Inside General Catalyst’s $1.5B AI Roll-Up Machine

General Catalyst bets $1.5bn on AI-run services roll-ups

General Catalyst is betting that AI makes it cheaper to buy services businesses than to sell software to them. Services are a $16tn market, fragmented and sticky, and most of these companies earn thin margins because humans do the work. Automate 30% of the tasks and margins move from the mid-teens toward 40%, so the firm incubates a company, pairs an industry operator with AI engineers, funds it to acquire rivals, and holds for seven to ten years toward a public listing.

Takeaways

• General Catalyst put $1.5bn into AI roll-ups. General Catalyst put about $1.5bn into its Creation strategy in the last fund cycle. It incubates applied AI companies and, in roughly half the cases, funds them to buy distribution, client lists and data.
• Services are a bigger prize than software. Bhargava puts the global services market at $16tn against $1tn for software. Services businesses historically break even or earn 15 to 20% EBITDA margins, and automating their work can lift those margins to 30 or 40%.
• Four buckets of work can be automated. General Catalyst targets customer support, data entry and evaluation, content and marketing copy, and basic logic and reasoning, aiming to automate at least 20 to 30% of tasks. It looked at 70 services industries and rejected 60 of them.
• Crescendo turns call centre margins around. Built with Andy Lee, who ran Alorica for over thirty years, Crescendo automates 50 to 70% of call centre work. That takes a business earning 10% EBITDA margins to roughly 40%.
• Holds run seven to ten years toward IPO. Unlike private equity, General Catalyst adds engineers and LLM subscriptions rather than debt and cost cuts. It cites Danaher and Constellation Software as compounder templates and predicts plenty of $100bn AI-native compounders in public markets within a decade.
• It leads early rounds and keeps 20-25%. General Catalyst invests $100m to $150m per project across three or four funding rounds, leading at least the first two and bringing in other investors later. Founders should still own 10% at IPO, ideally 20 to 30%.
• AI makes teams more powerful, not smaller. Bhargava cites Hippocratic AI's nurse software, where one nurse oversees five agents, and points to shortages of nurses and accountants. He expects outsourcing hubs in India, the Philippines and Mexico to be hit hardest.

How General Catalyst builds a roll-up

1. Pick a fragmented industry where selling software is hard and contracts stick, such as HOA management with its two-year agreements, accounting, insurance or managed IT services.
2. Partner with an operator who ran that industry, as it did with Andy Lee at Alorica, and staff the company with AI-native engineers from Rippling, Scale, Stripe or Figma.