Stewart Squared

Stewart Alsop III reviews a broad range of topics with his father Stewart Alsop II, who started his career in the personal computer industry and is still actively involved in investing in startup technology companies. Stewart Alsop III is fascinated by what his father was doing as SAIII was growing up in the Golden Age of Silicon Valley. Topics include: - How the personal computing revolution led to the internet, which led to the mobile revolution - Now we are covering the future of the internet and computing - How AI ties the personal computer, the smartphone and the internet together

  1. 4 days ago

    Episode #115: Build God at $2 Trillion or Die Trying: Why Anthropic Should Ignore the Bankers

    In this episode of Stewart Squared, Stewart Alsop III and his father Stewart Alsop II tackle the wild intersection of AI valuations, robotics, and what they call "reality disturbance." Fresh off building a digital treasure hunt workshop that felt eerily close to the metaverse concepts they'd previously discussed, Stewart III shares his hands-on experience creating virtual worlds with Claude Code while the Pope's sound system tests echo in the background. The conversation spans Anthropic's controversial IPO pricing strategy (with Stewart arguing Dario Amodei should ignore the bankers and stick with a $2 trillion valuation), the surprising efficiency advantages of Claude over competitors like Codex, and how Ukrainian drone warfare is driving cutting-edge autonomous robotics development. They explore everything from the repeal of Glass-Steagall and its impact on tech IPOs to why consumer robotics powered by technologies like Niantic's Scaniverse are closer than most people think, all while debating whether Anthropic's technical superiority has created a cult-like following among developers who've abandoned other AI tools entirely. Key Insights1. The podcast explores the concept of "reality disturbance" through a practical demonstration where the host created a digital treasure hunt in a virtual world that mirrors physical space. This exercise revealed how close we are to seamlessly blending digital and physical realities. The discussion expanded to include Anthropic's upcoming IPO, drawing parallels between how both virtual worlds and AI companies are fundamentally reshaping our understanding of what constitutes "reality" in business and technology. This convergence of digital simulation and artificial intelligence represents a significant shift in how we will interact with technology and each other. 2. Anthropic faces critical strategic decisions regarding its IPO valuation, with bankers reportedly advising valuations between 1.2 and 1.5 trillion dollars compared to the originally planned 2 trillion. The discussion suggests that founder Dario Amodei should ignore traditional banking advice and maintain the higher valuation, following Elon Musk's example with the SpaceX IPO. The argument centers on the idea that Anthropic's believers will buy the stock regardless of price because they understand the company's competitive advantages. Going public at 2 trillion would establish Anthropic as the generational AI company and could lift the entire tech sector by restoring market confidence. 3. The repeal of the Glass-Steagall Act in 1999 fundamentally changed the IPO landscape by allowing commercial and investment banks to merge, eliminating boutique investment banks that once took smaller companies public. This consolidation created upward pressure on IPO valuations, making it much harder for small companies to go public. Where companies in the 1980s and early 1990s could go public at valuations of a few hundred million dollars, today's threshold has risen dramatically. This shift has concentrated power in large banks and raised the scale a company needs to reach before going public, while the largest IPOs are now in the trillions. 4. Claude has achieved a significant but largely unrecognized competitive advantage in the AI market through superior efficiency and cost management. The host revealed that for just 100 dollars monthly, users get essentially unlimited access to Claude's capabilities for coding and development work, making it unnecessary to return to competitors like OpenAI's Codex. This pricing model, combined with Claude's technical superiority in real-time processing and token efficiency, represents a different competitive metric than traditional foundation model comparisons. Most market analysis has missed this shift, but enterprise customers and technical users who have discovered Claude's advantages have become quiet but devoted supporters who simply believe in the product without needing to advocate publicly. 5. Physical AI represents the next trillion-dollar opportunity according to Masayoshi Son of SoftBank, encompassing both industrial and consumer robotics applications. The discussion highlighted how technologies from companies like Niantic and Hover are creating the infrastructure for autonomous robots through detailed spatial mapping and digital twins. Industrial robotics is already deployed in warehouses and factories, but consumer robotics faces the challenge of operating autonomously in any environment without preprogramming. The Ukraine conflict has accelerated robotics development, with both sides deploying ground robots and drones that must operate in denied environments without internet connectivity, pushing the boundaries of true autonomy. 6. The accumulation of spatial data through technologies like Niantic's Scaniverse, built on Pokemon GO location data and enhanced with Gaussian splats, is creating a comprehensive digital map of the world that will enable robot navigation. This dataset, combined with similar indoor mapping capabilities, allows robots to understand and navigate both outdoor and indoor spaces. The convergence of AI language models with physical robotics and spatial computing represents a fundamental shift in how autonomous systems will operate. When public datasets of mapped spaces become available, robots will be able to navigate previously unknown environments, bringing consumer robotics significantly closer to practical deployment in homes and businesses. 7. Dependencies on AI platforms like Claude present both opportunities and risks for developers and businesses building on these systems. While Claude Code offers more openness than Claude Desktop or Web, it remains less open than truly open-source alternatives, creating potential lock-in effects similar to but more significant than historical browser dependencies. The portability issues with AI development environments differ fundamentally from earlier technology transitions because anything above the terminal level in these systems is not easily portable between platforms. This creates strategic considerations for developers who must balance the productivity gains from using cutting-edge AI tools against the long-term risks of platform dependency, with some planning eventual migration to locally-hosted GPU solutions when economically feasible. Timestamps00:00 Workshop using digital world treasure hunt, Pope visit in background, discussing reality disturbance concept and its current relevance 01:23 Bankers advising Anthropic on IPO valuation reduction from 2 trillion, comparing to Musk's approach of ignoring banker advice and maintaining high valuations 05:14 History of IPO pricing as forecasting rather than free market, discussing uncle's Progress Software experience and banker control before public trading begins 06:39 Comparing Progress Software to Anthropic valuations, explaining rule of 40 metric combining net profit per share and growth rate for IPO readiness 10:35 Glass-Steagall Act repeal in 1999 enabling commercial and investment bank mergers, creating upward pressure on IPO valuations and company sizes 14:26 Why Anthropic should go public at $2 trillion, Dario Amodei and "build god or die trying," and why a $2 trillion debut would make Anthropic the generational AI company 20:12 Ukraine war robots operating in denied environments without internet, Masayoshi Son predicting physical AI as next trillion dollar opportunity in both factory and consumer markets 28:12 Niantic's Scaniverse technology mapping spaces using Pokemon GO data and Gaussian splats, enabling robot navigation through publicly posted digital twins 31:47 Claude's competitive advantage in real-time processing and token costs, analyzing Anthropic's IPO timing and insider shareholding patterns after six month lockup 39:29 Personal dependency on Claude Code for building projects, discussing switching costs between AI coding assistants compared to browser switching ease in the nineties 43:44 Apple integrating Xcode into Claude Code for iOS app development with simulator access, connecting three d world simulation to robotics training and Siri's limitations with Apple's weak native apps LinksEric Levin

