Andreessen Horowitz said it is restructuring its crypto investment platform into a broader frontier technology fund, widening its focus from blockchain to a larger set of emerging technical fields.
In an a16z podcast special released on Oct. 22, 2026, Chris Dixon said the firm’s former crypto fund is being expanded into a new vehicle called a16z Frontier. The discussion featured Dixon alongside a16z co-founders and general partners Marc Andreessen and Ben Horowitz, with Erik Torenberg, founder of Turpentine, serving as host.
a16z Frontier broadens the old crypto mandate
Torenberg opened the episode by framing the move as a major strategic announcement for the firm: the well-known a16z crypto fund is being expanded and repositioned as a frontier exploration fund.
Dixon said the team has spent the past several years rethinking how frontier technology cycles are evolving. When a16z launched a dedicated crypto fund in 2018, blockchain represented one of the most compelling edges of technology. That edge has since widened, he said, as generative AI, compute infrastructure and modern cryptography have started to spill into one another.
In his account, the strongest founders no longer fit neatly inside one category. Some are building decentralized networks while also working on embodied robotics, local AI agents, new compute clusters, synthetic biology and new immersive media. That shift is what led the firm to widen the fund’s remit.
Dixon said, 「From today onward, not just blockchain, but every field at the edge of human technological possibility, we will back through a16z Frontier.」 He added that founders should stop asking whether their ideas are too aggressive and start asking whether their ambitions are large enough.
Andreessen placed the move in a longer institutional timeline. Over 15 years, he said, a16z has repeatedly changed its structure and investment strategy as technology waves changed shape. The path from cloud and mobile to enterprise software, health and crypto was never linear. AI is now pulling together decades of accumulated work in hardware, physics and software, and Frontier is meant to reflect a mindset of continued exploration.
Horowitz said the most important breakthroughs often emerge in the gray zones between disciplines. If a startup is building humanoid robots powered by open-source foundation models while using onchain systems for data rights and micropayments, he said, it no longer makes sense to force that company into either an AI or crypto bucket. Frontier is meant to remove those artificial boundaries.
Is being too early still a fatal mistake?
Torenberg then raised a long-standing venture capital saying: in technology, being too early can look almost the same as being wrong. He asked Andreessen whether that still holds in the current cycle.
Andreessen said the warning made sense for decades because startups used to depend on external infrastructure that often did not yet exist. A company trying to build online video 20 years ago, for example, had to wait for broadband, smartphones and cloud-native data centers. Many businesses would run out of money before the environment caught up.
He argued that the timeline has now compressed sharply. Startups no longer need to lay cable, build chip fabs or assemble every layer from scratch. Public cloud infrastructure, open-source base models and dense mobile device distribution are already in place. AI itself, he said, is acting as a force multiplier, shrinking work that once took years — coding, prototyping, molecular simulation and hardware rendering — into days or even hours.
That changes the risk profile. In Andreessen’s view, the gap created by being too early has narrowed. The bigger danger now is not being bold enough. A company that is only six months ahead may find itself overtaken before launch by large technology firms or open-source communities moving at much higher speed.
Dixon agreed. He said many people once dismissed smart contracts and decentralized compute as premature, yet those early layers helped produce today’s censorship-resistant, high-throughput distributed settlement networks. For Frontier, he said, taking “too early” risk is part of the point because outsized returns and category-defining companies tend to appear before consensus is comfortable.
A historic gap between builders and the mainstream
Dixon described the present moment as one of the widest perception gaps in the history of computing. In his telling, mainstream society, many multinational executives and even a large share of regulators still see AI as a chat interface that writes office documents or edits resumes. They treat it as a productivity add-on.
That is not what early builders are seeing, he said. In top developer circles in San Francisco and elsewhere, teams are already using hundreds or thousands of autonomous agents to help write complex operating systems. Robots are training in simulation at speeds far beyond human learning cycles. Synthetic biology researchers are using deep neural networks to design highly active proteins that do not exist in nature, and doing so in a matter of weeks.
