Sam Altman’s Contrarian Investing Lesson from Graham and Thiel

OpenAI CEO Sam Altman used a recent appearance on the Invest Like The Best podcast to share an investing principle he picked up from Y Combinator's Paul Graham and early Facebook backer Peter Thiel: the best opportunities rarely look attractive when they first appear. Following trends can produce decent returns, Altman said, but spectacular results almost always require backing something most people are ignoring or actively avoiding.

What happened

Detail	Fact
Podcast	Invest Like The Best
Mentors cited	Paul Graham (Y Combinator cofounder), Peter Thiel
Core claim	The best investment opportunities almost never look popular at the time
Startup investing analogy	Flight instructor: hands-on guidance over lectures
Compute scramble context	After realizing GPT-4’s exponential improvement curve
Fundraising parallel	“Most people tell you no, but all you need is one or two yeses”

Speaking on the podcast, Altman put it plainly: “The very best companies, the very best investment opportunities are almost never the ones that look really popular.” He added that following a trend and getting in slightly early can produce adequate returns, but doing “spectacularly well” requires going against the crowd.

Both Graham and Thiel are cited as the source of this view. Graham, cofounder of Y Combinator (the accelerator behind Airbnb, Stripe, and Dropbox), reportedly framed startup investing as analogous to flight instruction. A good flight instructor sits next to you, points out mistakes in real time, and helps redirect your energy. Lectures alone do not work.

Why it matters for founders and operators

The lesson has a practical edge for anyone raising money or deploying capital today. Altman connected it directly to OpenAI’s own experience. After the team recognized GPT-4’s steep improvement curve, they urgently needed compute. They called cloud providers, chip fabricators, and energy companies. Most said no. A handful said yes, and that was enough.

The parallel to early fundraising is deliberate. Most investors will pass. Chasing consensus validation is a slow path. If you are building something that looks weird or premature to the majority, Altman’s framing suggests that is a feature, not a bug.

For businesses exploring AI integration right now, the same logic applies. The tools that look niche or unproven today, before enterprise adoption normalizes them, are often where the productivity gap opens up. Waiting for something to look popular usually means waiting until the advantage is gone.

The ChatGPT parenting tip that backfired

Separately, Altman posted on X suggesting parents use ChatGPT Work to generate personalized daily podcasts for their children’s morning school commute. The idea involved connecting family calendars and kids’ personal interests to produce audio summaries covering things like an upcoming soccer game, a sibling’s birthday, and current news.

The response was swift. Alex Hirsch, creator of the animated series Gravity Falls, replied with one line: “What if you just talked to your children?” Altman’s original post received roughly 4,000 reposts and 12,600 likes. Hirsch’s reply pulled in over 18,000 reposts and 197,000 likes, according to Times of India reporting based on Business Insider.

The ratio tells a story. Even among people who use AI tools daily, there is a clear line between genuinely useful automation and automating something that was never the problem to begin with.

Our take

The contrarian investing point is not new. Thiel made a version of it in Zero to One a decade ago. What’s worth noting is that Altman is saying it now, at a moment when AI is extremely popular and capital is flooding in. The implicit message: most of the money chasing obvious AI bets will do “okay.” The interesting returns are in the things that still look strange.

The parenting tip episode is a useful counterpoint. Altman is genuinely good at identifying where AI creates real leverage. He is also capable of suggesting AI for situations where the answer is just: be a parent. The two things coexist. Good judgment about which category a use case falls into is the actual skill, and no model ships that for you.

If you are evaluating where AI fits in your own business, our guide to open-weight AI models for business leaders is a practical starting point for understanding what is actually ready to deploy versus what is still hype.

What to do about it

1. Audit your current AI shortlist: if everything on it is already mainstream, you are probably pricing in the trend, not ahead of it.
2. Identify one workflow that looks too early or too niche to bother with, and run a small test anyway.
3. Apply Graham’s flight-instructor standard to any advisor or agency you work with: are they sitting next to you correcting mistakes, or sending slide decks?
4. Before automating any communication or relationship task, ask Hirsch’s question first.

The edge is usually in the thing that still looks a bit weird.

Source: Bing News · OpenAI