Martin Casado and Sarah Wang, general partners at Andreessen Horowitz, discuss the unprecedented dynamics reshaping AI investing with Alessio Fanelli and Sean Wang from Latent Space podcast. Casado, who created software-defined networking at Nicira, and Wang, described as the "best growth investor in the entire industry" for AI companies, bring unique perspectives from both venture and growth investing.

The conversation explores the capital flywheel effect where AI companies can raise massive rounds, convert capital directly into compute and capabilities, then use breakthrough results to raise even larger subsequent rounds. This dynamic differs fundamentally from traditional software development constraints described in The Mythical Man-Month, where engineering progress couldn't be accelerated simply by adding resources.

Key topics include talent wars reaching $5 billion acquisition prices, the blurring lines between infrastructure and applications, whether frontier model companies can consume entire ecosystems built on their APIs, and why "boring" enterprise software remains underinvested despite strong fundamentals.

The AI Capital Flywheel: From Zero to Billion in Record Time

"A model company can raise money and drop a model in a year with a team of 20, and produce something with immediate demand" - Martin, contrasting with The Mythical Man-Month where engineering couldn't scale with resources.

Unlike the internet buildout where "investors put money into fiber that nobody used," AI has "no dark GPUs" with immediate demand for every compute dollar invested.

The emerging strategy: raise money for compute, achieve breakthrough, funnel into vertically integrated applications, gain massive user share while potentially subsidizing, then repeat the cycle.

"If Frontier Labs can raise three times more than the aggregate of every company built on top of them, they may consume the entire application layer" - creating unprecedented market concentration.

Talent Wars and $5 Billion Acquisitions Reshape Industry

"Very rarely can you see someone get poached for $5 billion. That's hard to compete with" - Sarah on Meta's aggressive talent acquisition strategy.

A16Z struggles to compete for talent with "active offers for 10 million a year" even for L5 engineers, fundamentally changing early-stage founder economics.

2025 talent wars may have been "just like a blip" as Meta assembled their team, but inflated compensation continues trickling down across the industry.

Historic amount of M&A for "basically hires" creates positive venture outcomes through effective acqui-hires despite headline disruption.

The Underinvestment in Boring Software

"If you're not basically growing from zero to 100 in a year, you're not interesting, which is the silliest thing to say" - Martin on current investment mania.

Traditional enterprise software companies in large markets growing 5x annually struggle for investor attention despite being "great investments for anybody."

"Boring software, boring enterprise software" represents the most underinvested sector, with companies that would deliver strong returns but lack AI growth narratives.

LPs want "3x net over the life cycle of a fund" - returns easily achievable by solid software companies in big markets growing 5x annually.

Every Task is AGI Complete: The Coding Model Lesson

"What happened to all the special coding models? None of them worked" - specialized models failed because coding requires full conversational intelligence.

"When I'm coding, it's not just code. It's everything" - Martin describes using models for compliance, web search, history, and brainstorming while engineering.

"There's no such thing as a coding model. You're talking to another human being and it's good at coding, but it's got to be good at everything."

OpenAI will likely release "GPT-5 and GPT-5 Codex" - "one for Riz and one for Tiz" - suggesting two-dimensional specialization rather than task-specific models.

Market Structure: Oligopoly vs Fragmentation

Two potential futures: infinite market with fragmented value creation, or general models that "consume everything beyond it" through superior capital access.

Current models are "gross margin positive" on existing capabilities but "gross margin negative" on next-generation training runs, borrowing against future fundraising.

"If you can raise more money than the aggregate of everybody that uses your models, that doesn't even matter" - capital advantage trumps technical specialization.

Companies like Anthropic with 60-80% API margins can potentially outspend entire ecosystems built on their infrastructure through successive fundraising rounds.

Investment Thesis: N-of-One Founders and Capability Breakthroughs

A16Z invests in "n of one founders" with demonstrated prior breakthroughs: Ilya Sutskever at SSI, John Schulman as "godfather of reinforcement learning."

"When there's a real capability breakthrough, the demand is there. And so the revenue growth is much faster than we've ever seen once it's turned on."

One unnamed company went from product GA to "tens of millions of revenue" in weeks, compared to traditional SaaS companies taking seven years for similar levels.

"We don't think it's a zero-sum game" - specialization creates value as seen with ElevenLabs maintaining audio leadership despite numerous competing models.