The Frontier Ideas a16z Wants to Fund | Marc Andreessen, Ben Horowitz & Chris Dixon

A16Z Frontier broadens the firm’s crypto investing into areas where crypto now overlaps with fintech and Wall Street. Stripe’s use of stablecoins and financial firms’ moves into tokenized stocks and other assets make it harder to define companies as simply “crypto.” Ben Horowitz said Frontier fills a gap in a16z’s structure, letting it back compelling founders and ideas beyond existing verticals. The fund aims to invest early, accepting category risk; an idea could later grow into a larger practice or fund. The partners said a16z ultimately bets on entrepreneurs more than categories or products.

Chris Dixon described frontier investing as looking for second-order effects: improved smartphone components, for example, helped make drones and Oculus possible, while crypto followed from open-source software and services. The partners see AI expanding the range of ideas founders can pursue, across software, hardware, science and entertainment. Startup formation is already rising and could accelerate as AI lowers the cost of building companies and ambitious products. The key question, they argued, is increasingly whether an idea is ambitious enough. A 30-person team may now be able to build a full-stack device company, compared with roughly 1,000 people a decade ago, while the gap between AI-native startups and typical incumbents is unusually wide and growing.

The most transformative AI products may enable new behaviors rather than simply improve familiar tasks such as coding, math, search and knowledge work. Such products carry greater risk because multiple things may have to go right, but could also offer greater upside; the partners pointed to new social and dating experiences and internet agents that might eventually outnumber people. They said technology, culture, business conditions and politics all affect timing, and successful ideas sometimes had failed precursors or arrived about a decade too early. VR illustrates the uncertainty: despite Oculus’s acquisition, it had not become as mainstream as hoped 13 years later, partly because technical problems remained.

AI could let independent developers make games with capabilities once available only to large studios, while leaving room for human creativity. One speaker called adoption inevitable because AI offers a better way to build games and benefits consumers. The partners were less certain about AI’s effect on movies, but said it could lower production costs, improve effects and make previously impractical stories viable. Horowitz predicted AI films would become a distinct form of media alongside more efficient conventional filmmaking. As an example of the economics at stake, one speaker said Digger represented a $200 million cash loss for its studio despite a production budget of about $180 million, and suggested that at a $2 million cost it might have been “insanely profitable” and returned investors’ money ten times over. A recently released open-source tool, R-E-A (“reverse engineer, anything”), was described as using AI to reverse-engineer hardware and generate source code, potentially letting people customize devices such as thermostats.

The partners argued that open-source AI can widen access to expertise and prevent power from concentrating in a few companies or government. Marc Andreessen said open-source AI ecosystems are active in the US and China alongside proprietary systems, and estimated that open models were six to 12 months ahead of closed models, depending on the comparison. He said OpenRouter data showed open-model token consumption growing faster, without giving a rate. He also said the current US administration strongly supports American open-source AI and China supports open-source AI for the world, while warning that governments have threatened severe restrictions or bans. Horowitz identified reward hacking, when an AI takes an unintended route to obtain its reward, as a major safety problem. Open model weights, he argued, would let researchers study the problem and possible defenses.

Dixon warned that AI is already disrupting the internet’s original exchange: content sites let platforms such as Google index their work in return for referral clicks, but search and social services increasingly answer users directly or keep them on-platform. AI replacements have coincided with steep traffic declines at Stack Overflow and other websites, putting businesses dependent on attention, advertising and conversions under pressure. Dixon predicted a major internet reconfiguration over roughly the next decade, more consequential than mobile’s expansion from about 500 million internet users to 5 billion. He expects AI voice to help unlock wearables and AI-generated immersive content to help unlock VR; collaborative networks of agents could move beyond individual interactions toward tasks such as designing buildings or making scientific breakthroughs. The partners also pointed to AI’s potential in fields from biology and design to classics, including deciphering the Herculaneum scrolls, and urged people to use the tools and work at the cutting edge.