Podchemy

Highlights from this episode with @dylan522p x @dwarkesh_sp > AI labs may need to more than double their compute capacity within a single year just to support the inference demands of their projected revenue growth. > The Alchian-Allen effect suggests that as the fixed cost of compute rises, users will gravitate toward the highest quality AI models because the relative price gap narrows. > The ultimate constraint on AI scaling by the end of the decade will be the production capacity of EUV lithography machines by ASML. > While data centers can be built in under a year, semiconductor fabs require two to three years of construction, creating a significant lead time disparity. > Nvidia demonstrates extreme financial leverage by turning a small fraction of TSMC's hardware investment into $160 billion in annual revenue. > Unlike other hardware monopolies, ASML links its price increases directly to improvements in machine throughput and accuracy rather than pure market demand. > The primary challenge in scaling AI is the efficiency loss that occurs when moving data between hundreds of interconnected chips. > AI demand is causing a memory crunch that could triple the component costs for smartphones, leading to significantly higher retail prices. > AI data centers can access more power by using batteries to manage peak loads, which unlocks the 20% of the grid that usually stays idle. > Power costs are a minor factor in AI infrastructure because the value generated by improved models far exceeds the expense of even doubling electricity prices. > Modularization allows data centers to scale despite labor shortages by shifting complex wiring and plumbing from construction sites to specialized factories. > GPUs have high failure rates, often requiring physical repairs that make remote or space-based deployments economically and logistically risky. > Compute efficiency gains from research can make model training ten times cheaper annually. Because of this, labs prioritize research over massive pre-training to achieve the fastest possible technological takeoff. > Apple is losing its status as TSMC's most favored customer as AI companies begin prepaying for chip capacity and manufacturing costs. > The semiconductor industry has a circular dependency where the tools needed to make chips require the very chips they produce. > The bottleneck for AI development has shifted from chip design to securing the entire infrastructure stack, including power and land. Link below for notes to the full episode!

.@dylan522p gives a deep dive on 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. 0:00:00 – Why an H100 is worth more today than 3 years ago 0:24:52 – Nvidia secured TSMC allocation early; Google is getting squeezed 0:34:34 – ASML will be the #1 constraint for AI compute scaling by 2030 0:56:06 – Can’t we just use TSMC’s older fabs? 1:05:56 – When will China outscale the West in semis? 1:16:20 – The enormous incoming memory crunch 1:42:53 – Scaling power in the US will not be a problem 1:55:03 – Space GPUs aren't happening this decade 2:14:26 – Why aren’t more hedge funds making the AGI trade? 2:18:49 – Will TSMC kick Apple out from N2? 2:24:35 – Robots and Taiwan risk Look up Dwarkesh Podcast on YouTube, Apple Podcasts, or Spotify. Enjoy!

Highlights from this episode with @jasonfried x @davidsenra > The only competition a business truly has is its own costs because you cannot control your competitors, but you can control your own spending. > Software lacks the physical constraints of the real world, which often leads to infinite expansion and a decline in quality over time. > Instead of pursuing infinite growth, businesses should aim to reach orbit. After the initial push to overcome gravity, the focus should shift to maintaining a sustainable and enjoyable level of quality. > Long term planning is often an illusion. It is more effective to plan in short cycles and course correct like a squirrel crossing a field. > A great life is a string of great days. Focus on getting the next 24 hours right instead of worrying about a five year plan. > Seeking inspiration from unrelated fields like architecture or nature provides a fresh perspective that industry-specific products cannot offer. > Writing a product story during development helps define its purpose before the features are finished. > Insights are like turning the dial on an old radio. The other frequencies are always there, but we are usually tuned into only one channel. > Distribute profits based on employee longevity rather than job title to reward loyalty with real cash instead of speculative stock options. > Business post-mortems often lead to false certainty because the variables involved in a project outcome are impossible to truly isolate. > Profitability is the ultimate form of independence because it allows a business to survive and make decisions without external approval. > The tech industry often sells regressions as progress by replacing simple, intuitive interfaces with unnecessarily complex ones. > A durable business consists of many small, equal customers rather than a few outliers that the company cannot afford to lose. > Management layers can act like a game of telephone where information is lost between leaders and the people doing the actual work. > True peace comes from being comfortable with what you have built and realizing that it is enough, rather than chasing the constant cycle of growth and serial entrepreneurship. > The most effective way to evaluate an employee is to ask if you would hire them again knowing everything you learned during their first year. > A business should be a thin shell that holds a thick product. High organizational mass makes it difficult to change direction and distances the company from its customers. > Real learning occurs through future action rather than past analysis. If you do not like a previous result, the most effective lesson is to simply try a different approach next time. Link to full episode notes below!

