Why a16z's Martin Casado Believes the AI Boom Still Has Years to Run

Casado discusses how AI has most surprised him in coding capabilities, comparing current AI energy to the 1996 dot-com boom. He argues we're still early in the cycle, not near a late-90s bubble, because companies are generating real revenue. He introduces his market-first investing philosophy, emphasizing that markets create companies, not the other way around.

• •AI coding capabilities have exceeded all expectations - Casado never thought it would be this good despite developing his whole life
• •Current AI wave feels like early 1996, not late 90s bubble - companies have real revenue and hundreds of billions on balance sheets
• •Market-first investing approach: always start with market analysis, then ask if this is the right founder for that market
• •The reality is the market creates the company in most cases, not the other way around

Casado shares his unconventional path from writing budget video game engines in the 90s to computational physics at Lawrence Livermore National Lab. He discusses his transition to Stanford PhD work and the difficult decision to continue with Nicira during the 2008 financial crisis, revealing how responsibility to his team and personal temperament drove the choice.

• •Paid for undergrad by writing game engines for budget games that received lowest-ever PC Gamer scores
• •Transitioned from game physics to computational physics at Lawrence Livermore National Lab starting in 1997
• •2008 financial crisis forced a critical decision: pursue Nicira or take faculty position at Cornell
• •Still codes nightly using AI tools, which has rekindled his passion for programming by eliminating framework learning overhead

Casado recounts meeting Ben Horowitz when interviewing him as a potential CEO for Nicira. Despite Horowitz saying he was 'too rich' for the role, a 45-minute conversation taught Casado more than years of other advisors, particularly about pricing as the single most important business decision. He explains his decade-long journey with networking and eventual transition to venture capital.

• •Ben Horowitz's insight: pricing is the single most important decision for company valuation, affecting both growth and margins
• •The pricing conversation from 2009 remains relevant today as software shifts from recurring licenses to usage-based billing
• •Learned as much at VMware scaling to $1B globally as during the startup phase
• •Career moves in decade epochs: 20s focused on code/papers, 30s on building products/business, 40s on portfolio-level investing

Casado describes joining a16z as the ninth general partner when the firm had ~70 people who could all sit around one table. He explains the firm's dramatic evolution into a 600+ person organization with specialized funds, clear progression ladders, and systematic processes - all driven by the question of how to scale venture capital beyond traditional partnership models.

• •A16z grew from 70 people (all generalists, one table on Mondays) to 600+ with specialized funds and clear career progression
• •Infrastructure practice focuses on products where the buyer/user is technical - the 'stuff to build the stuff'
• •Team hiring prioritizes product experience over pure technical prowess - nearly everyone has built a company or run a product team
• •Market understanding requires product background; technical knowledge enables critical conversations but isn't sufficient alone

Casado articulates his systematic investing approach: identify spaces where multiple strong founders are working (they're smarter than investors), determine the leader in that space, and invest accordingly. He emphasizes evaluating founder-market fit over generic founder qualities, focusing on earned knowledge and product-focused CEOs who understand market insertion.

• •Assume 3-4 strong founders working in a space signals opportunity - they're risking time and family wealth
• •Investment decisions require comprehensive analysis despite social media's desire to reduce VC to five simple rules
• •Critical factors: founder-market fit, tactical approach, market growth (not size alone), ability to hire and fundraise
• •Growing markets are essential - even huge markets that are flat or shrinking create terrible business conditions

Casado makes a bold claim that Broadcom's Hock Tan may be the best infrastructure CEO ever, outside of Jensen Huang. He highlights Tan's exceptional employee retention, complex acquisition execution, and legendary work ethic - noting that Tan personally serves as the entire M&A integration committee even for massive deals like VMware.

• •Hock Tan may be the best CEO infrastructure has ever seen, rivaling only Jensen Huang
• •Unbelievable employee retention and ability to execute incredibly complex acquisitions
• •For acquisitions like VMware, the integration 'committee' is just Hock Tan himself
• •Knows every detail of his business and all the numbers - stays away from limelight to focus on execution

Casado explains his controversial tweet about non-consensus investing being dangerous in early stage, which managed to 'piss everybody off.' He clarifies that ignoring follow-on capital and later-stage consensus is incomplete analysis. The tweet sparked debate because it challenged VC orthodoxy while appearing to confirm outsider criticisms of pattern-matching.

• •The 'non-consensus' mantra can be dangerous if you don't consider follow-on capital and later-stage dynamics
• •As you go later stage, VCs become more consensus-driven due to larger capital deployment and need for predictability
• •The tweet was meant as obvious guidance but became a 'Rorschach test' - everyone found a reason to be offended
• •AGI has become a conversational crutch that encourages lazy thinking and obscures meaningful technical discussions

Casado draws detailed parallels between current AI energy and the late 90s dot-com boom, arguing we're in early 1996, not near the 1999 bubble peak. Unlike the 90s, today's companies have real revenue, and the infrastructure is funded by companies with hundreds of billions on balance sheets. He pushes back on premature bubble concerns.

