We judge leaders by what’s visible now, not by what they set in motion years ago.

Dan Luu wrote a piece in 2024 about Steve Ballmer being underrated. The thesis was simple: the bets that made Nadella look like a genius were placed by Ballmer. Azure, Office 365, the enterprise sales machine. Ballmer took the criticism for years while building infrastructure his successor would harvest. His record shows how consistently we misjudge foresight in real time.

Jensen Huang spent years building CUDA when GPUs were for gamers and researchers nobody talked to. The bet looked niche until it didn’t. We retroactively call it visionary, but at the time it was a weird fixation on parallel computing that analysts questioned repeatedly.

Sam Altman pushed OpenAI toward scale when the AI safety community wanted caution and the tech community thought transformers were interesting but not revolutionary. The bet was that capability would arrive faster than anyone expected and that being at the frontier mattered more than waiting for consensus. Whether you think that was wise or reckless depends on your priors, but you can’t deny the foresight about timing.

Real foresight often looks wrong. A bet that looks obviously correct is a consensus bet, which doesn’t produce outsized returns.

This creates an uncomfortable problem for evaluation. The leaders making the best long-term decisions often look the worst in the present. They’re spending resources on things that seem irrelevant. They’re ignoring what everyone says they should focus on. The feedback they get is negative because the feedback loop for strategic bets operates on a different timescale than quarterly earnings.

Microsoft’s internal politics under Ballmer killed promising products before they could mature. But Ballmer also cleared the worst political actors before he left, giving Nadella a cleaner board. That’s not a decision you get credit for. It’s invisible infrastructure. The absence of obstacles rarely makes headlines.

The comments on Ballmer’s retirement revealed something else. Pundits kept saying “he should have done X” where X was something that was only obviously correct in retrospect. Should have bought Google. Should have killed the iPhone. Should have seen mobile coming. That’s hindsight cosplaying as strategy.

Foresight requires positioning. You don’t need to know exactly what’s coming. You need to be building capabilities that give you options when it arrives. Ballmer’s Azure bet wasn’t “I predict cloud will be huge.” It was “I predict Microsoft needs to be able to compete in whatever the next platform is, and cloud infrastructure gives us that option.”

Apple’s restrained AI bet is another version of this options argument.

We can’t evaluate foresight in real time. By definition, we don’t have the information yet. The best we can do is notice who’s making bets that look wrong but have internal logic, and who’s making bets that look right because they’re following consensus.

Those unconventional bets can still fail. Their appeal is the chance of being right before everyone else.