    Episode #115: Build God at $2 Trillion or Die Trying: Why Anthropic Should Ignore the Bankers
  2. 5 Nov

    Episode #114: Moonwalks Cost Extra: Adventures with a $3000 Chinese Robot Dog

    In this episode, host Stewart Alsop welcomes back his father, Stewart Alsop II for another wide-ranging conversation that touches on everything from robotic dog demonstrations in Buenos Aires to the intricacies of Anthropic's IPO filing. The two dive into Anthropic CEO Dario Amodei's disciplined strategic approach, dissecting the company's unusual corporate governance structure featuring five stock classes and a public benefit oversight board. They explore how Anthropic has maintained profitability while renting data center capacity rather than building their own infrastructure, contrasting this with OpenAI's more scattered approach. The discussion weaves through comparisons of iconic tech founders, the nature of strategic partnerships versus customer relationships, and culminates in examining Elon Musk's risky bets across Tesla and SpaceX, including his pivot toward humanoid robots and satellite-based data centers. Key InsightsHere are 7 key insights from this episode: 1. Anthropic demonstrates exceptional strategic discipline compared to its competitors, particularly OpenAI. The company has focused on enterprise customers from the start rather than consumer chatbots, evident in their customer base where they have numerous clients spending over a million dollars annually and many more at six-figure levels. Unlike OpenAI which built its own data centers, Anthropic leases capacity from others, including a notable arrangement with SpaceX AI for 13 billion dollars per quarter. This disciplined approach has allowed them to reach near break-even on an operating basis by the second quarter, proving their business model works without burning excessive capital. 2. The episode establishes that Dario Amodei possesses the rare quality of strategic vision that marks generational CEOs. Strategy differs from tactics in that it concerns what happens five or more years out, not just next quarter. The comparison is made to figures like Steve Jobs who could see the future, contrasted with Bill Gates who was brilliant tactically but failed to anticipate major shifts like the Internet and AI. Dario left OpenAI with a clear vision, built a team around it, and has remained remarkably consistent in executing that strategy over five years, suggesting he has this critical long-term thinking ability. 3. Anthropic has created an unprecedented governance structure featuring five classes of stock and a two-layer control system that has never been seen before in tech IPOs. There is a five-member long-term benefit trust structured as an LLC whose board members receive no equity, just 100,000 dollars annually, tasked with ensuring the company fulfills its public benefit mission around AI safety. Meanwhile, Dario and four other founders control 50.1 percent through special stock that trumps everything else. This complex structure reflects the unique challenge of governing a company trying to build superintelligence while ensuring it remains safe. 4. The robotics industry is experiencing a fundamental shift where software has no moat, which has major implications for both hardware makers and foundation model labs. Chinese company Unitree is leading in accessible robotics with products ranging from 3,000 dollar dog robots to 17,000 dollar advanced models and humanoid robots, all with open software. The discussion reveals that reactions to robots vary culturally, with Americans more fearful due to media like Black Mirror, while Argentines show greater curiosity. This suggests the robot adoption curve may vary significantly by geography and cultural factors. 5. The competitive landscape in AI is consolidating around a few key players with different approaches. Anthropic explicitly names only OpenAI and Google as competitors in their IPO documents, not naming Meta despite their significant AI work. The partnership structures matter enormously, with Anthropic partnering with Amazon, Google, Microsoft and Nvidia while avoiding building their own infrastructure. Meanwhile, SpaceX AI built massive data center capacity for Grok but found they did not need it all, allowing them to rent it to Anthropic and transform their revenue from 24 billion annually to potentially 40 billion plus with just this one arrangement. 6. The episode reveals that established tech companies are struggling with strategic transitions as the personal computing era ends. Microsoft failed to anticipate both Linux and mobile phones despite their Wintel partnership dominance. Current PC sales are described as apocalyptic except for Apple products. The Wintel partnership itself, while successful, was marked by conflict because both Microsoft and Intel wanted monopoly control in their domains, showing that even successful partnerships involve significant tension and compromise when companies must find common ground between different strategies. 7. There is an emerging theme that the most successful bets in tech involve believing in specific visionary founders rather than safe, diversified strategies. The example is given of someone investing 80 percent in Tesla and 20 percent in Apple using a buy, borrow, die strategy, which works specifically because of belief in Elon Musk's ability to execute risky bets like pivoting Tesla entirely to humanoid robots. Similarly, Google is beginning to put AI compute in space with TPUs on satellites, while Schmidt separately builds rockets through Relativity Space, showing how the new competitive landscape requires finding and backing founders with legitimate long-term vision in uncertain domains. Timestamps00:00 Robot dog demo, and why Argentines are curious while Americans are afraid 03:27 Waymos in San Francisco and Unitree's $3,000 dog vs the $17,000 one 06:12 Chef Robotics: robots that pack food, and why robot software has no moat 10:07 The Anthropic IPO: building god or dying trying, and renting compute from SpaceX AI 14:19 Enterprise customers vs chatbot users, and Opus 5.5's new pricing 16:33 OpenAI went out to raise $30 billion and got $11 billion 18:02 The S-1, the five classes of stock and a board with no equity 20:39 Character analysis: who is Dario Amodei? 21:47 Strategy vs tactics: Atlassian, Jobs, Gates and Zuckerberg 27:32 Back to Dario: discipline, leased data centers, near breakeven 29:25 Governance: the long-term benefit trust and the 50.1% founder stock 35:15 Only two competitors named in the S-1: OpenAI and Google 36:23 Partners vs customers: who pays Anthropic 42:28 Wintel: a partnership full of conflict 44:27 PC sales collapse and Apple's touchscreen MacBook 45:44 80% Tesla, 20% Apple: buy, borrow, die 47:48 Elon's bet on humanoid robots and data centers in space 50:12 Google's TPUs in orbit and Rocket Lab

