Josh Wolfe on Where Investors Will Make Money in AI

Episode Summary

Executive Summary: The episode examines where value will accrue in AI, arguing that the biggest near-term winners may be incumbents, infrastructure providers, and firms with proprietary data—not just model builders. Josh Wolf of Lux Capital explains how hype, regulation, and shifting technology stacks are shaping AI investing, highlighting hardware, open-source frameworks, vertical models in finance/healthcare, and biology as the next major frontier.

Main Topics: AI hype vs. real monetization (Priority: 5/5): The hosts and guest distinguish genuine technological progress from marketing noise, noting that AI mentions are now ubiquitous in corporate earnings and investor narratives. They ask where real profits will come from and whether current enthusiasm is justified. Infrastructure and chips as early winners (Priority: 5/5): Wolf argues that GPUs, Nvidia’s CUDA stack, and the broader semiconductor supply chain have been foundational to AI’s rise. He stresses that compute scarcity, chip costs, and hardware/software lock-in have created major investment opportunities. Model layer vs. application layer (Priority: 5/5): The discussion contrasts general-purpose foundation models with narrower vertical applications. Wolf suggests many consumer-facing AI wrappers are likely to be features, while durable value may come from specialized models in finance, healthcare, and data-rich industries. Data as the new scarce resource (Priority: 5/5): As compute and algorithms become more abundant, proprietary and reliable data becomes the key competitive advantage. Examples include Bloomberg, banks, insurers, Amazon, Spotify, and healthcare systems with compliant, high-quality datasets. Corporate alliances and regulation (Priority: 4/5): The episode explores how Big Tech is partnering with model makers instead of acquiring them outright, partly due to antitrust scrutiny. These investments recycle cash back into cloud/computation spend and may shape both competition and exit paths. Open source, frameworks, and competitive shifts (Priority: 4/5): Wolf points to Meta’s PyTorch and OpenAI’s Triton as important software layers that may weaken Nvidia’s dominance by making models more hardware-agnostic. Meta’s open-source strategy is framed as both strategic and reputational. The next frontier: biology and human communication (Priority: 4/5): Wolf identifies biology as a major future AI domain because large context windows can handle genomic-scale data. He also argues that human-to-human communication may become scarcer as AI-generated text, voice, and media flood daily life.

Key Arguments: AI enthusiasm was catalyzed less by entirely new technology than by a compelling user experience that felt like “magic,” making advanced capabilities visible to the public. The deepest and most durable AI winners to date have been incumbents like Microsoft and, potentially, Google and Meta—not the startups many expected five years ago. Nvidia’s early dominance comes from both hardware leadership and the CUDA programming ecosystem, but that moat is potentially vulnerable as PyTorch and Triton expand hardware-agnostic development. Training costs for frontier models are rising sharply, but not necessarily because that is a permanent law; cheaper distributed compute and alternative stacks could undercut incumbents. Most consumer-facing AI applications are likely to be transient wrappers or features embedded into larger platforms, not standalone enduring companies. Vertical AI models in finance and healthcare may be more defensible because they can be trained on proprietary, high-quality datasets and require higher accuracy than general web-trained models. Corporate venture and strategic partnerships are becoming central because direct acquisitions face regulatory barriers; investments can still create exit optionality and strategic alignment. Open source is not just ideological; it is strategically useful for companies like Meta that want ecosystem reach and protection against regulatory criticism. Biology may be the most important long-term AI opportunity because large context windows can process genomic-scale data for protein design and drug discovery. AI will likely commoditize coding and increase the importance of prompt skill, while also flooding communication channels with synthetic content and increasing demand for trusted human interaction.

Data Points: Stock Movers report length: 5 minutes or less - Bloomberg promo describing short audio market updates Lux Capital investment in Zoox: $25 million - Wolf describes early backing of the self-driving car company Nvidia market cap growth: from $15 billion to $1 trillion - Used to illustrate the rise of GPUs and AI hardware demand Intel market cap comparison: $150 billion - Context for Nvidia’s rise relative to Intel Intel acquisition of Nirvana Systems: about $350 million - Wolf cites a prior AI-related exit for Naveen Rao’s company Databricks acquisition of Mosaic: $1.3 billion - Recent exit mentioned as another example of AI company value creation H100 chip price: $100,000 - Wolf notes scarcity and expense of Nvidia’s high-end chips GPT-3 training cost: a few million dollars / maybe $10 million - Illustrates how AI model training costs have increased over time GPT-4 training cost: low hundreds of millions of dollars - Shows escalation in frontier model training spend Next frontier model training cost: about $1 billion - Wolf’s estimate for what comes after GPT-4 OpenAI context window: about 8,000 tokens - Used to explain limits on input size and why biology is compelling Anthropic context window: 100,000 tokens - Compared as a much larger capacity for long inputs Anthropic funding from Google: about $300 million - Example of strategic investment structure Runway financing: $140 million - Illustrates strategic corporate participation from Google, Nvidia, and Salesforce Lux Capital fund size: $1.2 billion - Wolf says the firm closed the fund in 10 weeks Cloudflare revenue/growth: about $1 billion revenue; 50%-60% growth - Wolf cites the company as an infrastructure winner Timeline for AI-generated film: within the next two years - Wolf predicts full feature films can be AI-generated soon

Pivotal Quotes: "the feeling that somebody had a superpower" — Josh Wolf: Describing what made ChatGPT-style AI feel like a breakthrough to the public "The great irony here is the flood of money and talent and productivity in AI and deep technology is probably going to bring us closer to our innate humanity." — Josh Wolf: On AI increasing the value of human, in-person trust and communication "we want people to agree with us just later" — Josh Wolf: On Lux Capital’s contrarian investing approach and waiting for consensus to catch up

Implications: AI investing may reward infrastructure, data-rich incumbents, and specialized vertical plays more than generic app startups. Expect more strategic partnerships, open-source competition, and growing demand for trustworthy data and human interaction.