A Second Opinion On Future GenAI Spending

Two weeks ago, before we began our nightmare travels to get to the 2025 edition of Nvidia’s GPU Technology Conference in San Jose, we put together an analysis of the AI server and storage spending forecasts put out by the good folks at IDC. Now, we have a similar, and much tighter, forecast coming out of the other good folks at Gartner, who like IDC give those of us who don’t have big bucks for market research a little taste of what it must be like for those who do.

This AI market is the biggest supercomputing market the world has ever seen, and that in itself is a remarkable thing. Which just goes to show you that you can never be sure how something might go commercial. The Internet started as a way to create a communications network that could still function despite nuclear strikes in major cities across a fairly large nation, and was used as a communications backbone for scientists and researchers before it became the conduit of comment and commerce. Those of us who grew up in the HPC simulation and modeling world have been hoping for our Jarvis companion to do great things, but it turns out we are going to get Max Headroom and C-3PO and R2-D2 instead. Perhaps that’s all most of us can handle anyway.

If Nvidia’s financial results are any indication – and the most certainly are – then the world, or at least a few dozen very eager customers hoping to get rich on AI, is spending a mind-boggling amount of cash on systems to run AI models and run inference against them.

The IDC forecast we wrote about two weeks ago looked at server and storage spending for all AI workloads, including classical machine learning as well as the GenAI variant underpinned by large language models and now “chain of thought” reasoning models that are comprised of many smaller, specialized LLMs that work together to generate text, images, and video when prompted and that are like the Q&A systems we read about in science fiction for many decades. IDC is only casing spending for AI infrastructure every six months right now because it is hard to tease the data out of those who are making and selling this gear. That forecast shows just shy of $150 billion in AI servers will be purchased by organizations in 2025, with a thin sliver for storage.

In its forecast, the analysts at Gartner are only talking about the GenAI portion of the AI market, and they go further and break the spending on GenAI down into services, software, devices, and servers. In its overall IT spending tracking and forecasts, Gartner breaks calls servers “datacenter systems” and presumably this includes the storage and networking for the datacenter segment of the IT business.

Just for fun, we took the data from 2024 and 2025 that Gartner put out in its public statement and backcast it into 2023 using the growth rates it showed for 2024. So we get three years of that that way. And then we grabbed the most recent Gartner IT spending forecast, which we discussed back in January, and stacked that up against the GenAI spending to see what share GenAI was of overall spending. And for more fun, we plotted “calendarized” revenues in the Nvidia datacenter business as well, including our forecast of Nvidia spending, which we created for the IDC story a few months ago.

Nvidia finishes its calendar year at the end of January, so you have to lop off January revenues on the front end and tack on the previous January’s revenues on the back end to get something you can compare to an annual year. This is not a perfect method, of course, but it is a close enough approximation to say something useful.

The first thing that is annoying with the Gartner data – and even John-David Lovelock, the distinguished vice president analyst at Gartner who put together the forecast admitted – is that AI PCs are skewing the data. (Or maybe the are SKUing it?) In any event, as Lovelock said, people are buying AI PCs, which have local AI accelerators in them, but they don’t seem to be paying any premium for that. So as old PC inventories get flushed out, eventually “GenAI devices” will just be “all devices” and there will not really be any distinction. But as you can see in the table above, there are lots of them that sold in 2024 and nearly twice as much money will be spent on them in 2025 and they will comprise about half of all device revenue this year.

The same holds true for servers, but in this case, people are actually intentionally spending money on GPU-enhanced machinery to do GenAI work, and the share of the server revenue pie that comes from these GenAI machines will cross over above 50 percent sometime in early 2026 if current trends persist.

In 2023, GenAI servers represented 22.6 percent of all server spending, and in 2024 that rose to 41.2 percent. By Gartner’s numbers, that will rise to GenAI driving 44.5 percent of datacenter systems sales in 2025, which is $180.6 billion. This is a huge amount of money, especially for those of us who waited a long time to see datacenter systems spending (meaning servers, switches, and storage) cross over the $100 billion mark, which it did sometime in the mid-to-late 2000s, depending who you ask and how you do the math.

If you take devices (PCs, smartphones, and tablets) out of the picture, you get a better sense of what is going on with GenAI in the datacenter. In 2023, which was the first full year of the GenAI boom, datacenter spending on GenAI was $62.5 billion, but total datacenter spending (not including communication and data services, which we ignored) were $2.71 trillion according to Gartner. So that core datacenter GenAI business represented only 2.3 percent of total datacenter spending.

In 2024, the overall IT market grew by 9.9 percent to $3.74 trillion, and the part that was not devices and was focused on the datacenter grew by 10.8 percent to just a tad above $3 trillion. GenAI outside of devices was $165.4 billion, up by a factor of 2.65X and comprised 5.5 percent of overall datacenter spending.

In 2025, Gartner is estimating that overall IT spending will grow by 12.1 percent to $4.19 trillion, datacenter spending within that will grow by 12.5 percent to $3.38 trillion, and GenAI in the datacenter will grow by 48.5 percent to $245.5 billion.

Given this, our projection for Nvidia to have $174.4 billion in sales for its Datacenter division on a calendarized basis for 2025 might be a little high. But then again, maybe Gartner’s datacenter GenAI projection is too low. We shall just have to see when all of the beans are counted.