a16z

a16z is Andreessen Horowitz, the venture capital firm founded by Marc Andreessen and Ben Horowitz in 2009. It invests from seed through growth across technology sectors, including AI. Its AI work combines financing with analysis of business economics, experimental developer tools, and compute support for portfolio companies. The firm's operating platform provides talent, marketing, legal, and policy expertise alongside investment. a16z reported more than $100 billion under management across multiple funds as of April 30, 2026.[1][3][4][6]

That combination of specialized capital and operating support developed over years. The firm's earlier bio funds and separate early-stage and growth vehicles provide context for its later organization around AI models, applications, and infrastructure.[16][17][2]

a16z launched publicly in July 2009 with $300 million in initial capitalization. Horowitz later described its starting structure as one fund with two general partners. In 2015, the firm launched a $200 million bio fund; in December 2017, Jorge Conde and Vijay Pande announced a second, $450 million bio fund, investing from seed to growth at the intersection of biology and engineering. Their interests included computational biomedicine and AI and machine learning for diagnostics and therapeutics.[15][2][16]

Separate investment vehicles also emerged around company stage. In May 2019, Axios reported that a16z had closed two funds: a $750 million early-stage fund and a $2 billion later-stage fund led by David George, alongside its dedicated bio and crypto funds.[17]

In February 2020, Martin Casado and Matt Bornstein's The New Business of AI argued that AI companies combined elements of software and services. Cloud infrastructure and ongoing human support could reduce margins; customer-specific work and edge cases could complicate expansion; models and data might offer less defensibility than founders expected. They reported anecdotal AI gross margins often around 50–60%, compared with a 60–80%+ benchmark for comparable SaaS businesses. Their AI estimates came from founder interviews, rather than a representative sector study.[3]

Their concrete engineering example was the cost of maintaining separate models for each customer. That architecture can multiply training runs, data pipelines, and maintenance. They recommended sharing models where feasible to simplify deployment and reduce cloud costs. Even apparently similar customers could require different training data: two manufacturers doing defect detection might place cameras differently on their assembly lines. Such variation helps explain why adding customers can require continuing model work rather than simply distributing another copy of software.[3]

The a16z-infra Llama 2 chatbot, documented as an experimental July 2023 release, connects a Streamlit interface to Meta's models through Replicate. It maintains session chat history, offers 7B, 13B, and 70B model endpoints, and exposes temperature, top-p, and maximum sequence-length controls. Marco Mascorro, Yoko Li, Rajko Radovanović, Matt Bornstein, and Guido Appenzeller are credited as authors. The application uses Apache 2.0 licensing, separately from Meta's model license.[4]

AI Town provides an MIT-licensed, deployable starter kit for a virtual town where AI characters interact. Inspired by Generative Agents: Interactive Simulacra of Human Behavior, it supplies an extensible JavaScript/TypeScript framework rather than originating that research. Convex provides the game engine, database, and vector search, supporting shared global state, transactions, and simulation. The application machinery coordinates a running world around language-model calls; its intended scalability is a design goal rather than a measured performance result.[5]

On April 16, 2024, a16z announced $7.2 billion raised across American Dynamism ($600 million), Apps ($1 billion), Games ($600 million), Infrastructure ($1.25 billion), and Growth ($3.75 billion). Horowitz explained the specialized structure through the different expertise and networks each category required: founders building AI foundation models needed different capabilities from founders developing biotech therapies. Dedicated funds paired investment teams with operating platforms focused on their segments.[2]

a16z announced its investment in World Labs in September 2024. Founded by Fei-Fei Li, Justin Johnson, Ben Mildenhall, and Christoph Lassner, World Labs was developing a foundation model to generate interactive three-dimensional worlds for spatial reasoning. Casado and Sarah Wang described Casado working from its offices, learning from the AI and graphics team and helping balance deep research with product development. The model development belongs to World Labs; a16z's role was investment and support.[9]

In October 2024, a16z described Oxygen, a compute-support program led at that time by Anjney Midha and Guido Appenzeller. Its rationale was a mismatch between startup experimentation and GPU contracts favoring large customers with long commitments. Training capacity can be planned for a defined run, while post-launch inference demand depends on customer usage. Committing too early can leave a startup paying for idle resources. Oxygen helps portfolio companies obtain training and inference capacity with shorter commitments and greater flexibility.[6]

On January 9, 2026, a16z announced more than $15 billion raised: American Dynamism ($1.176 billion), Apps ($1.7 billion), Bio + Health ($700 million), Infrastructure ($1.7 billion), Growth ($6.75 billion), and other venture strategies ($3 billion). This was a multi-strategy raise, rather than a single dedicated AI fund or a measure of capital already deployed.[11][14]

In their February 2026 Latent Space interview, Wang and Casado described frontier-model financing as combining an early-stage bet on technical founders with growth-scale resources. Deals could join financial and strategic investors with negotiations over equity, compute contracts, and commercial partnerships. Wang described a financing cycle of raising capital, buying compute, producing a capability breakthrough, gaining application users, and raising again. She made its key assumption explicit: capital translates into capability only while scaling laws hold and capabilities continue improving.[7]

The firm's market tracking also changed. In Olivia Moore's March 2026 consumer AI ranking, eligibility expanded beyond AI-native products to established products where generative AI had become central, including Canva and CapCut. The sixth edition ranked web products by unique monthly visits from SimilarWeb and mobile apps by monthly active users from Sensor Tower, both for January 2026. These are different usage proxies, not measures of technical quality or revenue; broader eligibility also limits direct comparisons with earlier editions.[10]

In their August 27 analysis, Tugce Erten and Sarah Wang distinguished raw model access from applications combining proprietary data, tools, orchestration, integrations, and workflow logic. They recommended pricing applications around recognizable work—such as an account brief or implemented code change—or outcomes when results are measurable and attributable. Their framework separates what work is worth from what it costs to deliver, allowing efficiency gains from routing and caching to improve margins. Token pricing remains appropriate for model access, with pass-through sometimes useful for unusually volatile costs.[13]

On August 28, a16z announced $1.1 billion raised for the Machine Age Fund. Its stated scope includes chips, memory, networking, storage, data centers, robotics, and home AI appliances. The fund formalized a hardware investment direction alongside earlier investments: the firm had led Skydio's Series A in 2016, first invested in Anduril in 2019, and participated in Waymo's 2020 raise. It extends AI investing into the physical infrastructure and systems on which applications and models depend.[12]

On August 31, David George and Raghu Raghuram announced additional capital closed, bringing a16z's fifth Growth fund to $8.5 billion, updating January's $6.75 billion amount. They also expanded Growth operating support across sales, marketing, pricing, revenue operations, and AI-native go-to-market work. The authors reported helping more than 100 companies through scaling transitions over more than seven years. The expanded services address decisions such as moving beyond founder-led sales, introducing usage-based pricing, and managing AI margins.[18][11]