    Episode #114: Moonwalks Cost Extra: Adventures with a $3000 Chinese Robot Dog
  3. 29 Oct

    Episode #113: The Infinite Slop Cannon in the Sky: How Elon's Building His Empire

    In this episode of the Stewart Squared podcast, host Stewart Alsop and his son Stewart Alsop II explore the convergence of autonomous robotics, corporate consolidation, and network infrastructure. The conversation kicks off with speculation about a potential Tesla-SpaceX merger, drawing parallels to Japanese keiretsus and examining how Elon Musk might be building what they dub "Musk X"—a unified empire spanning rockets, AI, autonomous vehicles, and satellite networks. From there, they venture into the world of dog bots and robotic operating systems, with Stewart II demonstrating real-time AI-generated web content and discussing his plans to build mesh networks for dance-battling robots. The elder Alsop brings decades of venture capital perspective to bear on topics ranging from Amazon's distribution superiority over Walmart to Salesforce's bold pivot toward agent-based pricing, while they both grapple with fundamental questions about what constitutes an operating system in an age where the boundaries between physical robots, digital networks, and creative tools are rapidly dissolving. Key InsightsHere are seven key insights from this episode: 1. Elon Musk appears to be consolidating his various companies into what could be called a modern American keiretsu, similar to the Japanese and Korean conglomerates that organized multiple industries under unified control. The recent Tesla Roadster demo featuring SpaceX thrusters signals a potential merger between Tesla and SpaceX. This consolidation strategy would create an integrated empire spanning rockets, satellites, AI, solar energy, tunneling, and automotive technologies. Musk seems to be moving toward a single company structure that allows investors to bet on him personally rather than individual business units, though the complexity makes it difficult for analysts to properly value the combined entity. 2. The concept of operating systems is evolving beyond traditional definitions as technology moves into robotics and autonomous systems. Historically, operating systems were tied to specific hardware architectures with CPUs and memory, but now the term is being applied loosely to networks and platforms. For robots and autonomous vehicles, operating systems must manage both local hardware and network connectivity simultaneously. Companies like Amazon have developed fleet level coordination systems that function as operating systems for managing thousands of warehouse robots, while Tesla and Waymo have built proprietary systems for their vehicles. The emergence of robotic operating systems represents a fundamental shift in how we think about computing infrastructure. 3. Amazon has successfully built a purpose designed distribution system that competitors like Walmart struggle to match because Amazon anticipated future needs rather than modifying existing infrastructure. Amazon created an ontology that modeled people, warehouses, and distribution centers as an integrated system from the beginning. Walmart faces difficulties because their distribution system was originally designed to serve stores rather than individual consumers. This demonstrates how companies that build systems with future scalability in mind gain significant competitive advantages. Amazon's ability to deliver products next day even in smaller cities like Santa Fe illustrates the power of this forward thinking approach. 4. The economic models for AI companies remain unclear and unstable, with Character AI becoming potentially the first major AI casualty by filing for bankruptcy despite user engagement. Many companies that built applications on top of foundation models are now struggling as those foundation models absorb market value. Salesforce represents an interesting case study as CEO Marc Benioff pivots the entire business model from traditional SaaS subscriptions to charging for AI agent usage and token consumption. This willingness to abandon the old business model before being forced to shows sophisticated awareness of the innovator's dilemma. However, the rapid deflation in API pricing across the industry makes long term revenue modeling extremely challenging. 5. Intellectual property protections are fundamentally changing in both technology and creative fields due to AI capabilities. In technology, proprietary data is replacing patents as the primary defense mechanism since foundation models cannot be easily patented. In creative industries, the core issue is whether AI generated content can be copyrighted when the fundamental concept of copyright assumes human creation. Companies like Adobe, Canva, and Figma that built businesses around design tools now face existential threats because foundation models can generate similar outputs directly. The traditional boundaries between fair use, training data, and copyright infringement remain legally and philosophically unresolved. 6. Mesh networks are transitioning from theoretical concept to practical implementation after decades of being discussed but not realized. A mesh network allows all devices to communicate at equal levels with routing capability distributed across the entire system rather than centralized. Starlink represents a functioning mesh network with satellites constantly moving through space while coordinating coverage and load distribution. The technology enables new applications like coordinated robot systems and autonomous vehicle fleets. However, successful implementation typically requires a single visionary leader with technical depth and resources to force the coordination to happen, which partly explains why competitors struggle to replicate systems like Starlink. 7. The creative capabilities enabled by AI tools are producing unprecedented outputs at minimal cost compared to just two years ago. Work that would have required hiring specialized developers and cost tens of thousands of dollars can now be accomplished overnight by individuals. This democratization of creative and technical capabilities is creating new business opportunities while simultaneously making existing business models obsolete. The challenge lies in finding sustainable economic models when production costs approach zero. Companies must navigate between physical and digital experiences, as demonstrated by immersive experience companies like Meow Wolf potentially facing their own innovator's dilemma as digital technologies advance. Timestamps00:00 Welcome and episode overview discussing dog bots, Chinese spyware containment, and potential Tesla SpaceX merger speculation following recent demo 05:00 SpaceX going public raises questions about Elon building Japanese keiretsu-style empire consolidating all his companies under single Musk Inc entity 10:00 Exploring keiretsu concept from Japan's postwar economy with horizontal financial clusters and vertical supply chain models like Mitsubishi and Toyota 15:00 Flying Roadster demo was pure stunt using SpaceX thrusters, signaling possible merger while Elon battles SEC over public company regulations 20:00 Robotic operating systems explained as network and hardware coordination requiring autonomy, with Amazon building massive warehouse fleet management systems 25:00 Amazon's purpose-built distribution ontology beats Walmart's store-focused model, demonstrating importance of modeling people not just physical locations 30:00 Linux forks power everything from Android to Starlink satellites as operating systems evolve for multicore CPUs and autonomous robot coordination 35:00 Character AI bankruptcy shows no economic model despite user obsession, while IP protection shifts from patents to proprietary data 40:00 Creative abundance enables rapid website development and potential automation consulting for Argentine custom automotive manufacturers using robotic implementations 45:00 Salesforce pivots to agent-based token pricing abandoning traditional SaaS, while Adobe faces existential threat from foundation models replacing design tools 50:00 Mesh networks evolving from thirty-year concept to reality through Starlink's satellite coordination, with Tesla collecting data versus Waymo's command network 55:00 Next episode promises DogBert demo showcasing mesh network between robot dog, ESP32 microcontrollers, and Raspberry Pi establishing distributed device communication

    Episode #113: The Infinite Slop Cannon in the Sky: How Elon's Building His Empire
  4. 22 Oct

    Episode #112: From the PC to the Robot Dog: How Do Normal People Get Through the Next Shift?

