Stewart Squared
Episode #55: From Justin.tv to Claude Hacks: Lessons in Tech, Money, and Security
- AI & Agents
- CapEx
- depreciation
- revenue
- cash accounting
- GAAP accounting
- Nvidia chips
- server farms
- AI infrastructure
Listen
Download MP3About this episode
In this episode of Stewart Squared, Stewart Alsop III sits down with Stewart Alsop II to explore the financial and technical foundations shaping today’s AI and cloud economy, from the staggering scale of CapEx and depreciation schedules to the sustainability of investments by Microsoft, Meta, OpenAI, and Anthropic. The conversation traces historical precedents like the fiber boom, Google’s rise, and the pivot from Justin.tv to Twitch, leading into a discussion of venture capital shifts, IPO trends, and the enduring importance of the “rule of 40.” They also examine Cloudflare’s emerging role in the open internet economy, the rise of agents and Amazon’s use of reinforcement learning gems, and pressing security challenges around AI scraping, ITAR data, and national infrastructure.
Timestamps
00:05 Stewart Alsop introduces the theme of CapEx and depreciation, setting the stage with numbers on massive 2025 infrastructure spending.
00:10 Stewart Alsop II explains depreciation schedules, cash vs GAAP accounting, and how fast AI infrastructure like Nvidia chips and server farms lose value.
00:15 The discussion shifts to Microsoft’s Azure strategy, OpenAI’s spending, and comparisons to the 1999 fiber boom where dark fiber overbuilds reshaped the internet.
00:20 Meta’s dual front in VR/AR and AI is questioned for sustainability, as acquisitions and billion-dollar hiring sprees raise risks.
00:25 Historical precedents emerge: Google’s speed in search, Facebook’s real-time newsfeed infrastructure, and the rise of Twitch from Justin TV through Emmett Shear’s pivot.
00:30 Venture capital lessons are highlighted, from early struggles to explosive growth, with reflections on Series A–C shifts, ZIRP, growth equity vs private equity, and the rule of 40.
00:35 Tesla vs Rivian valuations anchor a risk discussion, then focus moves to Cloudflare, intermediaries, AI web crawling, and pay-by-crawl monetization.
00:40 The episode closes with agents, RLGems, universal verifiers, Amazon and Apple’s data advantages, security concerns with ITAR breaches, and the future of an open internet.
Key Insights
1. Depreciation shapes the economics of AI infrastructure.
Stewart Alsop II explains how massive CapEx spending—such as $392 billion in 2025—must be matched against depreciation schedules, which spread the cost of assets like Nvidia chips and server farms over years. The challenge is that AI hardware becomes obsolete much faster than traditional assets, making the schedule a judgment call that influences sustainability.
2. Microsoft’s position differs from AI-first labs.
Unlike OpenAI or Anthropic, Microsoft already had Azure and enterprise infrastructure in place, so their incremental AI spending built on existing investments. This makes their approach more sustainable and less risky than startups burning cash to compete.
3. Meta faces a “two-front war.”
Meta’s massive CapEx is split between VR/AR hardware bets and AI infrastructure, stretching resources and raising questions about whether its cash flows from social media can continue to fund both without weakening the core business.
4. Historical precedents highlight today’s risks.
The fiber boom of the late 1990s, Google’s breakthrough with fast search, and the pivot from Justin.tv to Twitch show how infrastructure-heavy investments can collapse or succeed depending on timing, user demand, and business model clarity.
5. Venture capital dynamics have shifted.
Seed rounds remain risky and contrarian, but later rounds resemble private equity with safer bets and higher valuations. The “rule of 40” has become a standard measure for balancing growth and profitability when evaluating public companies.
6. Cloudflare positions itself as a gatekeeper.
With 80% of AI companies crawling the web through its network, Cloudflare’s pay-by-crawl model could redefine how publishers monetize access to their content, creating a new intermediary in the AI-driven internet economy.
7. Agents and security are the next frontier.
Amazon’s RLGems and universal verifiers illustrate the push to give AI agents personalization and autonomy, but this shift also heightens security risks. Breaches like ITAR data leaks underscore that the AI-driven world may be even more insecure than today’s internet.
Episode transcript
Everyone is wrong about AI but if you know the history of technology, you can be wrong and still be right.
I know the history of technology since I started with the personal computer in 1980. So listen in while we explore what is right about AI.
Welcome to the Stuart Squared podcast. Today we are going to do a deep dive into the money the financial business foundations of how to think about investing huge amounts of capex, including $392 billion in 2025 alone, matched against depreciation of $40 billion a year. But that's what I want to ask you later as what is depreciation? Why does it matter? Versus $20 billion in revenue? And this is all of them together, all the foundational labs, that's all they spent. Is this something that is sustainable?
So you mentioned why you were saying we will talk about that later but. For some reason my limitless app came up go away. The so let's start with depreciation and understand what it is because that's kind of the basis for what the question you're asking the so simple accounting. You know, if you're a sole proprietor and person, what money you spend has to be put against whatever income you have. So income and outgo. That's very simple. Cash. It's called cash accounting. Ironically, the US government is on cash accounting. So every time they buy a $3 billion aircraft carrier that's EBCo, they don't account for the how long that thing will last. And that's where the concept of depreciation comes in is in business accounting, double entry accounting and GAAP accounting. The idea is to account for how the benefit of that expenditure over time for buying something that lasts. So so there's various schedules of depreciation for different kinds of things that you buy a car normally depreciates over maybe a year actually because as soon as you buy a car it's worth less. House depreciates over well there's. You don't really buy houses business but a building depreciates over 20 years maybe. So whatever. There's a schedule of depreciation for different classes of things. And so the debate in the AI world is how long will this investment in Nvidia chips and server farms be valuable and AI moving so fast including the generations of Nvidia chips and you know, so you pack those into a server farm like they're doing down in Texas and a bunch of them all over the place. But the value of those Nvidia chips declines over time because there's next generation and Then there's competitors. And so to build foundational models, the value of a chip and its implementation in a server farm might not be worth depreciating over a very long period of time. That's a judgment call. Right, I know, because we don't really know where all this is headed. But normally chip factories and server farms depreciate over a fairly long period of time because there's a perception that the value is persistent. Over what? So pick a number, say 10 years. And, and that's the debate, right? So if they're spending what you say, $20 billion, but you said, you know, you said a much larger number.
Your first full transcript is free. After that, an email opens every transcript in the index — a list of readers we can write to, not a guest book.