie daisy, we do need the Blackwells.” (48:05)
- China’s self-imposed limitations on Blackwell chip imports may widen the gap between US and Chinese AI research.
5. Edge AI as a Bear Case and the Arc of Compute Demand
-
Edge AI Risk:
- If devices achieve sufficient on-device inference speeds for consumer-grade AI, cloud compute demand could stall.
- “Edge AI is to me by far the most plausible and scariest bear case.” (29:22)
- If devices achieve sufficient on-device inference speeds for consumer-grade AI, cloud compute demand could stall.
-
On S-Curves and Frontier Model Utility:
- Progress is now less obvious to laypeople, and the industry’s next phase hinge more on improved usefulness and task automation than raw intelligence.
- “We need to shift from getting more intelligent to more useful.” (31:48)
- Progress is now less obvious to laypeople, and the industry’s next phase hinge more on improved usefulness and task automation than raw intelligence.
6. Physical Constraints: Power, Data Center Buildouts, & Space Renaissance
- Power as the Ultimate Governor:
- Watt limits benefit the most advanced hardware; locations with abundant, cheap energy (natural gas, solar) will attract more AI buildout.
- “Having watts as a constraint is good for the most advanced compute players... If you could get 3x more tokens per watt, that is 3x more revenue.” (60:28)
- Watt limits benefit the most advanced hardware; locations with abundant, cheap energy (natural gas, solar) will attract more AI buildout.
- Data Centers in Space:
- Details the logic and economics of eventually moving data centers into orbit, leveraging solar, free cooling, and high-speed laser links.
- “Data centers should be in space... In every way, data centers in space, from a first principles perspective, are superior.” (53:15)
- “In space, cooling is free. You just put a radiator on the dark side of the satellite, it's f***ing gold.” (53:22)
- Details the logic and economics of eventually moving data centers into orbit, leveraging solar, free cooling, and high-speed laser links.
7. AI, Productivity, and the Real-World ROI
-
Productivity Gains:
- Early Fortune 500 examples (e.g., C.H. Robinson improving freight matching via AI) demonstrate concrete ROI, not just cost savings.
- “With AI, they're quoting 100% and doing it in seconds... the stock went up 20% and it was because of AI-driven productivity.” (36:18)
- Early Fortune 500 examples (e.g., C.H. Robinson improving freight matching via AI) demonstrate concrete ROI, not just cost savings.
-
SMB and Startup Impact:
- Startups and tech-forward firms adapt fastest; VCs see productivity gains ahead of public investors.
- “For a given level of revenue, a company today has significantly lower employees than a company of two years ago. The reason is AI is doing a lot...” (35:43)
- Startups and tech-forward firms adapt fastest; VCs see productivity gains ahead of public investors.
8. VC and Talent Dynamics in Semiconductors and AI Natives
- Semicon VC Renaissance:
- Driven by Nvidia’s growth, veteran semiconductor designers are founding startups to address every facet of the supply chain.
- “Your average semiconductor venture founder is like 50 years old. Nvidia has single handedly ignited semiconductor venture.” (64:44)
- Driven by Nvidia’s growth, veteran semiconductor designers are founding startups to address every facet of the supply chain.
- AI Native Entrepreneurs:
- Next-gen founders are more skilled, more rapidly, thanks to native AI usage—using models to research, handle HR, and strategize sales.
- “They're just so impressive in all ways, and they get more polished faster... because they're talking to the AI.” (63:18)
- Next-gen founders are more skilled, more rapidly, thanks to native AI usage—using models to research, handle HR, and strategize sales.
9. SaaS at an Inflection Point
- Agent Opportunity & Margin Compression:
- SaaS firms must accept lower gross margins for true AI features or risk obsolescence, just as brick-and-mortar retail missed out on e-commerce by protecting legacy margins.
