The AI Frontier and How to Spot Billion-Dollar Companies Before Everyone Else — Elad Gil

The current technological landscape indicates a period where being consensus-driven, especially in AI, is a smart approach, rather than overthinking contrarian strategies. The AI world is experiencing significant shifts, including a new phenomenon of talent acquisition and compute constraints. Recently, Meta aggressively started bidding on AI talent, which was a rational strategy given their multi-billion dollar compute investments. This led to a unique "personal IPO" for a class of 50 to a few hundred top AI researchers across Silicon Valley, whose compensation packages dramatically increased. This phenomenon is unusual, with a similar historical precedent seen only in crypto in 2017. The implications are that a subset of these individuals may now focus on large-scale science projects for humanity or personal quests, or simply disengage. Compensation packages for these top AI researchers are rumored to range from tens to hundreds of millions of dollars, reflecting the immense economic value and transformative potential of AI. A major constraint in the AI industry is compute, specifically a type of memory largely produced by Korean companies. This memory constraint is expected to last for about two years. This bottleneck limits the size and performance of AI models that can be scaled up in the short term, effectively creating a ceiling on how far any single lab can pull ahead of others. This means that major AI labs like OpenAI, Anthropic, and Google will likely remain relatively close in capabilities for the next two years. Beyond memory, future constraints could include data center construction or power and energy supply. Despite these constraints, AI companies are experiencing unprecedented growth, with OpenAI and Anthropic rumored to be at a $30 billion run rate, a speed to scale that far outpaces previous technology generations. In the context of this rapid growth, the question of market structure in AI arises: will it be a winner-take-all scenario or an oligopoly? Historically, in every technology cycle, 90-99% of companies fail. The internet bubble of the 90s saw thousands of companies go public, with only a dozen or two surviving. This pattern is expected to repeat in AI. Founders of successful AI companies should critically evaluate whether now, or in the next 12-18 months, is the optimal time to exit, as their offerings may become commoditized or outcompeted by larger labs. A handful of companies, primarily the core labs, are expected to have durable advantages. These labs, like OpenAI, Anthropic, and Google, are likely to form an oligopoly, especially with current compute constraints preventing one from pulling too far ahead. For application-layer AI companies, durability depends on several factors: how much the product improves with better underlying models, the depth and breadth of the product offering, integration into existing workflows (making it hard to remove), and the use of proprietary data. The issue for AI adoption is often change management, not technology quality. Exit options for AI companies are numerous, given the unprecedented market capitalization of large tech companies. These multi-trillion dollar entities have immense buying power, making large acquisitions feasible. Potential buyers include major AI labs, hyperscalers (Apple, Amazon, Google), large incumbents in specific verticals (e.g., Thomson Reuters for legal tech), or even mergers of competitors to consolidate market share. The speaker's investing philosophy emphasizes market first, then team strength. While exceptional teams can sometimes overcome market challenges, a strong market is generally more critical. This is exemplified by investments in Perplexity, where the founder's exceptional talent was a key factor, and Anduril, which addressed an underserved defense tech market. The speaker also avoids "science projects" – companies focused on cool but commercially unproven technologies – in favor of those with clear market potential. The rapid rise of AI, particularly with the advent of GPT-3 in 2020 and the underlying transformer architecture (2017), signaled a massive step function in capabilities. The key insight was the generalizability of these models and their accessibility via APIs, allowing anyone to integrate powerful AI with just a few lines of code, bypassing the need for complex MLOps teams. This ease of access and broad applicability transformed AI from a specialized, difficult-to-implement technology into a widely usable tool. When analyzing investment opportunities, especially in later stages, the focus shifts to identifying the "one single question" that determines a company's future success. This core belief, rather than complex multi-page models, is often the most critical factor. Examples include Coinbase as an index on crypto growth, Stripe as an index on e-commerce growth, and Anduril's focus on AI and drones for defense. The venture capital landscape has drastically changed since 2011. Historically, companies