Inside Thrive Capital: Investing in OpenAI, Wiz, Cursor, Nudge, Physical Intelligence

Philip explains his journey from studying physics to computer science, his fascination with technology's impact on history, and how a semiconductor-focused Substack led to meeting Josh Kushner and joining Thrive Capital in 2022.

• •Switched from physics to CS after realizing the timeline to impact was too long, chose venture as path to partner with generational companies
• •Met Thrive through a semiconductor Substack exploring reshoring chip manufacturing to America
• •Philosophy: be a partner to transformational technology companies rather than starting one without a great idea
• •Early investment in Mesh Optical (data center interconnects) shows continued interest in hardware infrastructure

Details how Philip met Cursor's Michael when he was pivoting from a CAD tool for mechanical engineers, the 'electric energy' of great founders, and how the product grew from tens of thousands of users to hundreds of millions in ARR in just 18 months.

• •Met Michael in late 2022 on the day he decided to pivot away from mechanical engineering CAD tool
• •Recommended focusing on software over customer support - team eventually built first AI IDE
• •Spent a year building relationship before investing in May 2024
• •Product demonstrates 'one-way door' magic - once you use it, you can't go back to traditional coding
• •Growth from low single-digit million ARR to 'many hundreds of millions' in 18 months shows compressed AI-era timelines

Philip recounts flying to Tel Aviv during the Israeli-Gaza war to invest in Wiz, a cloud security company that was later acquired by Google for the largest sum ever paid for a startup. Discusses Wiz's unique developer-first approach to security.

• •Met Wiz founders in Tel Aviv in 2022, spent a year building relationship before investing in December 2023
• •Philosophy: 'People who win deals are the ones who want to win them most' - flew to Israel when no other investors would
• •Wiz had unprecedented ratio of deal size to speed: closing 6-figure deals in weeks/months vs. industry standard of months/years
• •50% of users were developers, not security people - operated more like infrastructure platform (Datadog/Cloudflare) than traditional security
• •Near-zero churn and ability to scan cloud environments in minutes created instant value demonstration
• •True platform with adoption across code security, cloud infrastructure, and runtime security after only 4 years

Philip makes the case for why hardware investing is entering a golden age, driven by dramatically cheaper sensors (LIDAR down from $75-80K to $200), software intelligence layers, and a generation of founders trained at SpaceX, Anduril, and Tesla.

• •LIDAR sensor costs dropped from $75-80K (2014) to ~$200 (2024) - 99%+ cost reduction enabling new applications
• •Software intelligence (AI) making hardware far more capable - self-driving cars as proof point
• •New generation of hardware-trained founders from SpaceX, Anduril, Tesla creating talent pool that didn't exist for earlier pioneers
• •Hardware businesses harder to build but more durable once scaled - very difficult to disrupt SpaceX or Anduril
• •Hardware companies can actually accelerate growth at scale (Anduril, SpaceX, Tesla) vs. software's typical deceleration
• •Larger deals become accessible as companies prove themselves - Anduril went from $10-100M contracts to multi-billion dollar programs

Deep dive on Nudge, a neurotech company using non-invasive ultrasound to stimulate specific brain regions to treat depression, pain, and addiction. Founded by Jeremy (ex-Neuralink) and Fred Ehrsam (Coinbase co-founder).

• •Uses same ultrasound technology as pregnancy imaging to stimulate brain activity - very safe, non-invasive approach
• •Targets specific brain regions associated with neurological problems to change neural activity
• •Goal: treat depression/addiction easier than taking a pill - just wear headset on temples
• •Differentiated from Neuralink (invasive implant for reading brain signals) and TMS (unpleasant electroshock therapy requiring clinic visits)
• •Founded by 'life's work founders' - Jeremy from Neuralink engineering, Fred Ehrsam from Coinbase
• •Future vision: everyone uses it for mood, energy, focus optimization, not just treating illness

Philip explains Thrive's philosophy of concentrated bets on exceptional companies, driven by alignment with life's work founders and the mathematical reality of power law returns where top 7 stocks represent 50%+ of NASDAQ market cap.

• •Concentration creates authentic alignment with founders - 'all wins should feel great, all losses should hurt'
• •Portfolio of one for founders should mean near-portfolio of one for investors with real skin in the game
• •Power law reality: MAG 7 represents 50%+ of NASDAQ market cap - few companies drive disproportionate value
• •Strategy reflects belief in investing in companies on right tail of distribution that reshape how world works
• •Venture has 'worst beta, best alpha' of any asset class - index returns mediocre, best companies exceptional

Philip reveals that Thrive was one of only two firms to give OpenAI a term sheet at $29B valuation in early 2023, before ChatGPT's launch. Discusses seeing GPT-4 demo in November 2022, the company's evolution, and why they've continued concentrating capital through $500B valuation.

• •Only 2 firms gave OpenAI term sheet at $29B in early 2023 - required belief, not competitive dynamics
• •Saw GPT-4 demo in November 2022 (pre-public launch) - 'passed the Turing test' and was astounding
• •Original investment model had no mention of ChatGPT - company totally transformed with 1B+ users
• •Every major paradigm in large language models originated at OpenAI - from pre-training to RLHF to reinforcement learning
• •Unprecedented talent density: attracts star researchers but also develops young talent (Sora team median age in low 20s)
• •Confidence increases with each round as company proves dominance - appropriate to invest more at higher valuations

Technical discussion of AI's paradigm shifts - from pre-training (GPT-1/2/3) to post-training techniques (supervised fine-tuning, RLHF) to current reinforcement learning era. Philip compares it to semiconductor evolution from single-core to multi-core to GPUs.

• •AI evolution mirrors semiconductors: many waves of innovation to extract better performance
• •Pre-training era (GPT-1 through GPT-4) followed by post-training (RLHF made models helpful vs. alien)
• •Current shift to reinforcement learning as low-hanging fruit from previous paradigms exhausted
• •More paradigm shifts coming - 'figuring out alchemy of turning metal into thinking machines in real-time'
• •OpenAI's advantage: constantly discovering next paradigm while productizing current one
• •Next 12 months: expect RL to make its way into production software products beyond just labs

Philip argues AI is augmenting rather than replacing workers, with no portfolio companies laying off engineers due to AI tools. Instead, tools like Cursor make everyone a '10x or 100x engineer' and free humanity to work on more important problems.

• •Empirical evidence: zero portfolio companies have laid off engineers due to AI coding tools
• •AI allows companies to tackle wider range of problems without proportional headcount growth
• •More augmenting than substituting - makes existing engineers dramatically more productive
• •Humanity should reallocate brainpower to unsolved important problems: oncology cures, sustainable mining, space exploration
• •AI frees people from less important work to focus on highest marginal use of creative potential
• •Not about job destruction but problem expansion - more important work to do than ever before

Philip shares Thrive's approach to measuring success through decision processes rather than near-term outcomes, drawing on Ray Dalio's philosophy. Emphasizes being 'paranoid but patient' and investing against fundamentals, not momentum.

• •Ray Dalio principle from Bridgewater: 'focus on the swing, not where the ball goes'
• •Evaluate decision quality and process, not short-term performance fluctuations
• •OpenAI example: looked like it went nowhere 2017-2020 as nonprofit/capped-profit, now most valuable private company
• •Worst investments: ones where you'd feel bad if numbers change one quarter - suggests investing on momentum not fundamentals
• •Be 'paranoid but patient' - always thinking about risks but not rushing into decisions
• •Right inputs matter more than near-term outputs - confidence in decision process enables long-term holding