- “If you are trying to preserve an 80% gross margin structure, you are guaranteeing that you will not succeed in AI. Absolute guarantee.” (69:00)
- “This is a life or death decision and essentially everyone except Microsoft is failing it. Their platforms are burning.” (73:40)
- SaaS firms must accept lower gross margins for true AI features or risk obsolescence, just as brick-and-mortar retail missed out on e-commerce by protecting legacy margins.
10. Cycles, Gluts, and Natural Bottlenecks
- Lessons from Semi Inventory Cycles:
- Historical overbuilds are mitigated by caution at Taiwan Semi and natural constraints in other supply chain areas.
- “I think Tawan Simi's in the process of making a mistake. But they're just so paranoid about an overbuild and they're so skeptical.” (58:02)
- DRAM Supply as a Governor:
- A real DRAM price spike—as in the 90s—could significantly slow AI’s ascent.
- Historical overbuilds are mitigated by caution at Taiwan Semi and natural constraints in other supply chain areas.
Notable Quotes & Moments
On Scaling Laws and the State of AI
- “No one on planet Earth knows how or why scaling laws for pre-training work. It's actually not a law. It's an empirical observation."
- Gavin Baker (08:21)
On Data Centers in Space
- “Cooling is free. You just put a radiator on the dark side of the satellite, it's f***ing gold. And it's as close to absolute zero as you can get.”
- Gavin Baker (53:22)
On the Economics of AI
- “AI is the first time in my career as a tech investor that being the low cost producer has ever mattered.”
- Gavin Baker (13:35)
On SaaS and AI Margins
- “If you are trying to preserve an 80% gross margin structure, you are guaranteeing that you will not succeed in AI. Absolute guarantee.”
- Gavin Baker (69:00)
On Investing and the Search for Truth
- “At some level, investing is the search for truth. If you find truth first and you're right about it being a truth, that's how you generate alpha. And it has to be a truth that other people have not yet seen. You're searching for hidden truths.”
- Gavin Baker (78:32)
On Raising Entrepreneurship Talent
- “These kids who are growing up native in AI, they are just proficient with it in a way that I am trying really hard to become.”
- Gavin Baker (64:28)
Timestamps for Key Segments
| Segment Topic | Start Time | |-------------------------------------------------------|------------| | How Gavin processes new AI developments | 04:58 | | Scaling laws: meaning and importance | 08:04 | | Nvidia Blackwell vs. Google’s TPU | 11:35 | | Temporary advantages and cost of tokens | 13:22 | | Frontier models, checkpoints, and data cycles | 42:49 | | Edge AI bear case and model usefulness | 29:22 | | Power, supply gluts, & natural regulatory governors | 60:13 | | Data centers in space—first principles and economics | 51:49 | | AI productivity and startup/enterprise adoption | 35:03 | | VC, semiconductor renaissance, AI-native founders | 63:18 | | SaaS crossroads—agent strategy & margin contraction | 68:28 | | Niche “bubbles” (nuclear, quantum), market lessons | 74:03 | | Investing philosophy, behavioral culture at Atreides | 77:01 | | Gavin’s origin story in investing | 81:22 |
The Final Story: Gavin’s Own Investing Journey
Gavin reveals the unusual path that led him to investing—via a deep interest in history, a stint as a ski bum, and parental urging to get “just one” professional internship. A serendipitous role reading stock research for a wealth management firm exposed him to the game-like nature of markets, fusing his love for history, competition, and argument.
“The way you got an edge in this greatest game of skill and chance imaginable was you had the most thorough knowledge possible of history, and you intersected that with the most accurate understanding of current events in the world to form a differential opinion on what was going to happen next.” (84:12)
Episode Tone & Takeaways
Baker’s enthusiasm for technology, investing, and discovery animates the conversation. The tone is analytical yet irreverent, especially when dissecting technical nuances or calling out strategic blunders in the industry. Baker is equally candid about his own limitations and passions. For investors, entrepreneurs, or anyone interested in technology’s trajectory, this episode delivers a dense but accessible primer on how the hardware, economics, and culture of AI is reshaping markets and careers on a global scale.
For deeper dives and episode archives:
Colossus | JoinColossus.com
