PNC's solution to rising AI costs: build its own AI

PNC kitbashes its own AI
The cost of running AI programs is coming into clearer focus. And it ain't cheap.

There have been a number of stories lately talking about the rising costs to businesses of using AI. The reason these contracts are suddenly becoming more expensive is because the tech companies that are selling them have changed their business model, charging now not by the seat but by usage (I touched on this on Monday). What companies are discovering is that they are burning through their annual budgets in months.

PNC has looked at this landscape and said, well, let's just build our own, as Nathan Place wrote yesterday. CEO Bill Demchak said the bank is building its own "AI factory," and it doesn't appear that he's just being figurative. They are, apparently, really planning to build a factory. They want to acquire their own data centers, buy their own chips, and build their own AI models designed for their specific uses and needs. The goal, Demchak said, is specifically about costs. If AI providers raise their prices – spoiler alert, they are – PNC doesn't want to be beholden to them.

This is a PNC story right now, but it's going to become an industry story. In fact, it really already is an industry story. Every bank is going to have to figure out how much it is spending on AI versus the value of that spending. In other words, they're going to have to justify the costs. Is it actually cheaper to build your own AI? It may be. There are plenty of kitbashed AI programs out there; we wrote about OpenClaw recently, but there are others. It's not impossible to do. The Chinese churned out DeepSeek and others that perform nearly as well as leading models and cost a fraction to build.

The biggest issue with AI may not be about security or firing employees or who owns the data. It may simply be cost.

Who's responsible for online fraud?
In the world of public figures, I am a very, very, very minor one. I am not getting invited to celebrity row for a Knicks game. I am not sitting two rows behind celebrity row. I'm not going to any Finals games at the Garden, in fact (I can't afford a ticket, of course.)  I am a public figure only in the sense that I write stories that carry my byline. That, though, is enough for a crook looking for an angle.

Over the years, given my background writing about markets and crypto, I have had my name and identity used by fraudsters online trying to scam people out of money. They will take my pictures and quotes from me and package it into something new designed to lure people into sending their money to Cambodia, or North Korea, or God knows where. One time a cousin of mine actually got a text message from somebody pretending to be me and trying to get her to invest with them. They gave up after she told them she was my cousin.

In my personal experience, the worst site for these kinds of scams is Instagram. The site seems to have absolutely no standards for verifying identity, and getting a fraudulent account shut down is difficult. And the people committing the frauds don't care if you do get an account shut down. They'll just open a new one and find new marks. And I've always thought the problem was so bad that the parent company, Meta, has been very negligent about that and somehow bears some responsibility. Apparently, a judge agrees with me.

A District Court judge allowed a case to proceed that claims Meta was, by the way it operates, aiding and abetting fraud, our Carter Pape reported. The plaintiffs claimed that fraudsters impersonating Bank of America's head of U.S. equity and quantitative strategy Savita Subramanian robbed them of $300 million. And, they blamed Meta, arguing that Meta's ad tools were actively involved in the fraud. The argument was that Meta's online tools made it a co-creator rather than neutral host. 

If you follow the markets at all, you know Subramanian. She's pretty much a fixture on cable news and is often quoted in the press. So her reputation was a natural attraction for grifters. And it worked, the grifters got people to part with $300 million. Would it have worked without Meta's ad tools? That's what the case is going to determine. This could be an important case. Social-media companies have long argued that they are not responsible at all for the content on their sites, that they are just passive hosts providing a service, and there is a law, Section 230 of the Communications Act, that backs them up. But where exactly to draw the line between passive and active support has long been debated. This case could shift that debate.