Lisa Su Tells MIT to Run at Hard Problems. Her Own Bet Returned 168x — MIT 2026

Lisa Su’s MIT address turns on a single line she was given at 25, working at IBM and wondering how anyone makes a difference inside a company of hundreds of thousands.

The advice

“One of my mentors told me something that I’ve never forgotten. Run towards the hardest problems. At the time, I’m not sure I really knew what that meant, but I now realize this was the best advice I’ve ever received.”

She then applies it to her own biggest decision, and is unusually candid that the people around her disagreed:

“12 years ago, I got a chance to put that lesson to the test. I had the opportunity to become CEO of AMD. AMD had a lot of potential, but the company had been through a few tough years. And some of my mentors thought taking that job was actually kind of risky.”

The closing frame is that outcomes like that are not luck in the passive sense:

“Luck is not just being in the right place at the right time. It is taking the risk to work on something really hard.”

What that bet actually returned

Su became AMD’s CEO in October 2014. The advice is delivered with the outcome already known, so the outcome is worth stating precisely rather than gesturing at.

Date	AMD close
October 2014 (Su becomes CEO)	$2.80
December 2016	$11.34
December 2020	$91.71
December 2024	$120.79
August 2026	$470.72

That is roughly 168x, or about +16,700 percent, over not quite twelve years. There is no reading of “kind of risky” under which that was the expected outcome in 2014, and the mentors who advised against it were not being stupid — they were pricing a semiconductor company that had been losing to Intel for a decade.

Why the advice is still worth taking

The survivorship objection is real and it is also not fatal, for a reason the speech itself supplies.

Su does not actually argue that hard bets pay off. She argues something narrower and more defensible — that hard problems change what you are capable of, independent of outcome:

“Hard problems really teach you what you’re capable of.”

And her account of graduate school is about the failure loop, not the win:

“I remember spending weeks in the clean room, fabricating devices, and then bringing my wafers up to the test lab, only to discover they didn’t behave the way I expected at all… little by little, I went from a new grad student learning about the field to someone doing original research.”

The claim there is about the confidence to proceed without the answer:

“Not the confidence that I would always know the answer, but the confidence that even when I didn’t know the answer, I could figure it out.”

That version does not need the 168x to be true. It survives the counterfactual where AMD failed, which is the test any career advice has to pass.

The part aimed at people entering an AI industry

The most quotable passage is not about careers at all. It is a hardware CEO drawing a boundary around what her own products cannot do:

“For everything that AI can do, AI can’t decide which problems are worth solving. It can’t make the hard judgments when the data is not there. It can’t take responsibility for the outcomes.”

Coming from the person selling the accelerators, that is a more interesting statement than the same sentence from a critic. It is also the connective tissue back to her main argument: if choosing the problem is the part that stays human, then “run at the hardest problems” is advice about the one skill that does not get automated.

What to take from it

Take the process claim, discount the outcome claim. Hard problems compound your capability whether or not the specific bet lands, and that is the part of Su’s argument that holds for someone without a semiconductor turnaround in their future.

Then apply her own framing honestly. She says the best people “find ways to make their luck.” The AMD number says something slightly different: she made a bet with a wide distribution of outcomes and got a very good draw from it. Both things are true, and only one of them is repeatable.