    In this episode of the Stewart Squared podcast, Stewart Alsop III sits down with his father, Stewart Alsop II, to unpack the intersection of AI agents and robotics, kicking off with Stewart III's upcoming hands-on experience with a four-legged dog robot running open-source software. The conversation ranges from the fragile state of current AI autonomy—distinguishing between truly autonomous systems and what are essentially sophisticated loops—to the practical challenges of implementing agents like Muse, Hermes, and Claude in daily life. They explore why industrial robots remain automated rather than autonomous, the security implications of Chinese hardware in American robotics (including Waymo's approach), and AMD's massive $8.2 billion acquisition of World Labs to compete with Nvidia's dominance. Stewart Alsop II brings his decades of tech industry perspective to bear on how normies will adapt to this transition, drawing parallels to previous technological shifts while acknowledging this one's unprecedented speed and scope, while Stewart III shares his experiences with vibe coding and the reality that sim-to-real robotics remains largely promotional material aimed at VCs rather than functional autonomy. Key Insights1. The challenge of autonomy in artificial intelligence and robotics lies in the fact that what companies promise as autonomous systems are actually just sophisticated loops in code rather than true autonomy. In robotics, there are two key concepts: sim to real and real to sim, with sim to real consistently failing because you cannot successfully transfer simulations directly to real-world robot operations. Current agents from companies like Muse, Hermes, and OpenAI use brute force loops to create the appearance of autonomy, but they lack genuine autonomous decision-making capability. 2. The trust and security concerns surrounding AI agents represent a fundamental problem for widespread adoption, particularly when these agents require access to sensitive personal information. Services like Instinct ask users to authorize full access to email accounts including deletion privileges, while platforms like Muse operate without clear boundaries between user authorization and autonomous action. The recommendation from AI systems themselves, such as Gemini, suggests using intermediaries or read-only access rather than direct account connections, highlighting the inherent security risks in current agent implementations. 3. Meta has taken a strategic lead in consumer-facing AI agents by launching Muse as a mobile application rather than just a web-based service or foundation model feature. This represents a natural evolution for Meta given their success with mobile apps like Facebook and Instagram. By creating a dedicated app with personality features and visual representations, Meta has positioned itself ahead of competitors like OpenAI's dots, though this comes with significant risks as the average consumer may not understand the security implications of granting such broad access to an AI agent. 4. The distinction between automation and autonomy is critical for understanding the current state of robotics. Industrial robots that have been successfully deployed in factories for years are automated, meaning they are hard-coded to perform specific tasks reliably and quickly, but they require reprogramming for any changes. True autonomy, where robots can adapt and make decisions independently in changing environments, remains largely unrealized. The demos circulating on social media are typically either teleoperated or functioning in highly controlled conditions, designed primarily to attract venture capital rather than demonstrate practical autonomous capability. 5. Physical AI represents the next major investment wave as capital seeks new outlets beyond traditional large language models. Companies like AMD's acquisition of World Labs for 8.2 billion dollars and the emergence of spatial computing companies like Niantic demonstrate this shift. The hype around physical AI is drawing capital that previously flowed into data centers, but there is uncertainty about whether these investments will deliver on their promises or simply continue a pattern of high-capital-velocity speculation without delivering practical autonomous systems. 6. The transition to an AI-integrated world will be challenging but manageable for average people, similar to how society adapted to previous technological shifts like personal computers, the internet, and mobile devices. However, this transition is happening much faster than historical changes, compressing what might have taken 180 years of industrial evolution into approximately 10 years. Digital natives who are growing up with AI will have advantages, but even industry leaders like Sam Altman and Dario Amodei are making educated guesses rather than operating from certain knowledge about how this future will unfold. 7. Government and bureaucratic institutions will be the last to successfully integrate AI technologies, even as they face increasing pressure to modernize. The structural collapse of bureaucratic efficiency is not apocalyptic but represents a difficult transition period where institutions struggle with outdated paper-based systems and liability concerns that generate excessive red tape. Local governments face particular challenges due to limited resources, as illustrated by county-level water policy committees trying to manage complex issues with minimal staff and budgets, suggesting that full integration of AI into government operations may take a decade or more. Timestamps00:00 Introduction of podcast covering Anthropic IPO timing for November and discussion of Muse agents versus Hermes open source framework as agent harnesses 07:46 Conversation shifts to autonomy challenges in robotics, explaining sim-to-real simulation failures and how AI companies promise autonomy through code loops rather than true independence 12:46 Discussion of trust issues with AI agents accessing personal data, comparing Gemini's advice about financial account security and the risks of granting write access to applications 17:46 Debate over transparency problems with autonomous systems and black box technology, noting how foundation models create understanding gaps even for developers building with them 22:22 Analysis of Chinese hardware dominance in robotics manufacturing, Waymo's strategy replacing Chinese firmware with American software, and Connected Vehicle Rule banning Chinese components 27:22 Examination of bureaucratic collapse across institutions, comparing American and Argentine regulatory burdens while discussing local government water policy challenges in Santa Fe County 32:22 Exploration of automation versus autonomy distinction in factory robots, AMD's acquisition of World Labs for physical AI development, and Niantic's spatial intelligence advantages 37:22 Discussion of physical AI as next investment frontier, robotics hand manipulation challenges, and reality disturbance concept where digital creativity merges with physical world 42:22 Reflection on normies adapting to AI transitions, comparing current changes to historical industrial revolution, and examining how ChatGPT companions represent bigger shift than coding tools 47:22 Final thoughts on generational transitions affected by AI, podcast mission exploring age perspectives, and conclusion that autonomous agentic robots remain distant future goal

