Horowitz says software's laws of physics have changed
At a16z's Connect/Fintech conference on April 14, 2026, Ben Horowitz argued that AI has rewritten the rules under software — collapsing moats, compressing product runway to weeks, and making US infrastructure the real constraint.
Ben Horowitz told a room of founders on April 14, 2026, that the runway a good software product used to enjoy — a decade, then five years — has compressed to “maybe five weeks.” Speaking at a16z’s invite-only Connect/Fintech conference in Park City, Utah, he argued that operators still running the old SaaS playbook “are definitely going to die.”
The talk, published to a16z’s YouTube channel and confirmed in same-week reporting by Fortune, was framed around what Horowitz called “AI anxiety” inside Silicon Valley — the specific fear, among founders and operators, that the playbooks that built the SaaS era no longer apply. It is the sharpest public articulation yet of a thesis that has been trade-press shorthand for most of the past year: that software’s terminal value is up for renegotiation.
The tension sits in the gap between how AI feels at the front of the curve and how much of the economy has actually adopted it. Goldman Sachs economists, per Fortune’s same-week writeup, put US establishment adoption of AI below 19 percent.
The moats, Horowitz argues, are gone
Horowitz’s core claim is that three classic sources of software defensibility — migration cost, data lock-in, and user-interface habit — have all been undermined at once. Code, he said, is now easy to replicate. Data is easy to move. And the user is increasingly not a human at all but an agent that is “really flexible on how they use user interfaces.”
He paired that with a related argument about headcount economics. The mythical man-month, Fred Brooks’s 1975 observation that throwing engineers at a late software project makes it later, no longer holds when the work is done by models. “If you have enough money and some good data,” Horowitz said, “you can buy enough GPUs and solve basically anything in software.” The binding input, in his telling, is capital, not coordination.
If you keep looking at it like the old world and it’s got completely different laws of physics, you are definitely going to die.
That is also, he conceded, the mechanism behind what the industry has taken to calling the SaaS apocalypse. “The reason why the SaaS apocalypse is happening is because there are doubts on terminal value,” he said — doubts that now land as valuation events rather than slow decade-long declines, because companies are staying private longer.
Infrastructure as the real constraint
Horowitz spent a substantial portion of the talk on the physical side of the stack. “America’s got to rebuild its entire infrastructure like right now,” he said, listing rare-earth minerals, electricity, manufacturing capacity, chips, and memory as simultaneous bottlenecks. “Nvidia will make enough chips, but then we won’t have enough memory. Almost everything is the bottleneck.” On power specifically, he was blunt: “We’re pretty much out of electricity now in the United States. Like not 12 months from now, like right now.”
He connected that diagnosis to a16z’s own posture. The firm raised $15 billion across five funds in January 2026, Reuters reported at the time, including a $1.7 billion AI infrastructure fund and a $1.12 billion American Dynamism fund. Horowitz referenced the raise on stage and said a16z had backed a physical power-transformer company — which he did not name — on the grounds that the transformer itself “hasn’t changed since we invented electricity.”
Crypto re-entered the talk as a narrow bridge off the infrastructure thread: proof-of-human, cryptographically signed content, and payment rails for agents acting as economic actors. It was a subplot rather than the main argument, but it fit his larger point that the next stack has to be rebuilt at the physical and protocol layers together.
What to watch
Horowitz closed on an anti-dystopian note, reaching for the industrial revolution and electricity as parallels and predicting that in 15 years Americans “will live better than the very best life” of 1980. The near-term test is narrower: whether the compression he described at the front of the curve — moats gone, runway in weeks — shows up in the adoption numbers Goldman is tracking at the back of it. The next few quarters of enterprise AI spend, and the first public earnings calls from legacy SaaS incumbents in this cycle, will be where the two curves either converge or keep diverging.