Investor Ben Horowitz speaks during the opening of the DLD (Digital-Life-Design) conference in Munich, Germany, 18 January 2015. At the Hubert Burda Media innovation conference, high-ranking guests speak about trends and developments of digitization for three days. PHOTO: TOBIAS HASE/dpa | usage worldwide (Photo by Tobias Hase/picture alliance via Getty Images)
picture alliance via Getty Images
The rules that made software companies defensible for a decade have collapsed in roughly 18 months. In a wide-ranging interview, Andreessen Horowitz co-founder Ben Horowitz argued that AI has simultaneously destroyed traditional software moats, compressed product lifecycles from years to weeks, and created an identity crisis so severe that blockchain infrastructure has potentially become a structural requirement. The implications for public markets, private valuations, and the $280 billion that flowed into North American startups last year are severe and will take place shortly.
In January 2026, the firm closed $15 billion in new funds, its largest haul ever, including $1.7 billion specifically for AI infrastructure and $3 billion for additional venture strategies. The fundraise arrived against a backdrop where startups in North America raised $280 billion in 2025, up 46% year over year per Crunchbase, with the majority of that capital directed at AI. Horowitz framed the fund's thesis plainly: "Our mission is ensuring that America wins the next 100 years of technology. That starts with winning the key architectures of the future, AI and crypto." The pairing is not rhetorical.
For two decades, enterprise software operated on two reliable axioms. First, you could not simply outspend a competitor into existence: adding engineers to a late project made it later, a principle known since Fred Brooks codified it in The Mythical Man-Month. Second, data lock-in, interface familiarity, and migration friction created durable competitive moats. Horowitz's position is that both are functionally gone. Sufficient GPU access and proprietary training data now let a well-capitalized challenger replicate almost any software product rapidly. AI agents can navigate interfaces and port data at a pace that eliminates the switching costs incumbents relied on. What once took a new entrant three years can now take three months or less.
The consequence is a compression of product lifecycles that public markets are struggling to price. Horowitz described what he calls a "Saaspocalypse," where the terminal value assumptions embedded in software multiples are being revised downward. A product with a five-to-ten-year competitive runway may now have five weeks. The rational response for founders is to stay private: private companies can pivot, cut costs, and restructure AI strategy without the immediate penalty of a quarterly earnings call.
The AI buildout is colliding with physical limits that no software fix can resolve: shortages in rare earth minerals, memory chips, and most acutely, electricity. The US grid relies on power transformers whose core engineering has not materially changed since the system was built. Rebuilding that infrastructure to support AI compute demand is not a 12-month project. The Department of Energy has projected that data center electricity demand could double by 2030, driven largely by AI workloads. a16z's infrastructure fund thesis is a direct bet on this gap: the firm positions AI infrastructure broadly to include foundational models, networking security, and applications targeted at technical buyers, not consumers.
The least intuitive part of Horowitz's argument may be the most consequential for investors. He contends that the proliferation of AI-generated content, deepfakes, and hyper-personalized spam has made traditional digital communication structurally unreliable. In a recent interview, Horowitz cited an estimated $450 billion lost to fraudsters during US pandemic stimulus programs as a baseline for what happens when identity verification fails at scale. The solution, in his framing, is not a new centralized database run by Google or the US government but cryptographic verification with "mathematical game-theoretic properties" that do not require trusting any single institution.
The second half of the convergence thesis is economic agency. For AI agents to function as autonomous actors, they need a native transaction layer. Credit cards require human authorization. Bank accounts require legal entities. Crypto, specifically stablecoins and bearer instruments on public blockchains, provides the infrastructure for an AI to receive payment, pay for services, and operate economically without human approval at each step. Columbia Business School researchers summarized Horowitz's position this way: in a world where AI agents need to purchase goods or prove authenticity, blockchain offers tools that traditional financial systems cannot.
On job displacement, Horowitz is skeptical of confident predictions in either direction. In a February 2026 interview, he argued that the assumption underlying mass-unemployment forecasts, that tomorrow's job categories are predictable from today's, is historically unsupported. He pointed to agriculture, where nearly all jobs were eventually automated, yet the population shifted into roles that would have been impossible to anticipate from within an agrarian economy. "The idea that we could imagine all the jobs that are going to come, sitting here now, that AI is going to enable, I think is low," he said. That position puts him to the optimistic side of a debate where figures including Anthropic CEO Dario Amodei and computer scientist Geoffrey Hinton have warned of significant displacement.
Horowitz's framing posits that removing the capital and technical barriers to creation means that the total addressable market for new products expands to anyone with an idea. The constraint shifts from access to capital or code to quality of judgment. Whether that creates the abundance Horowitz predicts or concentrates value in the infrastructure layer, as some critics argue, depends on which bets the current wave of multibillion-dollar funds turns out to have placed correctly.
Three practical conclusions follow from Horowitz’s framework. Founders in traditional SaaS need a credible answer to the moat question that does not rely on switching costs or data lock-in, because neither holds. Infrastructure bets, in electricity, chips, memory, and physical manufacturing, are no longer niche theses; they are a key part of the bet that the AI wave continues. And the AI-crypto convergence is no longer a dream: with US stablecoin legislation advancing, the plumbing for AI-native financial infrastructure is being built in real time. Investors who treat these as three separate sector bets may be misreading how tightly they are coupled in the transition Horowitz is describing.
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。