AI to the rescue for server and storage spending in the first quarter

Here’s a fun question and don’t cheat by asking ChatGPT. What is more valuable, an ounce of gold or an ounce of Nvidia Hopper H100 GPU accelerator?

The two are closer in price than one might imagine, but it looks like gold is still winning, and it also seems like GPU is worth more than an ounce of silver, the other base precious metal. However, the value of these and other GPUs, which are added to AI inference and training systems as well as HPC clusters, has been the main driving force driving revenue growth in the server and storage racket.

IDC just released its consolidated server and storage revenue estimates for Planet Earth and revised its 2023-through-2027 forecast for server and storage infrastructure sales by use case and industry sector. And that got us thinking about the gold and silver bars and the digital gold that companies like Nvidia and AMD and for its sake we hope Intel are also mining them in the hills of Big Data.

The fattest PCI-Express version of the H100 accelerator weighs 1,286 grams, or 45.4 ounces. It probably has a street price somewhere around $20,000, but who’s to tell with so many big-buck buyers chasing whatever limited offerings from Nvidia. That equates to $441 per ounce, which is roughly where gold traded in 2005. Gold is trading at about $1,929 per ounce as we go to press, and is therefore about 4.4 times more valuable per unit. by weight compared to a PCI H100 -Express card. Silver, by contrast, is as bright as a new server, but is trading at only about $23 an ounce this afternoon. Like GPU accelerators, gold and silver are valuable assets because they are relatively rare, but the difference is that the value of the GPU doesn’t just passively rise and fall as it becomes available and can be implemented with software to actually create new value in his right.

That’s the whole argument behind a massive expansion of AI systems that is underpinning servers and storage racks in the first quarter. Without it, volumes would be down and revenues would likely be down much more as well. We don’t know for sure because IDC isn’t ramping up sales of servers and storage acquired to run AI workloads separately from other workloads. But he said cloud infrastructure sales outpaced non-cloud infrastructure revenues in the first quarter, as they have for some years now, and that the cloud portion of server and storage sales is the strongest and largest part of the pie. data center infrastructure right now unit shipments are down a dramatic 11.4 percent, but average selling prices are up 29.7 percent thanks to price inflation (server makers feel confident pass on some costs eventually) and a higher concentration of GPU-accelerated systems being deployed by cloud service providers.

We believe that GPU-enhanced servers have sustained server revenues for just over a decade in a material way, and that this has increased as the AI ​​revolution has picked up its momentum. Clearly, in the wake of large language models and deep learning recommendation models going mainstream, demand has exploded and cloud builders and hyperscalers are buying as many as possible to try and chase the cloud opportunity. AI. It is fair to say that due to supply issues and the high cost of AI training systems, many organizations looking to scale AI into their workloads will need to start in the cloud, which creates a virtuous cycle that favors the cloud and their rental model over buying systems and installing them on premises with notable exceptions where large enterprises and startups have the money to buy their own and don’t have to pay premium cloud to get the capacity of the GPU to train your models. This strain will force many to start with pre-trained models and modify them, which is probably a good thing given the time it takes to train a model and the scarcity of hardware and people to make it.

In the first quarter, which is the most recent for which data is available from IDC, server and storage sales across all workloads and customers totaled $35.3 billion, an increase in 8.2% year over year. It’s important to remember that this is not an indicator of the revenue that cloud builders and other service providers make from renting capacity on the machines they buy.

Breakdown by delivery model, server and storage sales for shared clouds, what we call cloud builders and what many still call the public cloud even though it’s not public services at all increased by 22.5% to 15.7 billions of dollars.

IDC breaks into another market segment in dedicated clouds, which includes servers and storage sold to host applications more statically across service providers, as well as equipment sold by OEMs under lease agreements (such as Hewlett Packard Enterprise GreenLake, Dell APEX, Lenovo TrueScale and Cisco+). Iron ore sales to support dedicated clouds decreased 1.5% to $5.8 billion and still represent a relatively small portion of the market. But don’t get the wrong idea. Over the past twelve months, equipment sales in dedicated cloud use cases increased 18.7% compared to 21% for shared cloud infrastructure.

Add that up, and in Q1 2023, server and storage sales across all cloud use cases increased 14.9% to $21.5 billion. And sales of non-cloud legacy platforms think IBM System ze Unix mainframes and proprietary systems running database and transaction processing systems fell nine-tenths of a point to $13.8 billion. Over the past twelve months, however, sales of these non-cloud systems have increased by 10.1%.

The question now is this: Will rising interest rates by global central banks and general uncertainty in national and regional economies cause server and storage sales to decline in 2023, or fuel underlying growth in existing applications and add emerging AI bubble to do well enough in 2023 and beyond?

If non-cloud equipment sales are a leading indicator, then it’s not looking good. IDC expects server and storage revenues from this use case class to decline 6.3 percent to $60.4 billion in 2023. We don’t think this is a great leading indicator, but rather a reflection of the upgrade cycles for IBM System ze Power Systems Iron as well as other X86-based systems to do similar work.

While the overall server and storage market is forecast to grow just 1.1% this year, to $156.8 billion, cloud infrastructure sales are expected to grow 7.3% to $96.4 billion. dollars, and within this shared cloud infrastructure, things sold to Amazon Web Services, Microsoft Azure, Google Cloud, Meta Platforms, Alibaba, Tencent and Baidu are expected to grow 8.4% to $68 billion. Dedicated cloud infrastructure will grow a more modest 4.8% to $28.4 billion.

Sales of non-cloud equipment to businesses, government, and universities appear to be the culprit, if you run a few Venn diagrams of the data. Here is a consolidated table we created based on updated IDC data for 2022 plus forecasts in 2023 and 2027:

IDC expects server and storage spending in enterprises, governments and academic institutions to drop 5.1% to $62.3 billion this year, which is not good news and also suggests that when it comes to AI, many of these organizations are looking to the cloud to get started. This makes sense given the double uncertainties surrounding the economy and how to weave AI into applications.

The overall forecast for server and storage spending worldwide between 2022 and 2027 remains quite rosy, according to IDC, with sales expected to have a 6.2% compound annual growth rate and reach 221.7 billions of dollars by the end of the forecast period. Sales of cloud infrastructure in 2027 will be, as you can see from the table above, a little smaller than all infrastructure in 2022, which is remarkable in itself. But again, much of that growth is coming from investments in AI training and inference hardware, both for on-premises use and for use through a cloud.