This AI game hides in plain sight. It also happens to pay a large dividend.

There are reasons why the market is skeptical of Big Blues’ position in AI, but the omission is short-sighted. Shares of chipmaker Nvidia (ticker: NVDA) are up 172% this year, boosted by the success of its chips used in generative AI applications. Microsoft (MSFT) and Alphabet (GOOGL), the major players in AI software, each added 40%.

IBM (IBM)? It’s down 8%.

That astonishing decline comes just as IBM has made a makeover over the past three years under the steady hand of CEO Arvind Krishna. IBM probably has a deeper knowledge of AI than almost any company, yet shares trade at less than twice forward sales, compared to Nvidia at nearly 20 times. IBM also has a dividend yield of over 5%.

The company has been working on AI applications for at least four decades. In 1997, two years before Nvidia was released, an IBM supercomputer called Deep Blue defeated world chess champion Garry Kasparov in a six-game match. In 2011, IBM’s Watson supercomputer famously defeated human champions Danger. Among Watson’s opponents was the legendary Ken Jennings, now host of the show. I, for one, welcome our new computer overlords, wrote Jennings under one of his Finals Danger answers.

To be sure, IBM made missteps. After the Jeopardy stunt, he made a big push to use Watson in healthcare, for drug discovery and other applications. It never quite worked out, and IBM sold the Watson Health business for a reported $1 billion in 2022. Some people seem to have interpreted the sale of Watson Health as IBM giving up on artificial intelligence, but that’s far from it. how true.

Last week, I had a long conversation with Krishna about the company’s approach to the AI ​​business. He had a lot to say.

First, IBM recently launched an entirely new version of Watson called Watson X. The new offering has three parts: Watson.ai works with customers to create new models or datasets. Watson.data acts as a data store, putting the company in competition with Snowflake

(SNOW), among others. And Watson.governance monitors AI models to make sure they’re accurate and accountable, not filled with false or offensive information.

What IBM won’t do is create the next ChatGPT. Public models are incredibly powerful, Krishna said. What Google, Facebook, Microsoft are doing fits perfectly into this mode. They’re building very, very large models that serve everyone.

But Krishna thinks public-facing AI applications are only a small part of the opportunity. It’s like an iceberg, he says, with chatbots like Microsoft Bing and Google Bard above the waterline. There are multiple use cases that won’t benefit from a large public model.

IBM’s strategy is to help clients build their own AI applications, to get more value from their data. In some cases, the company combines open source models with proprietary data. For some customers, IBM is building private models specifically for their data.

While IBM doesn’t plan to build a large general model like ChatGPT, Krishna says it is building a family of domain-specific datasets. For example, he’s building a chemical model, based on publicly available information.

If I take one of my chemical industry partners, say Dow, Mitsui Chemicals or BASF, they have proprietary data on how they make chemicals, he says. Would any of them put their proprietary data into a public model? Obviously not. But they would like to extend the chemical model to include their own data… they could provide faster answers to customer and internal questions and come up with new formulations. It will be an accelerator of their business model.

In one case, Krishna says, IBM is working with a bank to get better compliance and audit data for its staff, in part to show regulators it has the right controls in place.

IBM has built 20 of these domain models beyond chemistry and banking, the list includes one for writing code, another for making IT operations more efficient. There’s a geospatial model that IBM is adapting to combine with climate data from NASA, to improve weather modeling. The company that owns open source software giant Red Hatis isn’t shy about using open source training models. Notably, IBM has partnered with Hugging Face, an AI start-up with a library of 130,000 models.

The complicating factor for investors is that IBM won’t say how much of its revenue is related to artificial intelligence. Krishna says the number is impossible to calculate. Our mainframes have AI circuits, so is it the mainframe’s corporate AI? Will I back up storage with AIis storage backup AI in the future? Maintenance applications use artificial intelligence. All cybersecurity will use artificial intelligence. Before the next five years are up, everything will have AI fused into it.

Krishna cites an estimate from PwC that shows AI could generate $16 trillion in value for the economy by 2030. And Krishna has a long-held view that IBM should grow revenue in the mid-single digits over time. .

What the CEO won’t say directly is that AI opens up new opportunities for IBM and could be additive to sales. It’s a potential accelerator, he says.

Investors should read between the lines.

Write to Eric J. Savitz at eric.savitz@barrons.com