Andreessen compared the disconnect to the early 1990s, when many people saw the internet as little more than an electronic newspaper. They did not imagine it would reshape finance, supply chains or geopolitics. For founders, he said, that kind of gap can become a moat. While incumbents interpret new systems through old frameworks, builders can use new tools to rebuild the base layer.
Who captures AI profits: incumbents or startups?
Asked whether AI’s economic upside will mostly accrue to existing giants or to a new generation of startups, Horowitz said the answer is mixed.
On the surface, he said, companies such as Microsoft, Google, Amazon and NVIDIA have clear advantages. They already control distribution, compute budgets and enterprise relationships. Plugging AI models into existing products like office suites or search engines is the fastest route to revenue.
Still, he argued that technology history points to a recurring limit: incumbents are constrained by the innovator’s dilemma. Their organizations, sales channels and business models are built for the old world. When a new paradigm threatens the profit engine they already own, hesitation follows. In the early internet era, telecom companies controlled the fiber and phone networks and looked well positioned to dominate. The companies that emerged instead were Netscape, Amazon, Yahoo and Google. New categories do not win by improving old businesses by 10%, he said. They win by starting with a different model from day one.
Dixon said that dynamic may be even stronger in the areas Frontier wants to fund. If the future internet is made up of autonomous agents settling value and allocating resources among themselves, then centralized systems built around card payments and ad clicks stop making sense. Agents do not need banner ads. They need low-friction micropayment rails and cryptographic credit onchain.
The agent economy and onchain identity
Dixon said the internet could look very different within three to five years. In that scenario, humans may no longer make up the majority of active participants online. Tens of millions of AI agents could be acting on behalf of people in asset allocation, data retrieval, code deployment and supply-chain bidding.
That creates a basic engineering and financial problem: how do agents establish identity, and how do they settle across borders without traditional bank accounts? Dixon said centralized banking systems are not built to issue debit cards to autonomous code. That is why, in his view, crypto protocols and decentralized agents belong together. Agents need their own cryptographic keys, their own wallets and smart-contract-based ways to coordinate and settle in permissionless networks.
Andreessen extended the idea into social platforms. The last 20 years of social media were built around person-to-person following graphs, he said. The next version may be a mixed ecosystem in which agents have their own personalities, knowledge bases and economic goals. They could debate, create content and even form decentralized collaborative groups. That would alter how people think about both community and media.
Andreessen’s critique of “Product Manager Brain”
One of the sharper parts of the conversation centered on what Andreessen has recently called “Product Manager Brain.” He described it as a bureaucratic style of thinking that grew out of the mature mobile internet era: optimize retention curves, run multivariate A/B tests, change a button from dark blue to light blue and celebrate a tiny gain in click-through rate.
He said that approach may squeeze some profit out of mature software markets, but it does little at the frontier and can actively suppress major innovation. If Steve Jobs had built the iPhone through focus groups and A/B testing, Andreessen said, the result would have been a BlackBerry with a smaller keyboard.
Real breakthroughs, in his view, come from taste, artistic instinct and engineering obsession, especially in AAA games, film storytelling and new hardware categories. AI matters here not because it helps managers produce more dashboards, but because it gives small teams with strong taste access to powerful generative tools. He said a 10-person studio may eventually be able to build an interactive universe with visual detail and world-building depth beyond top-tier Hollywood productions.
Horowitz agreed and said business is not just an algorithmic loop. It is also about reaching human curiosity and emotion. Founders need to reclaim judgment and place real bets on unknown experiences instead of handing decisions over to dashboards.
Open-source AI as a counterweight to concentration
Torenberg brought up a line often associated with investor Peter Thiel from roughly a decade ago: crypto is libertarian, AI is communist. The idea was that blockchains distribute power while AI tends toward concentration because of compute and data requirements.
