Idea · carried by 1 source
AI Value Accrues to Distribution, Not AI Companies
The biggest winners from AI won't be the AI labs or model companies — just as the biggest winners from railroads weren't the railroad companies. Value accrues to whoever has the distribution and cost structures to harness the new technology. The fattest, most inefficient brands will be the biggest beneficiaries.
Where it comes from
William Hockey, analyzing where AI value creation will land, in highlights/william-hockey-bet-on-yourself (~1:06:00): “I actually don't think AI companies are very set up for success. That's actually probably not where value is going to accrue. The value is going to accrue in two areas... Do you have massive distribution? Because if you have massive distribution, if you have massive costs, AI is going to be a massive benefit to you.”
How it works
- The railroad analogy: Everyone thinks the railroads captured the value from railroads. But “the value in railroads accrued to the oil companies... Standard Oil was the biggest beneficiary of railroads by an order of magnitude.”
- The tech parallel: Value accrued to ISPs, mobile phone makers — but “the value actually accrued to the people that could harness that the best.”
- Who wins with AI: “The biggest, fattest, most inefficient brands are going to be the best beneficiary of AI. Because brands have a massive moat, and man, there's a lot of cost to cut there.”
- Financial services specifically: “Financial services and legacy banks tend to be the largest inefficiencies out there. They also are very hard to take over.” Banks have massive headcount and technology costs ripe for AI optimization, but “the business model's too good” so they've been slow.
- Railroads are a bad AI story: “99% of your money is spent on fuel, track maintenance, and people... you take conductors down from 1,000 to 500, it doesn't change the equation at all.” Banks, by contrast, are “pretty head count focused and pretty technology focused.”
Don't build an AI company. Build a company with massive distribution and a fat cost structure, then use AI to make it dramatically more efficient. The value is in what AI enables, not in AI itself.