Episode #83: The Focus Layer: Why Anthropic, NVIDIA, and Cloudflare Are Winning the Same War

In this episode of Stewart Squared, host Stewart Alsop III and his father Stewart Alsop II cover a wide range of interconnected topics, starting with a sharp critique of OpenAI's lack of strategic focus under Sam Altman and how that compares to Anthropic's disciplined, consistent approach — including Anthropic's explosive ARR growth from $14 billion to $19 billion in just three months. From there, the conversation moves into the slowdown in AI model progress and the role of training data scarcity, the rise of vibe coding and AI-assisted software development, the architectural differences between CPUs and GPUs (with a nod to Jensen Huang's revealing interview with Ben Thompson on Stratechery about NVIDIA's vision beyond graphics chips), the emerging threat of world models as an alternative to LLMs, the geopolitics of satellite internet and Elon Musk's control over Starlink, Cloudflare's role as a de facto network operating system, the state of robotics and what a personal robot revolution might look like, autonomous vehicles and the LiDAR vs. video-only debate, and the historical parallels between the personal computer era and where AI and robotics are headed today.

Links mentioned:
- Coco Robotics: https://www.cocodelivery.com
- Niantic Spatial: https://nianticlabs.com

Timestamps

00:00 - Stewart II unveils the new recording studio, built entirely through vibe coding without writing a single line of code.
05:00 - Stewart Sr. argues OpenAI is in serious trouble, citing Sam Altman's opportunistic rather than strategic leadership style.
10:00 - Discussion shifts to Anthropic's disciplined focus versus OpenAI's scattered bets, with Anthropic's ARR jumping from 14B to 19B in three months.
15:00 - Training data bottleneck explored, LLM progress stalling as internet datasets are exhausted, forcing companies to manufacture synthetic data.
20:00 - World models emerge as existential threat to LLM companies, with Jensen Huang and NVIDIA quietly preparing CPU architecture for the transition.
25:00 - Personal robot revolution compared to personal computer era, debating humanoid robots versus specialized machines and standardization challenges.
30:00 - Hardware reality hits as Stewart II confronts robot-building complexity, exploring the ESP32, servo motors, and robotic arm pathway.
35:00 - Starlink's satellite network dominance discussed, including Elon cutting off Russian terminals and geopolitical consequences for Ukraine.
40:00 - Cloudflare emerges as the Internet's de facto network operating system, layering security and control over global traffic.
45:00 - Self-driving cars framed as the proving ground for robot localization, debating Tesla's video-only approach versus Waymo's LiDAR strategy.

Key Insights

1. OpenAI's Strategic Drift Is a Critical Weakness. Stewart Alsop (the father) argues that OpenAI is in deeper trouble than most recognize, attributing this to Sam Altman's opportunistic rather than strategic leadership. OpenAI expanded into numerous side projects before abruptly reversing course, and its latest foundational model has fallen behind competitors like Claude and Gemini. Without the cash flow reserves that Meta or Google possess, OpenAI has fewer options to recover, raising serious questions about its IPO readiness and long-term viability.
2. Anthropic's Consistency Is Paying Off Enormously. Unlike OpenAI, Anthropic has maintained a disciplined, unchanged strategy since its founding. This focus is reflected in its annualized revenue jumping from $14 billion to $19 billion in just three months, largely driven by Claude Code's superior agent "harness" that competitors have struggled to replicate.
3. Training Data Scarcity Is Slowing AI Progress. Stewart Alsop II highlights that the internet has essentially been fully consumed as a training source, forcing AI companies to generate synthetic datasets through specialized firms. This bottleneck is a structural constraint on model improvement, not merely a talent or energy problem.
4. World Models Represent an Existential Threat to LLMs. Both Stewarts agree that world models—AI systems grounded in real-time, physical reality rather than static text—could fundamentally disrupt the current LLM paradigm. Notably, existing foundational model companies almost never mention world models publicly, suggesting awareness of the threat.
5. NVIDIA Is Positioning Beyond GPUs. Jensen Huang's conversation with Ben Thompson revealed that NVIDIA views itself as a full computing architecture company, not merely a GPU supplier. Through partnerships like their Groq CPU licensing deal, NVIDIA is preparing for a future where both CPUs and GPUs must coexist in AI infrastructure, particularly for world model applications.
6. Robotics Lacks the Standardization Needed for Scale. A true operating system for robots cannot emerge without standardized hardware at scale—a lesson drawn from how Microsoft's OS only succeeded after IBM standardized the PC. Current robotics remains fragmented across specialized applications, making a universal robotic OS premature, with the Roomba cited as the only truly mass-scaled robot to date.
7. Vibe Coding Is Democratizing Software Development. Stewart Alsop II built an entire podcast recording studio by speaking instructions to an AI without writing a single line of code himself, using Claude Code as his development engine. This signals a broader shift where the barrier between understanding software conceptually and actually building it collapses, potentially reshaping who can participate in technology creation.