I don’t doubt that AI will prove as revolutionary as steam and electric power, but I am wondering about the return on all that investment. Behind every stock analyst’s price target, and every CFO’s signoff on a billion-dollar data center, there’s an expected return-on-investment.

What do we know about AI’s ROI?

There isn’t a lot of detailed information out there. On Wednesday’s earnings call, Huang said that corporations—and even countries—are refitting a trillion dollars worth of data centers with accelerated computing stuff from Nvidia.

The company’s investor presentation says those data centers can earn $5 over four years for every dollar they invest in Nvidia computing and networking products. That suggests a 50% annual return.

The Nvidia presentation calls that ROI calculation an “illustrative example.” Some difficulties in getting real calculations of ROI include the secrecy of competing firms, and the sheer newness of the AI learning models they’re deploying. Marketed products based on OpenAI’s ChatGPT or Google’s Gemini are few. Microsoft charges $10 to $30 a month for its Copilot virtual assistant. Netflix

NFLX

1.74%

, Amazon, and Alphabet

GOOGL

0.83%

’s Google use AI internally to determine what they present to their searching audiences. But that’s a cost of doing business, not individually priced products.

An obvious place to look for information on AI’s ROI is, well, AI. And so, I asked ChatGPT to quantify the return on investing in AI.

“Quantifying the return-on-investment (ROI) for AI can be complex,” said GPT, “and depends on various factors such as the specific application, the industry, the maturity of AI implementation, and the goals of the organization.”

“However, in many cases, AI investments can lead to substantial returns,” the bot said, optimistically.

The answers from the chatbots at Anthropic and Perplexity were similar to OpenAI’s.

Google has just added an “overview” summary to the listed results of your search. The overview is generated by its Gemini chatbot.

Gemini’s answer was more nerdy than the other chatbots. “For AI initiatives,” said Gemini, “ROI is calculated by subtracting the cost of investment from the net return from investment, then dividing that number by the cost of investment and multiplying by 100.”

Costs listed by the chatbots include the infrastructure of the data center, computers, and networks, plus charges for using an AI model that can have hundreds of billions of variables. Data scientists cost money. And so do the data used to train the model on a particular task. Barron’s owner News Corp

NWSA

3.32%

—for instance—announced a deal this week that lets OpenAI use our news products.

The chatbots also listed benefits (i.e., returns) from AI investments: Cost savings, revenue growth, customer satisfaction, and risk mitigation. Cost savings always come first.

That may help explain why real world case studies that enumerate the returns from AI are hard to find. “AI can automate tasks that are repetitive or time-consuming, leading to cost savings in terms of labor,” says ChatGPT. “For example, AI-powered chatbots can handle customer inquiries, reducing the need for human customer support agents.”

In other words, AI can displace workers or affect their wages. Each of the chatbots makes this point. Organizations deploying AI are less likely to do so.

I’m no Luddite. Two hundred years after the industrial revolution began replacing physical labor, America has low unemployment. AI tools may make knowledge workers more productive, without cutting ranks. But if AI replaces me, I won’t weep. New jobs will be created on top of AI. Still, the politics of job loss aren’t great, which makes companies less eager to share the ROI calculus around their AI use.

Back in the Nvidia corporate presentation, it lists some jobs ripe for the “huge ROI from AI driving a powerful new investment cycle”: one billion knowledge workers; 50 million content creators; one million legal professionals; nine million educators; 15 million call-center agents; one billion agricultural and food workers. Let’s hope the impact is on the benefit side, and not the cost side.

Apart from the newness of AI deployments, their impact is all the harder to predict because new capabilities of these models emerge as they get bigger. We discover they can do things we didn’t design them to do.

That could lead to new drug discoveries. Or, in the hands of the wrong government, new kinds of harm. The short-term dislocation of workers may also fuel extremist politics.

At least we know what kind of return Wall Street expects from Nvidia stock. The average 12-month target price on FactSet is $1,169. That’s a 12% return from a recent $1,045 price. Not to mention the 800% return it’s generated since 2022.

Write to Bill Alpert at william.alpert@barrons.com