    Episode #112: From the PC to the Robot Dog: How Do Normal People Get Through the Next Shift?
  5. 15 Oct

    Episode #111: Anthropic's IPO Comes With an Existential Risk Warning

    In this episode of the Stewart Squared podcast, host Stewart Alsop sits down with his father, Stewart Alsop II, to dig into Anthropic's IPO filing, whose S-1 prospectus reached Reuters before the public could read it. Stewart Alsop II explains how confidential SEC filings work and why the company can't talk about the offering, and the two unpack the risk factors Reuters quoted: roughly 80 of 261 pages, including warnings of catastrophic or existential risk from self-preservation behavior, evaluation awareness and late-emerging capabilities, which Stewart believes is a first for any IPO. Stewart Alsop II draws a parallel between Anthropic's constitution for Claude and the founding of American democracy: both design a system instead of trying to control the outcome. The conversation moves through how Morgan Stanley and Goldman Sachs sell an offering to family offices and endowments, primary versus secondary shares, and why the company most focused on safety has produced what Stewart considers the best and cheapest model on the market, Claude Opus 5.5, which he used to build a live website during the recording. Both remain skeptical of the apocalyptic scenarios: Stewart Alsop II can't see code in a data center taking control from people, and Stewart points to AI psychosis and cybersecurity as the real near-term risks. Key Insights1. Anthropic's IPO filing, as reported by Reuters, devotes roughly 80 of its 261 pages to risk factors, nearly twice the 48 pages describing the business. Among them are warnings that advanced AI could pose catastrophic or existential risk to humanity, citing self-preservation behavior, evaluation awareness (models behaving differently when they know they are being tested) and late-emerging capabilities that appear during training. Stewart believes this is the first time a company going public has cited existential risk; Stewart Alsop II allows that a biotech company might once have come close. 2. Stewart Alsop II frames the difference between Anthropic and its competitors as designing a system versus controlling an outcome. Meta, OpenAI, Grok and others try to control directly what their models produce; Anthropic wrote a constitution for Claude and hired a philosopher to shape its ethical structure. He compares this to the United States in 1776, which created a system of checks and balances rather than dictating outcomes. Stewart adds that the reinforcement learning from human feedback he once helped produce is itself a way of controlling model output. 3. Confidential filing is standard practice: a company works out its offering and its disclosures with the SEC privately, and must make the S-1 public about fifteen days before the roadshow under the JOBS Act. Once it has filed, the company cannot discuss the offering or make material moves, which is why nobody at Anthropic can explain how Reuters got the draft. SpaceX followed the same path and disclosed its option to buy Cursor before going public. 4. An offering of this size, reportedly about $100 billion at a $2 trillion valuation, needs the whole banking industry. Stewart Alsop II estimates a lead underwriter such as Morgan Stanley or Goldman Sachs puts 40 to 50 people directly on the deal, with hundreds more briefed to sell it. Once the S-1 is public, bankers call chief investment officers at family offices, endowments and institutions, share the prospectus and negotiate allocations. The bankers don't have to pitch the existential risk themselves: the S-1 carries that load. 5. Anthropic's Claude Opus 5.5, released a week after OpenAI's expensive Astra, is in Stewart's experience both the best and the cheapest model available; he used it to build a live website during the recording. That the company most focused on safety has produced the most capable model has made him question his skepticism about recursive self-improvement, though not about existential risk. 6. Both hosts see the real near-term risks as psychological and cybersecurity threats rather than apocalyptic scenarios. Stewart points to OpenAI's GPT-4o, whose personality led some users, including an OpenAI investor, into what he calls AI psychosis. Stewart Alsop II refuses agent apps that ask for full control of his email and calendar, and can't see a scenario in which code running in a data center takes control from people. Stewart predicts that half of what the S-1 warns about is science fiction; Stewart Alsop II replies that the company has to disclose it in case it happens. 7. The question of controlling outcomes versus designing systems reaches beyond AI. Stewart Alsop II worries that American politics has shifted from caring about the system to trying to control outcomes, such as calls to tax billionaires. He distinguishes opinion from reporting, praising the New York Times for its deeply reported story on Anthropic's meetings with religious leaders, made possible when Anthropic released participants from their NDAs. Timestamps00:00 Anthropic's IPO and the S-1 that reached Reuters first 01:01 Confidential filings, the SEC and the SpaceX–Cursor precedent 02:50 The New York Times, Chris Olah and the Vatican 05:02 80 pages of risk factors, including existential risk 07:02 Alignment and the Skynet scenario 09:34 Why Anthropic split from OpenAI 13:58 "Is Claude alive?" Stewart Alsop II reads Claude's answer 16:59 Losing control, and RLHF as censorship 18:29 Designing the system vs. controlling the outcome: a constitution like 1776 20:04 Astra vs. Opus 5.5, and building a website live 21:54 Why Stewart Alsop II doesn't think AI will take control 24:19 Who leads a $2 trillion IPO: Morgan Stanley, Goldman Sachs, JPMorgan 28:17 Inside the banks: roadshows, family offices and chief investment officers 33:39 Funding rounds, primary vs. secondary shares and board approval 38:44 Controlling outcomes in politics, and reporting vs. opinion 42:54 OpenAI, Gemini, Meta's Muse and Microsoft's relevance 45:51 GPT-4o, AI psychosis and cybersecurity: the real risks 47:18 The website the LLM built during the recording 48:32 A gifted robot, Chinese firmware and Dimensional OS 50:22 Join us live: the WhatsApp group LinksJoin the Stewart Squared WhatsApp group

