Why 95% of AI Startups Are Building on Quicksand

The Brutal Reality

Harrison Miller, who helped build Amazon's first marketplace and invested in 550+ tech companies at Summit Partners, drops an uncomfortable truth: "Building an AI startup is 95% company building and 5% AI." While founders obsess over model performance and GPU clusters, they're ignoring the fundamentals that determine whether any company survives. Miller's track record—from launching Amazon's first non-book category to creating what became AWS—gives him unique authority to call bullshit on today's AI hype cycle.

The Post-Check Playbook

Lesson 1: Your Unit Economics Will Kill You Before Your Competition Does

Miller learned this the hard way at Amazon. After successfully launching the toy category and capturing market share, he found himself sitting on $60 million of rapidly depreciating inventory. "Toys in January are like rotting fruit," he recalls. The breakthrough came from doing something most AI startups skip: actual unit economics analysis. They discovered the brutal math—receiving, shelving, picking, packing, and shipping individual items would never be profitable at scale. This led to the Toys "R" Us partnership that flipped their business from burning $60 million a ually to making $60 million. The parallel to AI is stark: most startups are burning cash on compute and data acquisition without understanding their true cost structure.

Lesson 2: Talent Variance Is Your Biggest Risk

"The reason Amazon was so successful is that the mean level of talent there was very high and the variance around the mean very low." Miller's investing career reinforced this truth—every good decision came down to A-players versus B-plus players who looked good on paper. In today's bubble environment with inflated valuations and crazy compensation, the temptation is to hire fast. But Miller warns that character matters as much as capability: "It's how people perform in the foxhole when it's really going poorly that can be the make or break."

Lesson 3: Platform Shifts Create False Confidence

When Amazon's stock went from 20 to 120 in three months, Jeff Bezos constantly reminded the team: "You are not six times smarter than you were two months ago." Two years later, the stock hit 6. Miller lived through both extremes, watching as "all my staff's moms were calling, saying 'Go work at Boeing.'" Today's AI entrepreneurs face the same dynamic—$100 billion in venture investment last year alone, up 80%. The bubble creates an illusion of progress that masks fundamental business challenges.

What Actually Works vs. VC Mythology

Miller's framework for building in bubble environments cuts through the noise. First, embrace constraints through clear principles—Amazon's six core values weren't suggestions, they were hiring and firing criteria. Second, focus on creating new business models, not just applying new technology. The Toys "R" Us deal wasn't about better algorithms; it was about fundamentally rethinking who owns inventory risk. Third, prepare for the reversal. "For every poorly researched article about what geniuses we are, there's going to be a poorly researched article about what idiots we are. Neither are true."

The Unfair Advantage

The real insight from Miller's 25-year journey from operator to investor isn't about AI at all. It's about recognizing that technological revolutions—whether internet, mobile, or AI—follow predictable patterns. The wi ers aren't those with the best technology but those who nail the fundamentals: rigorous talent selection, honest unit economics, and the discipline to build sustainable business models when everyone else is chasing the next shiny object. As Miller puts it: "If the idea is a quick flip, yeah, there's some opportunities there. But if the idea is to build an enduring, great company, in many ways it's actually harder because of the pace and the hype and the frenzy of it all."