Anthropic, the artificial intelligence (AI) company behind Claude, is preparing to go public as soon as November in what will likely be the biggest IPO ever. 

But there’s a complication: Some of its employees say the technology could threaten humanity. Their own alignment science lead said there's a greater than 10% chance AI wipes out humanity within 10 years.

The warnings have sparked debate about AI safety, regulation, and whether the boom has gone too far. They’ve also raised a question for investors: Is this a reason to get out of AI stocks, or an opportunity to buy while others get nervous?

I (Chris Reilly) put that question to three of our analysts on our latest RiskHedge Reserve quarterly call. Here’s what they had to say.

***

Stephen McBride: “New technologies are always met with fear...”

If you go back through history, new technologies are always met with panic and fear about them taking all the jobs, killing us, and rotting kids’ brains. 

The Pessimists Archive on Substack goes through how people freaked out when the Walkman came out, when pagers came out, when the bicycle was invented, even when the calculator came out. These concerns are always present when new technology arrives.

AI stocks did go down on this latest round of warnings, but many of them are back to all-time highs. AI is going to save millions of lives. It will create many, many times more jobs than it destroys.

I think Anthropic makes a truly wonderful product. I use it all the time. But I think its warnings also fit a broader strategy of regulatory capture: “AI is so dangerous that it must be controlled by a few people, and we happen to be those people.” That helps Anthropic against competitors.

If I have to pay Anthropic $210 a month to use Claude, that’s wonderful for the company. But if I can get an open-source Chinese or American model for free or $10 a month, that’s a big problem for them.

There are real risks to the AI boom, though. One I’m watching is the backlash against data centers in America. More people would rather have a nuclear power plant beside their house than a data center. The industry has done a terrible job explaining the benefits, and communities are opposing projects all over the country.

That’s a real risk to the boom. Data centers can bring communities tax revenue, better facilities, and potentially lower tax bills. We need to tell that story better.

Chris Wood: “The bigger danger may be regulation, not extinction.”

I take the risks of AI seriously, but I don’t find the extinction scenario convincing. And I’m skeptical of the policy proposals that tend to come attached to these warnings, like creating a big new regulatory body.

Look at nuclear power. The risks from radiation are real, but the United States effectively froze nuclear progress for decades. Nuclear has one of the best safety records of any energy source per unit of power produced. The cost of that fear was more than a generation of clean, reliable power we didn’t build.

The nuclear analogy cuts both ways, of course. A nuclear bomb is a real catastrophic threat. But when I think about AI, the most immediate risks are people using a powerful tool for terrible things: cyberattacks, fraud, and other kinds of misuse.

The more serious version of the extinction argument isn’t Terminator. It’s the possibility that we build systems smarter than us that pursue goals we didn’t intend and that we can’t correct. I don’t dismiss that as a future possibility. But it’s a hypothesis, not evidence that we’re close to that point today.

We should judge these arguments on their merits, not simply on who is making them. There are incentives on both sides. Big AI companies may welcome rules they can afford that smaller competitors can’t. But people who argue against restrictions can have incentives, too.

I think AI safety is fundamentally an engineering problem. The people best equipped to solve engineering problems are the engineers competing to build the best, most trustworthy products, not a new government agency.

Companies already face laws around product liability, negligence, and fraud. Independent evaluators can test models, and insurers can price AI risks. In an open market, a company that ships something dangerous risks losing customers, partners, and its reputation.

A government agency can’t see the future better than the engineers building it, and it moves much more slowly. The rules in the United States also don’t bind China or anybody else.

So I don’t spend much time worrying about AI ending the world. I worry about heavy-handed regulation holding back progress. That’s the much bigger risk to our portfolios by a wide margin. 

In the meantime, I’m betting on the builders. History has rewarded that bet time and again.

Warrants expert John Pangere: “Look at the picks and shovels of AI.”

I think the backlash against AI and data centers could eventually fade as people understand the technology better. That’s what often happens with new technologies: People get scared, then realize the reality isn’t as bad as they thought.

I’m looking at the picks-and-shovels side of AI. Not necessarily the companies producing the programs and software, but the companies that make it all run.

Take bitcoin (BTC) miners that are shifting toward AI data centers. Riot Platforms (RIOT) made that transition, and its stock took off as investors saw the appeal of contracts tied to AI computing rather than a business so exposed to swings in bitcoin’s price.

We have [redacted] in our portfolio, and it’s pursuing a similar strategy. It has bitcoin mining facilities in the United States, many in the South, and it’s starting to convert some of those facilities into AI data centers. 

The company has signed multibillion-dollar contracts with clients that need computing capacity. Because the facilities already exist, the conversion can happen faster than building from scratch.

I visited one of [redacted] facilities because I wanted to see the data-center controversy for myself. There are campaigns against these facilities over noise and water use. But inside, the sound was barely noticeable. The company immerses its machines in mineral oil, which it changes only every few years. The water use I saw was minimal.

I’m watching [redacted] to see whether it can follow a path similar to Riot’s as it shifts toward AI infrastructure.

The other big opportunity is energy. Data centers need enormous amounts of power, and I expect more of them to build their own generation capacity, potentially using nuclear power or natural gas.

I’m also watching uranium miners. There’s a deficit in uranium supply, and I think demand could grow as more data centers come online.

It will take time, but I’m betting on the companies that supply the computing infrastructure and energy behind the AI boom.

***

Editor’s note: If you’re not getting John Pangere’s free weekly research letter Strategic Edge, you’re missing out. More than 1,000 of your fellow RiskHedge readers look forward to receiving his insights and notes from his research trips every week. Sign up here to get Strategic Edge for free.