Gavin Baker on Orbital Compute, TSMC, and Frontier Models
Taiwan Semiconductor's disciplined capacity expansion is single-handedly preventing an AI infrastructure bubble. While companies could theoretically buy $2-3 trillion in GPUs if supply allowed, TSM's controlled wafer production creates a natural brake on overbuilding—similar to how they prevented te
1h 22mKey Takeaway
Taiwan Semiconductor's disciplined capacity expansion is single-handedly preventing an AI infrastructure bubble. While companies could theoretically buy $2-3 trillion in GPUs if supply allowed, TSM's controlled wafer production creates a natural brake on overbuilding—similar to how they prevented tech bubbles in previous cycles. This constraint forces efficient capital allocation and keeps AI infrastructure buildout sustainable rather than debt-fueled and speculative.
Episode Overview
This episode features an in-depth discussion about AI infrastructure investing, focusing on the extraordinary growth of companies like Anthropic (which added $11 billion ARR in one month), the critical role of Taiwan Semiconductor in preventing market bubbles through supply discipline, and emerging opportunities in orbital compute via SpaceX's satellite data centers.
Key Insights
Anthropic's Unprecedented Revenue Growth Defies Capitalism's History
Anthropic added $11 billion in ARR in a single month—equivalent to what Palantir, Snowflake, and Databricks built collectively over 10 years. This represents the most extraordinary moment in the history of capitalism, with no precedent for this velocity of revenue generation in any previous technology wave.
Taiwan Semi's Supply Discipline Prevents AI Bubble Formation
TSM's controlled capacity expansion is preventing a massive AI infrastructure bubble. If they produced all the wafers Jensen Huang wanted, Nvidia could sell $2-3 trillion in GPUs annually, likely creating overcapacity. TSM's restraint forces efficient capital allocation and prevents the debt-fueled overbuilding that characterized previous tech bubbles.
Returns Concentrate at the Frontier Despite Open Source Models
Overwhelmingly, economic value at the model layer accrues to frontier tokens, not cheaper alternatives. Models that seemed incredible months ago (like Gemini 3.1 Pro) become "intolerable" as the frontier advances. This concentration of returns at the cutting edge is surprising but consistent across the AI landscape.
SpaceX's Orbital Compute Will Solve Long-Term Power Constraints
Orbital compute isn't giant floating data centers—it's GPU racks in space connected by lasers, with massive solar arrays providing unlimited power and radiators for cooling. SpaceX already operates 98% of all satellites and has proven expertise in both cooling and laser communications through Starlink, making this transition more practical than skeptics realize.
Strategic Tariffs Created Massive US Manufacturing Advantage
The Strait of Hormuz being closed proved surprisingly beneficial for America. Natural gas prices in the US dropped 20% while doubling or tripling everywhere else, dramatically improving US manufacturing competitiveness—especially for energy-intensive AI infrastructure—and aligning with administration goals for domestic production.
Notable Quotes
"Anthropic added their combined businesses in one month. That's just nothing like that has ever happened in the history of capitalism. Forget my career. Just the flatout history of capitalism, the history of business."
"I would say broadly speaking there are two kinds of draw downs. There are drawdowns where you're wrong, a company misestimates, your hypothesis was invalidated and you have to take your medicine and you crystallize that loss. And then there are draw downs or periods of underperformance where you you're underperforming because of companies you know really really well and where you profoundly disagree with the price action and you can lean in and instead of crystallizing negative performance you can kind of build pent up alpha pent up future performance."
"Listen, he's out there landing rockets. I don't see anybody else landing rockets. And the reality is is that 10 years later, no other company is consistently capable of landing and fully reusing an orbital rocket. And none of this works makes sense without reusability."
"Markets are efficient. They correctly understand that this is a foundational technology. There's what Mobison calls a breakdown in diversity. Everyone becomes bullish on this new technology. And I am beginning to worry a little bit about a diversity breakdown. And then you get a bubble."
"Being early is the same thing as being wrong."
Action Items
-
1
Monitor Taiwan Semiconductor's Capacity Decisions as Bubble Indicator
Watch TSM's quarterly capacity expansion announcements closely. If they significantly accelerate wafer production or lose discipline in response to demand, it could signal an impending infrastructure bubble. Their restraint is currently the primary factor preventing overbuilding in AI infrastructure.
-
2
Focus Investment on Pareto Frontier Models
When evaluating AI companies, plot them on the intelligence-versus-cost Pareto frontier. Companies dominating this frontier (currently Anthropic, OpenAI, and XAI with Grok 4.3) capture overwhelming economic value. Avoid companies inside the frontier, even if their absolute performance seems good.
-
3
Distinguish Between Wrong Drawdowns and Opportunity Drawdowns
During portfolio underperformance, categorize each position: Is your thesis invalidated (crystallize the loss), or do you profoundly disagree with price action on companies you know well (lean in and build 'pent-up alpha')? This framework helps avoid both stubborn losses and missed opportunities.
-
4
Track the Race Between TSM, Intel, and Samsung for Second-Source Emergence
Monitor whether Intel or Samsung can capture >30% market share at leading nodes. If TSM maintains its 9-15 month lead while expanding enough to prevent competitors from scaling, it creates the optimal 'Goldilocks zone' that prevents both bubbles and monopoly concerns.