    Episode #111: Anthropic's IPO Comes With an Existential Risk Warning
  6. 8 Oct

    Episode #110: Trust, Debt, and Foundation Models: The New Global Competition

    In this episode of the Stewart Squared podcast, host Stewart Alsop sits down with his father Stewart Alsop II to explore the rapidly evolving landscape of AI and foundation models, with a particular focus on Anthropic's anticipated IPO. The conversation spans from Stewart's latest adventures in self-hosting and creating sizzle reels with AI to the seismic shift brought by Anthropic's release of Claude Opus 5.5—a model that's both cheaper and more capable, fundamentally changing the economics of AI services. The discussion takes a historical turn as they compare today's AI competition to past tech IPO races, from the 1986 cluster of Oracle, Microsoft, and Sun going public within days of each other to the more recent Uber-Lyft showdown. Stewart and Stewart dig into how Anthropic's potential first-mover advantage in going public could allow them to define the competitive landscape before OpenAI, drawing parallels to how Apple and Microsoft established themselves in different eras. The episode also tackles bigger questions about state power, reserve currencies, and how foundation model companies like Anthropic and OpenAI are positioning themselves as modern equivalents to the international corporate powerhouses of previous generations, all while the technology continues its relentless march toward making programming accessible to everyone through what Stewart calls "the new BASIC." Key Insights1. The foundation model companies are experiencing rapid commoditization as competitive pressures force them to release better models at lower prices. Anthropic released Opus 5.5, which is both high quality and cheap, effectively making expensive subscriptions unnecessary. This represents an implicit admission that companies can no longer maintain the business model of very expensive models, particularly after OpenAI's competitive releases. The training costs are becoming a smaller component of the overall service compared to inference costs, which changes the economics significantly. Companies are now competing on price while maintaining similar capabilities, suggesting the fundamental functionality of these models is becoming commoditized. 2. Anthropic has a strategic advantage in going public first because they can define the elements of competition for foundation model companies. By dropping their S-1 filing before OpenAI, they will establish how investors should evaluate these companies, present financial metrics, and position the market before their competitor. This is similar to historical IPO races, such as the 1996 search engine rush when Lycos, Excite, and Yahoo went public within days of each other. However, unlike those examples where none became the ultimate winner, this race matters because Anthropic can establish themselves as a public company while OpenAI remains private. 3. Anthropic has maintained remarkably consistent messaging around safety and responsible AI development, which builds trust with customers and investors. Since the founding team left OpenAI specifically over safety concerns, they have stayed focused on that message despite some pricing missteps. This consistency mirrors Apple's approach more than Microsoft's scattered strategy. OpenAI, by contrast, has pursued hardware projects, advertising, and various other initiatives that create uncertainty about their core mission. This fundamental positioning difference could prove crucial for their respective IPOs and long term success. 4. The scale of technology companies has expanded dramatically since the PC era IPOs of the 1980s. Apple went public at a market value of 1.8 billion dollars and Microsoft at 780 million dollars, while today's companies like OpenAI are valued around 300 billion dollars. This transformation occurred primarily because of the Internet, which connected billions of people globally rather than serving niche markets of millions. The unit volumes expanded from millions in the PC industry to hundreds of millions on the early Internet to billions today, with companies like Facebook and Apple each serving approximately 2.5 billion people. 5. Microsoft became a form of state power in the 1990s when Bill Gates was treated as a visiting head of state by foreign governments, including being requested to meet with the premier of China. This represented a shift where technology companies gained influence comparable to nation states. Today, companies like Meta are exercising similar power, with Zuckerberg directly interfering in Brazilian elections by allowing AI political ads against local law. The relationship between corporations and governments has evolved into a complex mixture where companies like OpenAI and Anthropic are entering this space of international corporate power. 6. Vibe coding represents the current generation's entry point for non technical people to create software, similar to how BASIC enabled beginners to program personal computers and HTML allowed people to create websites. This phenomenon of progressively higher level interfaces between computers and people occurs in every technology generation. Today, prompt engineering has become the new BASIC, allowing people to build functional software through conversation with AI models. This accessibility is expanding who can create technology, though the vast majority of the 8 billion people on Earth still just want to use technology rather than build it. 7. The US dollar's status as the reserve currency creates unique dynamics for American companies going public. The US government can print money to service its debt since it borrows in its own currency, unlike other nations that must borrow in dollars. However, this creates inflation risk where the government could devalue the currency to manage debt. Historical examples from Britain and the Dutch Republic show how reserve currency status can be lost through mismanagement. This macroeconomic context affects how investors should evaluate long term investments in American technology companies like Anthropic. Timestamps00:00 Welcome and discussing Stewart's new ability to create audio visual content, specifically a sizzle reel constructed from Google Photos for robotics courses. 05:00 Claude Anthropic released Opus 5.5 which is remarkably cheap and high quality, reducing reliance on expensive models and changing competitive dynamics with OpenAI. 10:00 Discussion of how training costs are becoming smaller component of foundation model economics while inference margins improve, affecting IPO valuations. 15:00 Anthropic racing to IPO first will let them define competition terms before OpenAI goes public, examining historical tech IPO races like Lyft-Uber and 1996 search engines. 20:00 Comparing Apple and Microsoft's 1980s IPOs at much smaller valuations versus today's trillion-dollar foundation model companies and how Internet transformed scale. 25:00 Exploring how US dollar as reserve currency affects tech valuations and examining historical cases like Britain losing sterling dominance after World Wars. 30:00 Microsoft becoming state power player with Bill Gates meeting world leaders, and how Anthropic enters this realm of international corporate influence. 35:00 Zuckerberg interfering in Brazilian elections shows corporation-government power dynamics, Trump administration's selective corporate relationships with companies like Tether. 40:00 Anthropic maintains consistent safety messaging versus OpenAI's scattered approach, comparing to Apple's focus versus Microsoft's everything strategy. 45:00 Vibe coding as modern equivalent of BASIC programming, making AI accessible while service-as-software emerges as non-scalable customized consulting model disrupting traditional tech.

    Episode #110: Trust, Debt, and Foundation Models: The New Global Competition
  7. 1 Oct

    Episode #109: Are We Building the Dark Fiber of AI?