Andreessen said that reading had force in an earlier phase, when deep learning depended heavily on centralized data centers and a small number of large operators. If only a few companies and governments control the most capable machine intelligence, he said, AI can become a deeply centralized tool.
That is exactly why open-source AI matters, in his view. He pointed to Meta’s open-source Llama family and to the broader global community working on architecture improvements and quantization. Advanced intelligence is being compressed quickly, he said. A model that would have required supercomputer-class resources a few years ago can now run locally on a laptop or even a smartphone chip. Open source means the code is visible, the weights are not controlled by a single gatekeeper, and anyone can fork, fine-tune or deploy.
Dixon said the logic matches the spirit of cryptography. Cryptography removed privileged financial intermediaries; open models can remove intermediaries around thought and intelligence. He also said some entrenched interests in Washington and other regulatory centers are lobbying under the banner of safety to classify open-source software as illegal, creating compliance barriers that protect monopoly positions. He described that as a classic “Baptists and bootleggers” alliance. Frontier, he said, will continue backing open ecosystems, decentralized compute and open protocols.
The internet’s original bargain is breaking down
Dixon also returned to a theme from his book Read Write Own: the collapse of the internet’s original economic contract. Early Web2 platforms were built on an implicit bargain, he said. Independent creators, bloggers, news outlets and community contributors supplied large amounts of high-quality content for free, and platforms returned distribution, audience growth and monetization opportunities.
In his account, that bargain has now broken down in public. Once platforms secured network effects, they tightened algorithms and pushed creators toward paid distribution. In the generative AI era, model companies scrape years of human work from across the web, then answer questions directly inside search and chat interfaces without sending users back to the original sites.
Dixon said that cuts off the economic loop that sustained independent creation. If the pattern continues, high-quality original work will dry up and the internet will fill with low-grade machine-generated output. His proposed answer is a blockchain-based system of verifiable data ownership, where each use of data for inference, training or retrieval can be traced and compensated automatically. Rebuilding that contract, he said, is one of Frontier’s core missions.
What comes after the smartphone
On hardware, Horowitz said the smartphone has been an extraordinary industrial product, but its interaction model is not natural. People have to look down, pull attention away from the physical world and focus on a small glass screen.
He said the firm is especially interested in always-on wearable computing, particularly smart glasses and environment-aware hardware. A device with the weight of ordinary glasses, paired with a front-facing high-dynamic-range camera and directional audio microphones, would effectively give the user a companion AI that shares first-person context. It would know what the user is looking at, understand who the user is speaking with and provide assistance at the right moment without requiring a phone unlock.
Andreessen described that as a shift from tool-based computing to environment-aware intelligence. Spatial audio, retinal projection and haptic feedback are all maturing, he said. Future hardware will not just receive information; it will extend human cognition. He said the category could support a new hardware supply chain ecosystem worth hundreds of billions of dollars.
Advice for people who are 18 today
At the end of the episode, Torenberg asked what the three guests would study if they were 18 years old and just entering university or the workforce.
Andreessen said his first piece of advice would be to avoid standardized tracks that train people for routine work. If a field is built around fixed procedures and document handling, he said, large models may replace much of that work before students even graduate.
Instead, he urged younger people to return to the physical world and first principles: mathematics, theoretical physics, low-level computer systems, chemistry and materials. At the same time, he said, they should become highly fluent with the latest AI tools. In his framing, a young person who knows how to use generative systems and coding agents can now achieve the output that once required a 20-engineer software department. Shipping real products and learning from users matters more than credentials.
Horowitz added that leadership and responsibility remain scarce no matter how technology changes. He said younger builders should go toward the rough edges, learn to absorb failure, gather unusual talent and hold their intuition when the crowd dismisses them.
Dixon’s advice was to keep a hacker mindset and stay skeptical of overpackaged consensus. He placed open-source software 20 years ago, blockchain 10 years ago and today’s world of hard tech and agents under the Frontier banner on the same line of frontier-building. In his view, the next opportunity still sits where things look unfinished and rough.
