If we somehow raise enough money for AI, it will probably kill us

Join Lucky Lopez, Jon Frederick, and Jack Ballinghoff for an all-access look at what’s happening across automotive, breaking down the biggest headlines, industry shifts, and stories dealers need to have on their radar.

🔴 We’re LIVE today at 1 PM ET. Tune in to Automotive Views.

— Freddy

Last week, some of the people closest to building frontier AI started asking everyone to slow down.

Jacob Coxon, a former OpenAI and Anthropic researcher, publicly resigned and accused both companies of “racing straight to self-improving superintelligence and gambling with our lives.”

Anthropic’s Evan Hubinger put the chance of AI killing all humans at more than 10% within the next decade.

Then, a few days later, Anthropic CEO Dario Amodei argued that frontier labs should deliberately pace capability gains and let independent evaluators look deeper inside their systems.

But I don’t think they’re actually scared of the tech. They’re scared of the financials.

See, none of these companies can afford to back off while competitors keep racing, but staying in the race requires trillions of dollars in infrastructure. Meanwhile, the returns are still a big question mark.

That makes “maybe we should all slow down together” a much more interesting proposition.

If Anthropic decides tomorrow that superintelligence is getting a little too spicy and slows frontier development, OpenAI does not have to follow. Neither do Google, xAI, or anyone else.

Anthropic could lose researchers, customers, capital, and whatever advantage comes from reaching the next major capability first. That is why coordinated pacing matters. Slowing down is a much easier decision when your competitors have to do it too.

Safety rules also happen to favor the companies already at the frontier. Audits, model evaluations, security requirements, and compliance teams cost money. Anthropic, OpenAI, Microsoft, and Google can pay those bills much more easily than a company trying to catch up.

My POV: “This technology might become catastrophically powerful” and “this technology is going to replace half your back office” are essentially the same claim about capability, pointed at two different audiences.

But if these people actually believe what they’re saying, why are they still racing? The answer is simple: Just follow the money.

Morgan Stanley estimates that roughly $2.9 trillion will be spent on data centers globally through 2028. About $1.4 trillion of that can be covered by Big Tech’s own cash flow. But the remaining $1.5 trillion has to come from outside capital: bonds, securitized debt, private credit, and other investors willing to finance the buildout.

At some point, trillions stop meaning anything to me. But the important part is that everyone is making those commitments before we really know what the technology and infrastructure will earn.

That sent me down another rabbit hole.

Take Morgan Stanley’s roughly $2.9 trillion buildout and, purely for a thought experiment, round it to $3 trillion in investment. If that capital had to earn back its cost over an average of seven years, it would need to generate roughly $430 billion a year before electricity, cooling, maintenance, interest, etc.

The point is that trillions of dollars are being put into assets that now have to generate a very large, recurring profit pool.

The good news is that demand is not exactly struggling. Anthropic reportedly went from a $9 billion annualized revenue run rate at the end of 2025 to roughly $65 billion by July. OpenAI reached around $40 billion annualized in August.

At the same time, the Street may be overestimating pricing power. During a recent call with hedge fund manager, Michael Taylor, he explained that as more players catch up on programming capabilities, they will likely have to compete on price. Not to mention, AI is getting cheaper as it gets better. New chips, smaller models, caching, and routing keep reducing the amount of compute required to perform the same work.

Bottom line: Slowing down is starting to look financially useful.

As AI leaders started calling for a slower pace, Trump rejected the idea that the U.S. should pull back while China is still competing for the lead. His argument is basically that AI is too strategically important to risk coming in second.

And honestly, that is the same problem the labs have, just with much higher stakes.

Anthropic can look at another giant training run and decide it would rather wait to see whether the last few billion dollars actually paid off. But if OpenAI keeps going, sitting out starts looking dangerous. At the national level, Washington can decide that the AI race is getting too expensive or risky, but not if it believes China will use the pause to close the gap.

China, meanwhile, does not exactly view American calls for restraint as a group meditation on responsible technology.

China’s state-backed Global Times called the plan a “Cold War” tactic designed to constrain Chinese development. Meanwhile, China’s foreign ministry called for cooperation and warned against treating AI as a zero-sum contest.

At the end of the day, this is what keeps the money moving.

Everyone can look at the cost of the race and conclude that the pace is getting ridiculous. But as long as falling behind looks worse than overspending, the rational response is to keep funding the next model, the next cluster, and the next data center.

But they also happen to be running companies trapped in one of the most expensive arms races in business history.

For me, that changes what I want to own. If today’s magical AI capability becomes next year’s cheap API call, I would rather own the customer relationship, distribution, proprietary data, or industry expertise than build a business whose advantage is access to somebody else’s model.

Our friend and colleague Adam Gaedke was in a cycling accident near his home in Kohler, Wisconsin, and suffered a critical spinal injury. He underwent major surgery and is now at one of the best facilities in the country for intense physical therapy and recovery.

Adam means a lot to me. He came to Freddy Media after a long career as the COO of the Van Horn Automotive Group. Adam built our dealer services team and will continue to lead it for years to come. He has become one of my closest friends.

Here are some products we find useful and would encourage you to implement in your company.