went public within four years, a timeframe reflected in the standard four-year stock vesting schedule. However, with the rise of private capital and regulations like Sarbanes-Oxley, companies now stay private for much longer, sometimes 10-20 years. This has transformed venture capital into a growth investing industry, handling stages that were once public market domains. Growth-centric questions now focus on the core business's sustainability and potential ancillary drivers. The speaker's book, "High Growth Handbook," is a tactical guide for scaling startups from 10 to 10,000 people, designed for founders to consult specific chapters as needed. A new book is in progress, focusing on the "zero to one" phase of startups, covering topics like hiring first employees and raising initial funding. On the topic of boards, founders are advised to prioritize a better board member over a slightly higher valuation, as valuation is temporary but control (and the impact of board members) is long-lasting. Boards should be proactively composed of a portfolio of people who can genuinely help with strategic direction, talent acquisition, product, and customer introductions. Board members are likened to "in-laws"—people you'll interact with for many years, and if they're investors, they can be impossible to remove. Aggressive distribution is a critical, often overlooked, aspect of startup success. Many wildly successful companies, like Google (with its toolbar) and Facebook (with targeted ads), employed aggressive distribution tactics that are often omitted from their sanitized origin stories. TikTok's massive growth, for instance, was fueled by billions spent on advertising. Even in enterprise, companies like Snowflake invested heavily in sales and partnerships. Sometimes, a company wins not because of the best product, but due to superior sales, marketing, and distribution. Long-held dogmas in industries can be disrupted. For example, the belief that "fraud will kill you in the payment space" was overcome by companies like PayPal. In AI, the dogma that "selling to law firms is a crappy business" was overturned by Harvey, which shifted from selling tools to selling "units of labor" or "cognition," effectively augmenting lawyers. AI has opened up numerous markets previously closed to innovation, creating an environment where founders can achieve explosive growth. If an AI company isn't experiencing rapid growth now, something is fundamentally broken. Great markets are often identified by asking "why now?"—what regulatory, technological, or competitive shifts have created a new opportunity. Examples include Samsara benefiting from new regulations for in-cab monitoring and AI's immediate applicability to all white-collar work due to its language processing capabilities. Defining the total addressable market (TAM) accurately is also crucial, avoiding "fake TAMs" that overstate market size. The speaker uses a multifaceted approach to information consumption: X (formerly Twitter), technical papers/journals, and talking to smart people. Increasingly, AI models (OpenAI, Claude, Perplexity, Gemini) are used for research, with different models leveraged for specific tasks (e.g., Gemini for travel tips). This allows for deep dives into various topics, such as the dramatic increase in autism and ADHD diagnoses, by aggregating clinical trial data, primary literature, and summary charts across multiple models for cross-verification. In personal health, the speaker focuses on basic interventions like good sleep, exercise, and diet, along with Vitamin D and creatine. The discussion touches on more experimental longevity interventions like rapamycin pulsing and neurosensory aging treatments, while acknowledging the often-overstated efficacy of many "biohacking" trends. The concept of "rebooting" the body or brain, akin to rebooting a computer, is explored, with examples like Ibogaine for opiate addiction showing potential for system-wide changes in impulse control and even brain age reversal, though with significant risks. Brain stimulation and bioelectric medicine are highlighted as promising frontiers for treating psychiatric disorders and enhancing performance, potentially offering non-invasive, outpatient solutions. Looking ahead five years, the speaker acknowledges the high uncertainty of a period of rapid change, embracing the unpredictability as part of the "fun." A personal exercise involves developing a 10-year plan, not as a rigid prediction, but to define ambition and guide near-term actions, rejecting a defeatist view of an AGI-driven future. For aspiring investors, particularly in AI, the advice is to immerse oneself in the industry's cluster (e.g., the Bay Area for AI), help smart people, and organically build a track record. This traditional venture story remains viable. The idea that market entry strategy differs from market disruption strategy is also emphasized, with examples like Instagram evolving from a toy to a major social product, and SpaceX leveraging launch capabilities to disrupt with Starlink.