My conversation with @JasonFried, co-founder of @37signals. 0:00 Build Products for Yourself 1:40 Low Costs, Small Company, Enough Customers 3:06 Your Only Competition Is Your Costs 5:25 How 37signals Stays Lean 9:43 Rewriting Basecamp & Fighting Software Bloat 13:42 Why "Enough" Beats Growth 17:44 Product People vs. Business Shells 22:41 The "So What?" Mindset 27:45 Staying Close to Customers 34:43 The Reward for Good Work Is More Work 39:57 Six-Week Horizons & Compounding Decisions 45:20 Anti-Fragile Business With Tiny Units 50:55 Galápagos Product Design 52:44 Radical Authenticity Over Marketing Tricks 1:27:39 Rick Rubin & Intuition-Driven Building 1:42:25 Lightning in a Bottle & Knowing When to Stop 1:50:29 Defining Success: Pride in the Work 1:53:58 Independence Through Profitability 1:59:23 When Tech Adds Friction Instead of Value 2:04:11 Ruthless Editing & What Never Changes 2:08:14 Longevity as the Moat 2:17:28 Building by Intuition Includes paid partnerships.

Highlights from this episode with @jasonfried x @davidsenra > The only competition a business truly has is its own costs because you cannot control your competitors, but you can control your own spending. > Software lacks the physical constraints of the real world, which often leads to infinite expansion and a decline in quality over time. > Instead of pursuing infinite growth, businesses should aim to reach orbit. After the initial push to overcome gravity, the focus should shift to maintaining a sustainable and enjoyable level of quality. > Long term planning is often an illusion. It is more effective to plan in short cycles and course correct like a squirrel crossing a field. > A great life is a string of great days. Focus on getting the next 24 hours right instead of worrying about a five year plan. > Seeking inspiration from unrelated fields like architecture or nature provides a fresh perspective that industry-specific products cannot offer. > Writing a product story during development helps define its purpose before the features are finished. > Insights are like turning the dial on an old radio. The other frequencies are always there, but we are usually tuned into only one channel. > Distribute profits based on employee longevity rather than job title to reward loyalty with real cash instead of speculative stock options. > Business post-mortems often lead to false certainty because the variables involved in a project outcome are impossible to truly isolate. > Profitability is the ultimate form of independence because it allows a business to survive and make decisions without external approval. > The tech industry often sells regressions as progress by replacing simple, intuitive interfaces with unnecessarily complex ones. > A durable business consists of many small, equal customers rather than a few outliers that the company cannot afford to lose. > Management layers can act like a game of telephone where information is lost between leaders and the people doing the actual work. > True peace comes from being comfortable with what you have built and realizing that it is enough, rather than chasing the constant cycle of growth and serial entrepreneurship. > The most effective way to evaluate an employee is to ask if you would hire them again knowing everything you learned during their first year. > A business should be a thin shell that holds a thick product. High organizational mass makes it difficult to change direction and distances the company from its customers. > Real learning occurs through future action rather than past analysis. If you do not like a previous result, the most effective lesson is to simply try a different approach next time. Link to full episode notes below!

