# NVIDIA’s Core Business Threatened by TPUs, Sam Altman’s ‘War Mode’ & Sierra’s $100M ARR Source: https://www.youtube.com/watch?v=U1sy7t_Wns8 Recap page: https://rapidrecap.app/video/U1sy7t_Wns8 Generated: 2025-11-27T15:02:38.266+00:00 --- ## Quick Overview The primary threat to NVIDIA's profitability stems from its customer concentration, as major spenders like Google, which spends billions on compute, are incentivized to develop internal alternatives like TPUs to capture massive profit margins currently flowing to NVIDIA, despite the high difficulty and cost of building comparable chips. **Key Points:** - Anthropic secured up to $15 billion from Microsoft and NVIDIA, with commitments for $30 billion in Azure compute, pushing its valuation to $350 billion. - Google's Gemini 3 was trained on its proprietary TPUs, raising the question of whether every large vertically integrated company must now own the chip layer to compete with NVIDIA's GPU dominance. - If Google were buying NVIDIA chips for its $36 billion compute spend, it would be handing NVIDIA over $20 billion a year in profit at high gross margins, incentivizing them to build internal solutions. - The speaker argues that hyperaggressive mode is essential for success, stating, "I do not think you can push your team too hard. I think you should push them as hard as the business needs to go. And if they leave, it's great." - Sierra, Brett Taylor's customer support AI company, reached $100 million in ARR within two years, valued at $10 billion, though the speaker notes the enterprise AI support category is currently oversold. - The valuation for Sierra requires massive growth, needing to reach $5 billion in ARR in five years, which necessitates eating a huge slug of the $200 billion annual customer support labor market, not just stealing software spend from Salesforce's Service Cloud. - The PE ratio on NVIDIA is currently lower than the PE on Costco, suggesting that the valuation hinges primarily on whether the massive demand for compute is a steady state or a cyclical peak. **Context:** The discussion centers on major developments in the AI landscape, specifically the massive investment round secured by Anthropic from Microsoft and NVIDIA, and the increasing vertical integration by hyperscalers like Google developing their own AI hardware (TPUs). The conversation also touches on the internal operational philosophy necessary to compete, using Sam Altman's 'war mode' memo as a catalyst for discussing the need for 'hyperaggressive mode' in high-growth companies, and analyzes the rapid scaling of Brett Taylor's new venture, Sierra, in the customer support sector. ## Detailed Analysis The core debate revolves around the sustainability of NVIDIA's dominance, facing a significant medium-term threat from its largest customers developing in-house chips like Google's TPUs. Because NVIDIA's revenue is highly concentrated among a few massive spenders, these customers face a huge financial incentive to internalize compute to avoid sending billions in profit margins to NVIDIA; for instance, Google's potential $36 billion chip spend could represent over $20 billion in annual profit for NVIDIA if purchased externally. Although building a comparable chip is difficult, the potential profit capture makes it a necessary consideration for giants like Google, Amazon, and Tesla. On the operational front, the panel agrees that 'hyperaggressive mode' is non-negotiable for survival in the current AI race, with one speaker asserting, "I don't want to hear about your pilot. I want to smell that your team is in hyperaggressive mode," noting that velocity increases only under extreme pressure, even if it causes some turnover. Regarding consumer market share, while Google's core search business has proven more resilient than anticipated, OpenAI's ChatGPT has successfully carved out a defensible, paid subscription niche in the attention economy, leading to an expanding pie rather than a zero-sum game yet. Finally, Sierra's achievement of $100 million ARR at a $10 billion valuation highlights the immense faith in LLMs disrupting customer support; however, the speakers caution that realizing this valuation depends on the company successfully transitioning from selling software to displacing vast amounts of human labor, a process constrained by the inherent complexity and change management required for enterprise diffusion. ### Anthropic's Investment & Microsoft's Strategy - Anthropic secured $15B from Microsoft and NVIDIA, valuing it at $350B with $30B in Azure commitments - Microsoft's move suggests an 'open marriage' approach after OpenAI sought independence from its prior monogamous relationship with Microsoft. ### The TPU Threat to NVIDIA - Google trained Gemini 3 on its own TPUs, posing a risk due to customer concentration where 4-5 customers account for 80% of NVIDIA's revenue - If Google spent $36B on compute from NVIDIA, it would yield over $20B profit to NVIDIA, incentivizing internal chip development. ### Operational Mandate - Hyperaggressive Mode: Success requires being in a 'hyperaggressive mode,' as nothing happens otherwise, despite the risk of turnover - Signs of this mode include management teams universally reporting faster shipping velocity and increased execution across the board. ### Google vs. OpenAI Consumer Battle - Google's search decline is not precipitous; the head of search noted search volume is increasing as people ask more questions - OpenAI carved out a new category of paid subscription for advanced AI, suggesting the pie is expanding rather than being wholly taken from Google. ### Sierra's $100M ARR Valuation Analysis - Sierra hit $100M ARR at a $10B valuation, demonstrating significant belief in LLMs automating customer support, potentially solving 60% of calls - The valuation requires Sierra to grow to $5B ARR in five years by displacing the $200B annual labor market for support agents, constrained by enterprise physics. ### NVIDIA Valuation Drivers - NVIDIA's PE is lower than Costco's, meaning valuation depends on demand stability, not current earnings - The primary question is whether compute demand in 2025/2026 is a steady state or a cyclical peak.