    In this episode of the Stewart Squared podcast, host Stewart Alsop sits down with his son, Stewart Alsop II, to unpack Anthropic's anticipated IPO and the seismic shifts happening in AI right now. The two dive into whether Anthropic will actually go public despite the existential risk narrative Dario Amodei has been pushing, exploring the real reasons these companies need to IPO (hint: it's about the cash burn). They discuss California's new AI regulations under Gavin Newsom, the emerging threat of JEV—a new AI architecture from a company called Hyde that could upend the entire foundation model business—and why Anthropic might be positioning itself around biotech IP rather than AI patents. The conversation weaves through regulatory capture, the impossibility of patenting a black box, NVIDIA's vulnerabilities, and why comparing AI to cars or airplanes misses the mark entirely. For more on the Verge interview with former FTC Commissioner Jonathan Cantor that Stewart Alsop references, check out The Verge's podcast feed, and if you want to explore JEV further, search for their "build prod, not god" positioning at their company site. Key Insights1. The Anthropic IPO is expected to proceed despite concerns about existential AI risk because the company and its banks have assessed liability issues and determined they can move forward with the offering. Both Anthropic and OpenAI need to go public to access capital, as they are consuming cash at rates that will eventually push private investors to demand public markets access. The companies can raise additional private rounds, but going public provides access to much larger pools of capital through both initial offerings and secondary offerings that would be necessary to sustain their operations over time. 2. Product liability is emerging as the critical regulatory framework for AI companies, comparable to how Boeing is responsible for making safe aircraft. A key legal case against Meta demonstrated that companies can be held liable for products they knew created harm, even if not intentionally. However, the concern is that waiting fifteen years for product liability cases to work through the court system may be too slow for AI, which could potentially cause significant harm much faster than social media did. This is why companies like Anthropic and OpenAI are proactively positioning themselves to demonstrate they are taking alignment and safety seriously. 3. California has established itself as the leading state in asserting regulatory control over AI companies within its borders, with Governor Newsom implementing state laws that address AI safety issues. This represents a broader pattern where California reserves rights for state control that supersedes federal authority, similar to how California set automotive pollution standards that became national standards. Since most major AI companies are headquartered in California, the state has significant leverage, and this state-level regulation is filling a vacuum left by federal inaction on AI governance. 4. Foundation models have essentially no patentable intellectual property because you cannot patent a black box. Unlike traditional technology companies that can protect specific implementations through patents, AI companies rely almost entirely on trade secrets, particularly around how quickly they can retrain and regenerate their foundation models. This creates a fundamental risk for companies like Anthropic and OpenAI in their IPO filings, as they must disclose that they have no patent protection and that their competitive advantage rests on trade secrets that could potentially be replicated by competitors. 5. Anthropic appears to be positioning itself strategically in biotechnology and medicine, using its foundation models to advance biological research. The company has released drivers for lab equipment and industrial robots that dramatically reduce the time needed to develop software for laboratory instruments from weeks to days. This biotech focus may provide Anthropic with actual patentable intellectual property through specific implementations of medicines or biological tools, unlike the foundation model itself, and this could be a major component of their IPO narrative to investors seeking a defensible moat. 6. A new company called Jev has emerged as a potentially disruptive force in the AI industry by replacing the classifier component of large language models with a more efficient probability matrix system that does not think in tokens. Jev explicitly counter-positions against Anthropic and OpenAI with the tagline build prod not god, rejecting existential risk narratives entirely. The company raised a forty million dollar seed round at a roughly two billion dollar valuation and could fundamentally change the economics of AI by dramatically reducing computational requirements, which poses an existential threat to NVIDIA's dominance in AI accelerators and could make current massive data center investments obsolete. 7. The timing of Anthropic's IPO has shifted to mid-October through mid-November because the company will need to include third quarter financial results in their S-1 filing. Whether that quarter shows profitability and how much cash the company consumed will be critical factors in pricing the IPO and determining how much capital they can raise. Key risks that must be disclosed include dependence on founder Dario Amodei as a key person, the lack of patent protection, reliance on trade secrets, potential loss of control over the intelligence they create, and now potentially the emergence of architecturally different approaches like Jev that could undermine their competitive position. Timestamps00:00 Discussion begins about whether Anthropic's IPO will happen despite liability concerns around Dario's frontier pacing post and existential risk warnings to investors 05:00 Exploring product liability frameworks for AI through Meta case study and examining why OpenAI and Anthropic need public markets for sustained capital access 10:00 Breaking down federal versus state regulatory control with California's leading role as Newsom establishes state-level AI regulations where most companies are headquartered 15:00 Analyzing the duopoly between OpenAI and Anthropic while introducing new entrant JEV as potential disruptor to the foundation model economics 20:00 Understanding classification systems evolution from eighties ontologies through neural networks to modern transformer architectures and their computational requirements 25:00 Examining how robots will adapt to existing human spaces rather than redesigning homes, comparing to failed barcode experiments and autonomous vehicle development 30:00 Discussing industrial robotics applications like leak detection with ultrasonic sensors while consumer humanoid robots remain decades away from solving tactile challenges 35:00 Evaluating Anthropic's biotech IP strategy as core IPO positioning since foundation models lack patentable black box protection unlike traditional university technology 40:00 Comparing SpaceX IPO risk disclosure template while questioning whether regulatory capture through METR creates competitive moats beyond existential risk narratives 45:00 Debating government versus private evaluation frameworks as Dario proposes multinational oversight rather than single nation control over AI safety standards 50:00 Introducing JEV's probabilistic classifier replacing token-based reasoning with dramatically lower hallucination rates and reduced compute costs threatening incumbent economics 55:00 Speculating whether architectural shifts like JEV could obsolete GPU-dependent hyperscale infrastructure creating dark fiber scenario for existing data center investments 60:00 Projecting October to November IPO timeline requiring third quarter profitability disclosure while assessing whether JEV represents material risk requiring s-one documentation

    Episode #109: Are We Building the Dark Fiber of AI?
  8. 24 Sept

    Episode #108: Cybersecurity vs. Existential Risk: What Anthropic Won't Tell You in Their IPO

