The invasion of Ukraine. Inflation. Covid. The price of gasoline and oil. The supply chain. Climate change. Those are some of the news topics driving today’s headlines domestically and abroad. While not as staggering, technology has had garnered a few headlines of late, particularly pertaining to GPUs. Why the attention? Because GPUs are an important element in making the modern world function.
Look around. GPUs are everywhere. As Dr. Seuss might say, “In your car, in your phone, even for a high–tech drone. Used for gaming and for mining, but certainly not fine dining.”
For more than four decades, analyst Jon Peddie, president of Jon Peddie Research (JPR) and publisher of JPR’s “Tech Watch” news portal, newsletter, and reports, has been closely monitoring and regularly reporting on the graphics market. In fact, he wrote the book on GPUs, literally, as The History of the GPU will be available later this year. Recently, he was interviewed by a journalist from JW Insights, a Shanghai–based technology news agency, about his assessments concerning GPUs, the global GPU supply landscape, and the current state of the GPU industry.
As Peddie points out, the GPU as a processor type has found its way into almost every imaginable electronics device, from cars to rockets, from IoT to watches, from TVs to smartphones, and of course, game consoles and PCs. The demand for GPUs in the data center and for PC gaming has been steadily increasing for over a decade, he says, and shows no signs of slowing.
GPU prices have been fluctuating significantly since the second half of last year. We have even seen GPUs listed on third–party sites for well over MSRP. Is inflation to blame for the high cost of GPUs? Inflation has not had that big of an effect, says Peddie. He contends that inflation has added between 5% and 7% at most to the overall cost — and most of that has been due to fuel costs. The import taxes that have been added a little over a year ago, however, account for a 25% boost, and that accounts for the biggest increase.
And who pays that import tax? Not the U.S. or Chinese government, or any other government for that matter. “The consumers themselves are paying that. If they are complaining about the high prices, what they are complaining about is this burdensome tax that has been put on us,” Peddie said.
Before continuing, Peddie wanted to clarify the following point as it pertains to pricing: Consumers don’t buy GPUs; they buy add–in boards (AIBs) for a desktop PC, or they buy a laptop that has a GPU already installed. That’s at the tail end of things. The GPU starts out as an independent semiconductor that gets placed in a system along with other components, including software, so when the system arrives in front of the consumer, several steps have already occurred. The GPU, or semiconductor, starts out in the fab and then goes to board fabricators, which insert the GPU on the motherboard (for notebooks) or into AIBs (for desktops, workstations, and servers).
The board fabricators then ship product to two types of customers: either directly to OEMs such as Dell, Lenovo, and HP, or to AIB partners such as Asus, EVGA, MSI, and others, which then sell to consumers either directly from their web page or they ship product to Amazon, Alibaba, New Egg, and so forth. They also ship them to retail stores such as Best Buy in the U.S. and places like Xingguang Center in Beijing.
“The online distributors and retail shops are the ones that have raised the prices. It hasn’t been the add–in board partners. It hasn’t been the OEMs. And, it hasn’t been the GPU makers/suppliers. It’s been done at the far end where the consumer is involved,” Peddie said. “That has happened because of supply and demand. If there’s a lot of demand, the people with product for sale will raise their prices, and they’ll keep raising them until the demand falls off. That’s called elasticity, and that has just started to happen.”
One of the things that triggered that supply/demand imbalance was the use of software bots by scalpers and speculators unleashed to constantly roam the internet in search of inventory from online distributors and online retailers. When the bots located inventory, they snatched it up. Suddenly, inventory levels plummeted to zero. So, any consumer looking to buy a board online, well, they were out of luck. With no inventory in the channel for the consumer, the speculators then began offering the same board they had recently purchased (at the regular price), but for two or three times the manufacturer’s recommended price.
The companies that sold the boards to the distributors did not make a bigger profit on the sales; they sold them to the distributors at their regular price. Encouraged by what they saw in terms of supply and demand, the distributors and retailers then raised their prices, too. “That’s why the prices have gone up!” Peddie explains.
Things did not go exactly as planned, however, and not as many consumers purchased the AIBs as the speculators had expected. The scalpers and speculators believed that the crypto miners would pay just about anything to get hold of these boards for their mining activities. But some major governments, including China and Kazakhstan, banned mining. So, when the miners were not buying those boards any longer, the speculators were left holding the bag — err, boards — and were trying to find customers for them. So, the AIBs were offered on eBay for big bucks initially, but slowly the pricing continued to drop. Seeing this, the distributors and channel vendors dropped their pricing, too.
“So now we are in a slowly downward drifting price situation,” Peddie said. “And everyone is looking at the future. It’s expected that the prices will get back to somewhere normal by the end of the third quarter. Maybe. That could change if there is another interruption in the supply chain and demand started to increase significantly again. Bottom line: With no [abnormal] demand, prices will come down.”
