When most investors think of artificial intelligence stocks, Nvidia is a top-of-mind name. And rightfully so. Its hardware is at the heart of most AI data centers. Or maybe it's Microsoft or Alphabet's, both of which offer popular AI-powered virtual assistants free of charge.
IBM (IBM +0.98%), on the other hand, doesn't come up much during discussions of artificial intelligence's likely future. However, maybe it should. Smart investors would at least put the company on their AI radars anyway. Here's why.
Meet the modern-day IBM
To be clear, IBM doesn't hold a candle to Nvidia's share of the AI data center market. As noted by the Motley Fool's own in-house research team, Nvidia generated more than $35 billion worth of AI data center revenue for the final quarter of last year, while Intel, Advanced Micro Devices, and IBM each only reported on the order of $4 billion. Nvidia is also the only one of these hardware makers to see any meaningful net growth of their AI data center business over the course of the past several years.
Still, IBM is a budding artificial intelligence name worth watching. But first things first. IBM's single-biggest profit center these days isn't hardware. It's software, which accounts for more than 40% of its revenue and nearly two-thirds of the company's total gross profits. These numbers, however, come with an important footnote.
See, it's hardware sales that ultimately generate software and consulting revenue. As the company has regularly pointed out for some time now, for every $1 spent on its cloud hardware -- which makes up most of what the company categorizes as "infrastructure" -- an additional $3 to $5 is spent on software while another $6 to $8 is shelled out on services.
IBM even picks up an extra dollar or two on additional purchases of enterprise infrastructure. IBM's historical numbers bear this idea out, too.
Data source: IBM. Chart by author. Figures are in millions. Revenue and gross profit data are both cumulatively but separately "stacked."
Connect the dots. Artificial intelligence data centers will never be IBM's biggest business. However, if that business grows, then the rest of IBM will grow even more -- and there's plenty of reason to believe such growth is in the cards.
A tailwind is brewing
There's no denying that it hasn't happened yet. The company's enterprise infrastructure business is arguably stagnant, in fact, mostly weighed down by weak sales of its Z series of mainframes and its state-of-the-art Z16 model, in particular.
Although far from being the kind of mainframe computer system most people think of when the label became fairly common back in the 1970s through the 1980s, it's still a legacy concept that's distinctly different from most other AI data centers that largely utilize Nvidia's hardware or power apps like Microsoft's AI tool, which is called Copilot.
As time marches on, though, the best solutions always move into the mainstream. As it turns out, IBM's Z16, as well as its upcoming Z17 platform, are outstanding at a type of machine learning called inference. That won't mean much to many people, but it does matter. To date, most of the commonly used AI tools like Google's Gemini and OpenAI's ChatGPT utilize what's called a "training" approach.
That just means these platforms parse a massive amount of selected text-based information on a topic and then come up with an appropriate response to a query based on the sum total of all this aggregated information. It works fine for most purposes.
NYSE: IBM
Key Data Points
Inference, on the other hand, is a different approach to machine learning. With inference models, a platform considers all known information but is tasked with taking a new action or making a reasoned but unproven prediction based on this known data. As IBM explains it, "inference is an AI model's moment of truth," ultimately ending in "a test of how well it can apply information learned during training to make a prediction or solve a task."
It's a small, nuanced difference, but it's also a pretty big deal to the artificial intelligence industry. Now that AI platforms effectively know what they don't know but also know they might be asked to craft an appropriate response or solution, artificial intelligence itself is quietly entering a new era. That's why industry research outfit Market.us believes the worldwide AI inference server market is set to grow at an annualized pace of more than 18% through 2034, matching up with a similar outlook from Lucintel.
This prospective growth, of course, bodes very well for IBM, which specializes in the very kings of mainframe servers that are on the cusp of significant demand growth.
A bet worth making
IBM isn't the only name that is making artificial intelligence platforms capable of handling heavy-duty inference tasks, for the record. Most any company in the AI hardware business is able to optimize their tech for this relatively newer approach to machine learning.
IBM is arguably one of the better inference plays, though, if not the best. The Telum II processors and their on-chip Spyre accelerators, capable of handling 24 TOPS (tera operations per second), will be found inside Z17 servers. They are the end result of a long string of successful inference-friendly technological developments that aren't quite being matched by other names in the AI hardware business. Data center operators just need to recognize their value. Then, investors need to follow suit.
It's still arguably a bet worth making sooner rather than later, though, before anyone else realizes there's an impending shift ahead for how many of the artificial intelligence industry's platforms process data and respond to users' requests. Just a little more AI server revenue could generate considerably more high-margin software revenue for IBM. Most of this new revenue would also be recurring.