• •A real bubble is when taxi drivers give you stock tips - we're nowhere near that level of excess
• •Key difference from 90s: current AI companies have real revenue and sustainable business models
• •Tech giants (Google, Meta, Microsoft) have hundreds of billions to fund infrastructure vs. leveraged suppliers like WorldCom
• •Current wave allows capital deployment that converts to actual revenue and users - not purely speculative

Casado addresses the paradox of MIT studies showing 95% of enterprise AI deployments failing while individual AI usage explodes. He argues AI is fundamentally a prosumer technology attached to individual behavior - organizations get value through their users adopting ChatGPT, Cursor, etc., not through top-down IT projects.

• •AI is currently an individual prosumer technology, not an enterprise platform deployment
• •Enterprise value comes from users individually adopting AI tools, not from internal AI projects
• •Top-down 'we need more AI' initiatives from platform teams are failing at high rates
• •Better for enterprises to work with AI product vendors than attempt their own projects for now

Casado reveals his personal AI workflow: coding nightly with AI to avoid learning arbitrary framework decisions, using Grok audio mode for deep conversations about historical readings during dog walks, but maintaining strict boundaries against AI-assisted writing because 'writing is thinking.'

• •Codes almost every night using AI - it eliminated the barrier of learning new frameworks
• •Uses Grok audio mode to have synthesis/analysis conversations about historical books while walking his dog
• •Currently reading Eisenhower books and comparing Warren Court rhetoric to current Supreme Court debates
• •Never uses AI for writing - considers writing essential to thinking and won't outsource that process

Casado explains the Cursor investment thesis: half of a16z portfolio companies were already using it, the team executed exceptionally well on the IDE, and they had broad ambitions to change all of code. He calculates the coding market at potentially $3 trillion (30M developers × $100K average × 10% productivity gain).

• •Cursor was the clear leader in AI-powered IDEs with exceptional execution and half of a16z companies using it
• •The coding market could be $3 trillion: 30 million developers earning $100K average, with 10% productivity capture
• •AI has been most surprising in coding - Casado never expected it to be this good despite lifelong development experience
• •Multiple approaches will succeed: God models doing everything vs. composition of specialized models

Casado argues there's no inherent defensibility in AI itself - it just solves the customer acquisition problem because it's 'magic' right now. Companies must build traditional moats: two-sided marketplaces, integration advantages, or workflow lock-in. New behaviors give challengers an advantage over incumbents.

• •No inherent defensibility in AI - it's like 'inventing cold fusion' for customer acquisition but doesn't create moats
• •Must rely on traditional moats: two-sided marketplaces, integrations, workflow advantages
• •New behaviors advantage challengers over incumbents who struggle with execution
• •Priority zero for AI founders: find white space, not worry about defensibility until market slows

Casado describes World Labs' mission to convert 2D images into 3D scenes, founded by pioneers including Fei-Fei Li (ImageNet), Ben Mildenhall (NeRF), and others. The technology has applications in VR content creation, architecture, AR, and critically for robotics, where 3D representation is essential for embodied AI navigation.

• •Founded by true 3D pioneers: Fei-Fei Li, Ben Mildenhall (NeRF), Christoph Lassner, Justin Johnson
• •Converts 2D images to 3D scenes despite not being able to see everything - requires generative components
• •Applications span VR content creation, architecture, AR, and robotics navigation
• •Bringing marginal cost of 3D content creation to zero could create an infinitely large market

Casado argues VR's main problem isn't hardware but content scarcity - unlike deeply immersive video games, VR experiences are limited. Building 3D worlds takes teams of people years. World Labs enables individual artists to create in tens of hours what would take a team a year, dramatically expanding available virtual content.

• •VR's gating factor is content, not hardware - experiences are too limited compared to video games
• •Building large 3D worlds traditionally takes teams of people years of work
• •With World Labs, individual artists can create detailed worlds in tens of hours vs. a year with a team
• •Casado works at World Labs on Wednesdays as a beta user, writing code and testing the product

Casado explains his skepticism of AGI framing: we haven't figured out how the human brain works, and AGI has become a 'holding place for magic and fears' that encourages sloppy thinking. He advocates focusing on concrete problems, solutions, and technology trends rather than using AGI as a universal justification for any argument.

• •We don't understand how the human brain works - language models are likely just a small subset
• •Agrees with Yann LeCun: no straight path from current models to AGI by just adding compute and data
• •AGI becomes a conversational crutch - people use it to justify fears, concerns, or optimistic hopes without real arguments
• •Better to discuss concrete problems like labor impacts or marginal risks based on what's actually happening now

Casado expresses concern about US companies relying on Chinese open source models, arguing the US made policy mistakes by not investing in open source proliferation. China answered the call with phenomenal teams and many of the best models. He criticizes VCs who spoke against open source and academia's silence on the issue.

• •Many of the best AI teams and models are now in China due to their investment in open source
• •US VCs who should know better spoke against open source while academia remained silent
• •Policy risks like developer liability (SB 1047) and spurious litigation slow US innovation
• •Current administration's AI policy has been encouraging, but US has significant catch-up work to do

Casado frames AI progress as part of technology's continuous march rather than a discrete breakthrough. Even if AI research stopped today, enough has been unlocked to create tremendous value. He views investing the same way as five or ten years ago, focusing on productizing unlocked capabilities into real businesses.

• •AI is part of the same continuum of technological progress from 10, 20, 30 years ago
• •Even if AI research stopped now, enough has been unlocked to create tremendous value
• •Investing approach remains unchanged - focus on turning research into real businesses
• •We'll continue to unlock new things as part of the long march, not waiting for a single breakthrough