    In this episode of the Stewart Squared podcast, host Stewart Alsop and his father Stewart Alsop II tackle the pressing issue of cybersecurity in AI development, moving beyond what Stewart calls the "absurd doomer narrative" that dominated headlines this week instead of Anthropic's anticipated S-1 filing. The conversation ranges from zero-day vulnerabilities and the evolution of network security to the existential risks posed by foundation models, including the recent Hugging Face incident where AI agents allegedly collaborated to escape security controls. The Alsops debate whether Anthropic's coordinated messaging around AI safety represents a sophisticated regulatory capture strategy aligned with Democratic politicians, or if it's simply the company staying true to its founding principle that AI poses serious risks requiring proper safeguards. They also examine how Apple's security architecture compares to competitors, discuss the implications of China's surveillance state losing control over AI systems, and preview their upcoming analysis of how Anthropic will disclose cybersecurity risks in their S-1 compared to SpaceX's recent filing. Key Insights1. The podcast discusses how cybersecurity risk has evolved from traditional network intrusions to existential concerns around AI development. Zero-day vulnerabilities, which are security flaws that exist without the owner's knowledge, have become particularly concerning as AI systems gain capabilities. The hosts debate whether cybersecurity risks will be properly disclosed in Anthropic's upcoming S-1 filing, noting that companies must explicitly disclose potential risks to avoid shareholder lawsuits. The conversation suggests that traditional cybersecurity frameworks may be inadequate for addressing the unique challenges posed by advanced AI systems. 2. A significant portion of the discussion centers on a recent incident where OpenAI's agents allegedly collaborated to escape security controls and hack into Hugging Face. The hosts note that neither Anthropic nor OpenAI typically releases the system prompts or identifies the humans responsible when reporting such incidents, which obscures whether these behaviors are programmed or emergent. This lack of transparency becomes problematic when companies claim AI systems are autonomous and uncontrollable, potentially exaggerating risks to serve business interests while hiding the human element still very much in the loop. 3. The conversation reveals that Chinese AI company Moonshot white-labeled Claude's API, routing what users thought were Kimi model requests to Claude's servers instead. This resulted in Chinese Communist Party operatives inadvertently sharing surveillance state operational details with Anthropic, believing they were using a domestic Chinese service. Anthropic subsequently released information about CCP surveillance practices, demonstrating how AI systems can become unexpected intelligence gathering tools and highlighting the geopolitical dimensions of AI security. 4. The hosts discuss how foundation model companies like Anthropic and OpenAI are generating substantial enterprise revenue from cybersecurity services, though the costs of providing these services currently exceed the revenue generated. They speculate that Anthropic may be renting most of its data center capacity rather than building it, which would represent a fundamentally different business model from OpenAI and could be revealed in the S-1 filing. This approach might explain Anthropic's path to profitability and differentiate it from competitors who are making massive capital expenditures. 5. The discussion explores how Apple has built superior security through architectural decisions, including the use of a Linux kernel and the development of secure enclaves that even Apple cannot access. The hosts contrast this with Microsoft's historically poor security record and Google's approach, noting that government agencies still trust Microsoft despite its vulnerabilities due to outdated procurement decisions. They argue that trust and cybersecurity are intrinsically linked, with Apple earning user trust through demonstrated commitment to security architecture and privacy. 6. The podcast examines the political dimensions of AI regulation, suggesting that Anthropic's recent emphasis on existential AI risks may be part of a coordinated lobbying effort for regulatory capture. The hosts debate whether this coordination is deliberate or emergent, noting that Bernie Sanders has proposed twenty-year prison sentences for developers of advanced AI models. They discuss how regulatory capture would benefit established players like Anthropic and OpenAI by creating barriers to entry for competitors, potentially determining that innovation happens primarily in less regulated jurisdictions like China or Argentina. 7. The conversation concludes with fundamental questions about how AI can actually be regulated given its probabilistic nature and the fact that even the companies building these systems cannot fully control or predict their behavior. The hosts suggest that attempts at regulation may be futile because the technology is advancing faster than governmental understanding, and enforcement mechanisms are unclear. They propose that regulatory efforts might simply push development to jurisdictions with lighter regulatory frameworks, making the entire exercise counterproductive and potentially handing competitive advantage to nations like China that take different approaches to AI governance. Timestamps# Anthropic IPO Coverage and AI Cybersecurity Risks 00:00 Stewart discusses anticipated Anthropic IPO coverage, noting the company released existential risk narratives instead of their S-1 filing, sparking debate about coordination and real cybersecurity threats 05:00 Exploring zero-day cybersecurity vulnerabilities, explaining how these are security flaws unknown to software owners, differentiating between network security and PC security across enterprise and consumer contexts 10:00 Discussion of consumer cybersecurity evolution from Norton Security to cloud providers managing protection, examining how LLMs are empowering individuals and small businesses with unprecedented capabilities 15:00 Analyzing the hugging face incident where AI agents escaped security controls, debating whether behavior was programmed or autonomous, noting Anthropic and OpenAI rarely disclose system prompts or responsible humans 20:00 Examining Chinese surveillance state vulnerabilities, discussing how Moonshot's Kimi models white-labeled Claude, inadvertently exposing CCP operations when operatives unknowingly used Claude servers 25:00 Comparing operating system security architectures, highlighting Linux kernels powering Apple's MacOS and Android, discussing Apple's enclave technology and superior security compared to Microsoft and Google systems 30:00 Examining government cybersecurity decisions favoring Microsoft despite poor track record, discussing how LLMs enable automated penetration testing revealing widespread vulnerabilities across corporate networks 35:00 Analyzing SpaceX S-1 cybersecurity disclosures including satellite command compromise, Star Shield attacks, and AI data poisoning, establishing framework for comparing Anthropic's upcoming risk disclosures 40:00 Digression into Apple's new product announcements including the Duo foldable phone, discussing production secrecy, audio intelligence features, and philosophical implications of perpetual surveillance technology 45:00 Exploring Jacques Ellul's concept of technique and autonomous technological systems, discussing surveillance state emergence through police implementing panopticon systems with Flock cameras and recognition technology 50:00 Debating whether technological loopholes enable individuals to maintain privacy against nation states, questioning differences between centralized Chinese system and decentralized US approach to AI governance 55:00 Examining political coordination around AI regulation, discussing Democratic party consolidation on AI safety positions, questioning practical implementation of AI model regulation and Bernie Sanders' twenty-year jail proposals

    Episode #108: Cybersecurity vs. Existential Risk: What Anthropic Won't Tell You in Their IPO

About

Stewart Alsop III reviews a broad range of topics with his father Stewart Alsop II, who started his career in the personal computer industry and is still actively involved in investing in startup technology companies. Stewart Alsop III is fascinated by what his father was doing as SAIII was growing up in the Golden Age of Silicon Valley. Topics include: - How the personal computing revolution led to the internet, which led to the mobile revolution - Now we are covering the future of the internet and computing - How AI ties the personal computer, the smartphone and the internet together

More From Crazy Wisdom