My conversation with @JasonFried, co-founder of @37signals. 0:00 Build Products for Yourself 1:40 Low Costs, Small Company, Enough Customers 3:06 Your Only Competition Is Your Costs 5:25 How 37signals Stays Lean 9:43 Rewriting Basecamp & Fighting Software Bloat 13:42 Why "Enough" Beats Growth 17:44 Product People vs. Business Shells 22:41 The "So What?" Mindset 27:45 Staying Close to Customers 34:43 The Reward for Good Work Is More Work 39:57 Six-Week Horizons & Compounding Decisions 45:20 Anti-Fragile Business With Tiny Units 50:55 Galápagos Product Design 52:44 Radical Authenticity Over Marketing Tricks 1:27:39 Rick Rubin & Intuition-Driven Building 1:42:25 Lightning in a Bottle & Knowing When to Stop 1:50:29 Defining Success: Pride in the Work 1:53:58 Independence Through Profitability 1:59:23 When Tech Adds Friction Instead of Value 2:04:11 Ruthless Editing & What Never Changes 2:08:14 Longevity as the Moat 2:17:28 Building by Intuition Includes paid partnerships.

The one and only @nikitabier is today's guest on Out of Office. He's @x's head of product, a @lightspeedvp venture partner, and the world's leading expert on consumer product growth. He and I recently had a long conversation in Los Angeles where we covered the evolution of the X product, how he mastered growth by hacking and learning to "think like an adversary", and what it's like to work with Elon Musk. Enjoy. Chapters: 00:00 No Silver Bullet to Growth 00:18 The Evolution of X: From PayPal to Financial Network 01:25 A Tour of South Bay 01:56 Growing Up in South Bay: Internet Adventures 03:11 From Hacking to Growth Hacking 06:57 Joining X: The Journey and Challenges 09:14 Revolutionizing X: Growth and Product Development 15:44 The Role of a Product Head at X 29:13 Balancing Free Speech and Authenticity on X 34:58 The Future of X: AI, Links, and User Engagement 37:22 Solving Engagement Issues on X 40:42 Working with Elon Musk 50:13 Building Viral Apps: TBH and Gas 01:02:35 Consulting and Angel Investing 01:10:09 AI's Impact on App Development 01:12:49 Personal Anecdotes and Reflections

Please enjoy this Cheeky Pint / @dwarkesh_sp crossover with @elonmusk. Dwarkesh was most interested in how Elon is going to make space datacenters work. I was most interested in Elon's method for attacking hard technical problems, and why it hasn’t been replicated as much as you might expect. But we got into plenty of topics in this three-hour session. 00:00:23 Space GPUs 00:35:39 Alignment 00:58:48 xAI 01:15:01 Optimus 01:28:03 China 01:40:46 Management 02:16:38 DOGE 02:34:58 Space GPUs redux

Highlights from @abcampbell on Monetary Matters with @JackFarley96! > The silver market is prone to price squeezes because 75 percent of its supply is a byproduct of mining other metals, making production insensitive to silver price changes. > Over the last decade, more than 100 percent of returns for many commodities occurred during overnight hours, indicating that Asian markets are the primary price drivers. > Perfect arbitrage is a myth in physical markets because moving metal involves time, regulatory hurdles, and transportation costs. > AI is making software development so efficient that many companies will choose to build their own tools instead of paying for expensive subscriptions. > A significant portion of modern service costs in healthcare and education stems from an army of administrators navigating inefficient software rather than the core services provided. > China is using gold accumulation as a strategy to turn its currency into a global reserve without opening its capital accounts. > International investors looking to exit the US dollar are more likely to sell high-profile stocks than currency futures, creating downward pressure on the S&P 500. > Proprietary data is only a true moat if the collection process cannot be easily automated by newer software companies. > The decline of globalization and the move toward independent supply chains is a structurally inflationary trend that favors physical assets over financial paper. > The resurgence of local compute will allow businesses to run sophisticated AI models on their own hardware to ensure better security and lower long-term costs. Read full episode notes here: podchemy.com/notes/why-silv…