Lazard CEO Peter Orszag Talks AI in Banking

Peter Orszag · Lazard

AI data-center build-out is larger than past infrastructure waves and is pushing interest rates higher.

Orszag puts AI infrastructure investment at roughly 3 to 3.5% of GDP. He says that exceeds the scale of railroads, highways and electrification by a significant margin. He ties this to upward pressure on interest rates, since the build-out absorbs large amounts of capital and borrowing.

“investment, which is larger than railroads, highways, electrification by a significant margin, that's also putting upward pressure on interest rates.”at 3:22

Peter Orszag · Lazard

Debt-financed AI build-out carries more financial-distress risk than people appreciate if token spending falls short.

Orszag separates the technology, which he is a huge believer in, from the business model. Because so much AI infrastructure investment is debt-financed rather than equity-financed, the ecosystem is exposed. If token spending does not grow as the debt expects, borrowers face real problems. He says the risk of financial distress here is higher than people appreciate.

“I do think the risk of financial distress in this ecosystem is higher than people appreciate because when it's debt financed, if the token spend instead of doing this... Just does that, you've got a problem.”at 3:49

Peter Orszag · Lazard

US inflation is in a 'supply side era 2.0' of recurring supply shocks, so treating them as one-offs is a mistake.

Orszag argues that to understand US inflation today, one must look at supply shocks from a year or two ago. He concedes a piece of research suggests one-time effects will erode and inflation will tail off in the next few months. His main point is that new supply shocks will keep arriving, so the economy is in a 'supply side era 2.0' in which shocks hit on an ongoing basis. Treating each one as a one-off, as many analysts do, is a mistake.

“we are living in a new supply side era... there are going to be continual new supply shocks. So we're living in this era, supply side era 2.0, in which you're going to be hit by these things on an ongoing basis, and so treating them each as a one off thing is mistake.”at 1:30

Peter Orszag · Lazard

Rate environment has shifted, bond vigilantes are returning, and Treasury maturities should have been extended when rates were low.

Orszag says the interest rate environment today is different from three or four years ago. He regrets that the government did not use the earlier low-rate period to extend the maturity of Treasury debt, a step he says he, Bob Rubin and Joe Stiglitz had called for. Instead, the Treasury has been shortening maturities. He concludes that the economy is in a new era in which bond vigilantes, investors who push back against government borrowing, are returning.

“I definitely think that the interest rate environment is different today than it was three or four years ago... I think we are in a new era in which, the bond vigilantes are coming back.”at 2:34

Peter Orszag · Lazard

AI technology is advancing astonishingly fast; the business model and its financing are the real question.

Orszag says we need to separate the technology from the business model. He is a huge believer in the technology and says he runs 70 or 80 AI agents as we speak. He is astonished by how fast it is progressing. His concern is with the business model, specifically the debt financing of the build-out and whether token spending will meet expectations.

“we need to separate the technology and the business model here. I'm a huge believer in the technology. I use it I probably have 70 or 80 agents running as we speak. I'm astonished by how rapidly things are progressing.”at 3:34

Peter Orszag · Lazard

Open-weight models and local compute could cut token spending on closed AI models, threatening debt-financed AI build-outs.

Orszag argues that a drop in token spending is the key risk to debt-financed AI infrastructure. He says that shift could come from open-weight models, which have already been discussed at length. It could also come from local compute, where models run on the user's own hardware, sometimes air-gapped. He believes people are not appreciating this second channel.

“going from this to that could come from open weight, which we've already talked a lot about, but also from local compute. I think that's the part that people are not appreciating”at 4:06

Peter Orszag · Lazard

Social Security and Medicare financing problems will have to be addressed, but only after a crisis forces action.

Orszag says Social Security and Medicare both have looming financing problems that will have to be addressed. He puts them within an overall fiscal imbalance of 6 to 7% of GDP. History shows, he argues, that reform comes only once a crisis is imminent. He cites the 1983 Social Security reform, which came only when checks were about to stop going out, and says there is no such forcing event yet.

“Both Social Security and Medicare have looming problems associated with their financing. They will have to be addressed. This is part of an overall fiscal imbalance of six to 7% of GDP... history shows we don't deal with these problems until they're crises.”at 5:38

Peter Orszag · Lazard

AI will shift junior banker training toward simulators and trust-building, away from repetitive slide work.

Orszag argues that as AI develops, some tasks will be valued less and others more. He points to a book, Messy Jobs, on which skills will hold value. Senior banking is a messy job that requires trust and in-person skills, which Lazard wants to cultivate. He says repetitive grunt work teaches little, and that new technology, like flight simulators for pilots, can train junior bankers while still giving them enough real at-bats.

“the hundred and seventy sixth time that you're moving the boxes around on a PowerPoint slide, you're not learning that much, and so the question becomes how can we train people?”at 8:30