Dan Sundheim says late-stage private companies now offer many interesting opportunities, since fewer investors scrutinize them than public stocks.
D1 Capital originally invested in OpenAI's $125bn round when many doubted LLMs could work as a business, and Sundheim now puts the odds of an economic return at roughly 70-30.
Sundheim expects hyperscalers AWS, Azure and GCP to become a worse business model over time, while systems of record like ERP software stay hard to displace.
D1's Sundheim on Why AI Links Private and Public Markets
Dan Sundheim argues that private and public investing now feed each other, and AI is the clearest case. D1 backed OpenAI and Anthropic early because it saw durable, high-margin products, even though their capital needs have no precedent. The open question is whether scaling laws keep returns attractive.
Takeaways
Private and public investing now reinforce each other. Public markets are the most competitive in the world, while private markets draw fewer eyes on each situation. About 25% of private companies D1 has looked at since its founding had synergy with its public investing, and AI has made that overlap greater than ever.
LLM companies look like durable, high-margin businesses. Sundheim's early odds were roughly 65-35 to 70-30 that these businesses would earn an economic return, and the debate has moved on. Gross margins are high, API commoditisation matters little, and four or five providers should matter long term.
Capital intensity is the open risk for LLMs. These firms are more capital-intensive than anything in business history, so heavy financial and operating leverage cannot absorb two or three years of slow progress. He thinks betting scaling laws stop is unlikely, and expects large free cash flow within five to ten years.
Conviction came from founder clarity, not consensus. D1 entered OpenAI in the $125bn round when few believed in LLMs as a business. Anthropic was compared to Lyft against Uber. Dario Amodei's essays, which Sundheim likens to Bezos' 1997 shareholder letter, gave him the conviction.
Focus beats doing everything. OpenAI pursues a do-everything approach, while Anthropic went all in on enterprise after finding no consumer traction. Sundheim says he would err toward focus, and he pushed OpenAI to add ads about a year and a half earlier.
Hyperscalers and software face worse economics ahead. AWS, Azure and GCP may accelerate near term as AI labs grow, but are a worse model going forward because LLM companies may eventually insource compute. Software also weakens, though systems of record will be hard to displace.
SpaceX exceeded expectations while Rivian disappointed. D1 invested similar amounts in both. Rivian did not deliver the planned return. SpaceX surprised on the upside, with Starship a potential game changer that could cut launch costs about 97%, and its satellite engineering also impressed him.
Risks and views outside AI
Sundheim calls a collision over Taiwan and semiconductors the largest risk. He cannot think of a scenario in which China, Taiwan and the US are all happy, and says an unresolved issue could bring something akin to the Great Depression.
Tap a name to go there · tap a line to see why they are linked
On this episode 2
Show 1
Replicating the semiconductor supply chain carries its own risk. The US becomes less likely to defend Taiwan, so China attacks anyway.
He calls shorting a bad business that requires being intellectually stimulated by it, and says hedge funds are not a good business.
Europe is economically stagnant and the most inefficient market, which makes it easier for pure fundamental stock pickers. Germany, Korea and Japan have companies with excellent physical assets.