Despite the yo–yo effects caused by the speculators, JPR is forecasting a very consistent end–user AIB demand growth of 3.0–3.1% through 2026. As Peddie explains, demand is still high across the board, so to speak, in the areas of gaming, GPU compute (used for AI and so forth), and workstations and notebooks (which are currently very popular).
“It’ll go up and down a little bit, but we think the average growth over those years will be around 3%,” Peddie said.
That’s an important number, he points out, because it indicates that demand is higher than the expansion of population. Typically, markets will grow, but they will either grow because of some phenomenal development or they’ll grow just at a steady rate as the population grows. So, AIBs are growing faster than the population, which means that new customers are coming into the market.
“That’s a very important sign,” Peddie noted.
The fab picture
Recently, Nvidia CEO Jensen Huang stated that Nvidia is considering using Intel’s new foundry business, IFS, to build chips, which would diversify its suppliers (Nvidia currently uses Taiwan Semiconductor Manufacturing Co. and Samsung Electronics Co. to build its products). Would IFS be a good partner given that Intel is now a direct competitor with Nvidia on both the CPU and now GPU fronts with the launch of Intel Arc Alchemist GPUs little over a month ago? Yes, argues Peddie.
“I have known Jensen Huang and [Intel CEO] Pat Gelsinger for over 20 years, and I am constantly surprised and delighted by their courage, vision, and rule–breaking. That’s important — they are willing to break the rules. Pat will not pass up an opportunity to keep the ovens in his fabs warm and running. He’s not religious about who the customer is — if he can get customers from the outside to keep the fabs running, he’ll do that. He has no problem with selling parts to Nvidia, AMD, or anyone else. And Jensen, who is a master at semiconductors, will source chips wherever the fab can meet his quality and production demands,” Peddie said. “Expect the unexpected from these titans. They will continue to surprise you.”
In fact, about three years ago, Intel was buying AMD GPUs and putting them into their CPUs (that product was discontinued several months ago). So, the idea that Intel would buy AMD parts and use them and put them in their product illustrates and demonstrates the flexibility these companies have toward each other. Yes, they are competitors, Peddie explains , but they are also very good friends; they respect each other.
Despite the growing trend of cloud computing, Peddie points out that the volume for that use and in the data center is much smaller than it is for the consumer — including those who are not gamers. In fact, for every data center graphics board that’s sold, a thousand consumer boards are sold. But keep in mind, he points out, that the data center board can do 3 to 10 times more types of work than a consumer board can, which is why that volume isn’t as high. Nevertheless, the volume is dramatically different.
“The data center side is definitely increasing. It’s got a really nice growth rate to it, and everybody’s happy about that,” Peddie said. “And that’s due to the flood of data being processed and the development of AI. But in terms of unit shipments, it’s nowhere close to what consumer electronics are.”
As Peddie emphasizes, GPUs do more than just render graphics and video. GPUs are everywhere, he iterates. Aside from being in PCs and servers, AMD, Intel, and Nvidia are putting them in cars, buses, airplanes, boats and ships, smartphones, supercomputers, game consoles, and hyperscalers, for example. And even though all those platforms can — and do — exploit the power of the GPU, they all have a slightly different demand on the GPU. Some need very low power demand, such as smartphones, while others, such as supercomputers, need unlimited performance. One size does not, and cannot, fit all, according to Peddie.
Therefore, GPU builders need designs that are scalable in terms of power demand, performance, size, and price. They have to be competitive, after all. So, there are all kinds of trade–offs that have to be considered.
“These are super–processing companies building state–of–the–art processors. These three — AMD, Intel, and Nvidia — are just so far ahead of everyone else; don’t underestimate any of them,” Peddie said. He cautions that a person cannot look at them from the point of view where one wins and the other loses, and it’s game over.
“It just doesn’t work like that. One might get a little ahead on technology, or production, or price. But that and market share in one segment or another will shift from quarter to quarter; if you look back 20 years or more, they never stopped growing, never stopped introducing leading–edge technology, and never stopped delighting their customers.”
Where does this leave other vendors, particularly those in China that are receiving government funding to build GPUs? After all, they are entering a very competitive market dominated by the three giants, each with thousands and thousands of employees. And here, size does matter. It takes time, experience, and bulk to compete. Peddie poses this question: How can a start–up with maybe 200 people compete against AMD, which has 16,000 employees; Intel, which has 121,000; and Nvidia, which has 22,000?
“Nvidia has more people writing drivers than the start–ups have in total!” he said.
There’s no comparison, Peddie argues… for now. Yet, the Chinese start–ups do have something that AMD, Nvidia, and Intel never had: a home–court advantage — a large home economy that wants to buy made–in–China parts.
“That will fortify them and help them grow, and they will get bigger and get more and more technology. And one day they will have the mass and technological experience to move outside the borders of China and take on the big guys. But there’s no rush. There’s plenty of market for everyone,” Peddie said. “And competition favors the consumer, so cheer them all on and we will win.”