AI safety concerns and how it could impact Anthropic's IPO, the markets, and more. Former SEC chair Gary Gensler, he's currently a professor at MIT and the co-host of Power and Consequences. It's a podcast you should go on and download right now. Gary, it's great to see you. We were just having a riveting conversation with Steve Eisman who has a view that if in fact he thinks all of this is fake. I think he's with the president sort of in this sort of safety issues being a hoax. But more importantly his view is if Anthropic really is as concerned as they say they are, they should cancel or delay or reschedule their IPO. If you were in the seat running the SEC right now and this IPO was before you, what would you be telling them?
I would be telling them to work with the staff of the Securities and Exchange Commission to keep the politics out of it. And it's really a question whether they're making all the right disclosures and that they're allowed to go public as long as they tell the investors what the material risks are. But the material risks could be real and potentially the question is also what the projections around the company could ultimately be.
Oh, absolutely. But their most important risk are really that they and OpenAI are building on a premise that there'll be enough revenues coming in in the future, right? And they're competing with China that has much cheaper, almost as good foundation models. So Sam Altman told Fortune magazine last week that he believes this is not the right time for AI companies to go public. He didn't seem to be just suggesting, and that's true of his company obviously, and he may have a motive in that he may not be ready to go public until 2027. But the suggestion was that Anthropic shouldn't be either. What do you think of that?
I think that's competitive talk and that OpenAI might not be as ready. But even Anthropic is spending so much money, billions of dollars, Andrew, that they want to tap that capital market. And they saw what happened at SpaceX. So they want to catch that uplift in the markets.
Look, there's a big AI trade going on and it is unlikely to last. They're unlikely to be able to tap the markets as successfully a year, a year and a half from now.
So you think it really will be harder? Yeah, I do. I do.
Does that mean whatever valuation this thing comes public at today will not be its valuation in a positive way?
Let me more broadly say we're going to have a retrenchment at some point in time. You worked on a book about the 1920s and that build out and the retrenchment that was on a lot of other technologies, electricity, radio, consumer products. We had the internet bubble that there was a retrenchment. At some point in time, this huge capital expenditure build will plateau and reverse. And I think it's just a question of when and how steep.
Gary, we're coming off two weeks of talking about Skynet and talking, we had our researcher come and say that the end of humanity could be by 2030. So your biggest risk in the IPO is that there may not be enough money for the buildout. Give us some red meat, will you? The biggest risk is that the planet and all living things on it ends, doesn't it? Would you have you ever had a company where that would need to be in the IPO?
Listen, Dario Amodei just wrote about it on Saturday. So you take that language and you put that language inside the document.
How would you put that in the IPO that we may there may be a risk to human existence?
Well, it's a question of what's the risk ultimately.
You believe in you sound like Trump. You don't you don't really buy into any of that, do you?
No, I think there's risk. I think there's risk. I think that I don't know if you love you know science fiction and everything but sort of and it sort of borders a little bit on Star Trek when Kirk said hey I don't like the test was Kobayashi.
Oh yeah you can't it's an unwinnable test.
An unwinnable test, let's do something different. That's what happened to OpenAI. Similar, that's what happened to OpenAI and they deployed this model they thought in a sandbox and it got out of the sandbox, found its way on the internet and disrupted the technology at HUD.
It's so early that that was serious enough. What happens you know when the capabilities of these.
The capabilities are likely to grow but I would say this from the capital markets point of view and I know you're going to be surprised at this with the capital markets point of view, you don't want your government putting a thumb on the scale. You let investors, as long as they're not missing out on the material information, the risk, and you don't defraud the public.
What's the legal liability on something like this, though? How do you put that into the perspective?
Oh, there's real potential legal liability. Whether you're JP Morgan or whether you're OpenAI, if you deploy a model and the model disrupts other companies, you do have some liabilities, right?
But don't you think that a class action lawyer, I mean the fact that these statements have already been made publicly, we could destroy humanity, that quote is going to be used in court repeatedly over and over and over again.
Product liability or by the way a company deciding to sue OpenAI or an individual deciding to sue or a class deciding to sue.
Look, if you were the underwriters for Anthropic you might say Dario this might not have been the best time to say these things.
Well, but he's been saying them for the beginning of time.
But you can go public at that same point in time. If I was his investment banker, I'd say you want to catch this part of the market because at some point in time, the market's going to reprice. And it comes down to this, Andrew. The Chinese are spending about a hundred billion dollars a year as an economy building the data centers. We're spending 6700 billion dollars maybe going to a trillion dollars next year, right?
And those models from China are 2 to 6 months behind us. And so more and more US companies are going to say we're going to use the Chinese models rather than Anthropic or OpenAI. And so that's a real competitive thing. You had Steve Eisman here. He said it's a whole chain. But when users use the Chinese models and they get something good enough, look, not everybody needs a Maserati, they might just need a Ford or a Chinese BYD car. That's that's the that's.
We don't allow a lot of those Chinese models here. When you're talking about cars in.
Actually, they're here. Microsoft hosts them, Google hosts them, Amazon.
That's what I mean. If I'm talking about when you're talking Chinese models for these cars, we have said we're not going to allow.
