Fourth-quarter earnings are public, and four key players confirm that the data-center segment—especially AI processing—is hot, while the PC segment lacks luster and embedded sales falter.
Data Center
Nvidia’s AI-fueled data-center business (“Compute and Networking”) overtook Intel’s in 1Q23, as Figure 1 shows. By the end of the year, it was more than four times bigger. Nvidia sells cards and systems, enabling it to capture more value than if it sold only chips. Moreover, an Nvidia-based AI cluster will also include the company’s InfiniBand and Ethernet products, further boosting revenue. InfiniBand sales have grown in proportion to the rest of the business, and the company reports networking has hit a $13 million run rate (i.e., $3.25 million per quarter—as much as all Compute and Networking a few years ago).
Xeon processors dominate Intel’s data-center business (DCAI). Xeon sales have been squeezed by AMD’s Epyc on the one side and Nvidia GPUs on the other, with the latter capturing customer dollars that would otherwise go to CPUs. Intel claims more than $2 billion in the pipeline for accelerators (data-center GPUs and Gaudi NPUs). Although many chip businesses would love to have $2 billion in customer interest, “in the pipeline” is a vague achievement. Some of these opportunities won’t materialize, and we don’t know how strict Intel is in qualifying them—especially in light of the pressure to pump up this figure. Those that do materialize will translate into sales spread over multiple quarters.
Sizable as it is, a $2 billion pipeline is a paltry reward for Intel’s years of investing in data-center AI compared with Nvidia’s quarterly revenue. The forthcoming Gaudi3 is another attempt to catch Nvidia, and Intel will follow it with a new GPU-based architecture, Falcon Shores. To paraphrase Emerson (or is it Omar?), Falcon best not miss.
For its part, AMD is coming at both the AI- and the server-processor kings. Also squeezed by cloud service providers shifting spending away from CPUs to GPUs/NPUs, the company now has strong products in both categories. It indicates MI300 revenue is ramping nicely—enough to more than offset 1Q24 Epyc weakness. The MI300’s performance, GPU-based architecture, and availability in systems and boards resembling those built with Nvidia’s chips have struck a chord to a degree that Gaudi—despite its positive attributes—hasn’t. If recent AMD and Intel revenue trends continue, AMD’s data-center business will surpass that of Intel’s between 3Q24 and 3Q25 (inclusive).
PC
The PC-processor market has been a mess. If you look only at websites and YouTube channels for PC enthusiasts, you could be forgiven for thinking that AMD is catching up to Intel in the PC CPU market. The reality is that Intel CCG is far ahead—six times bigger, as Figure 2 shows. The company’s quarterly revenue swings are about as much as all of AMD’s PC (Client) revenue. For the year, AMD’s Client business broke even (with respect to operating income), and Intel’s reached 22% after a less-profitable 2022.
Both AMD and Intel chalked up weak 2023 quarterly revenue to customers’ inventory depletion. Having bought too much during the pandemic boom, they ordered less this past year. Both guide to a normal, seasonally down first quarter.
Meanwhile, Nvidia’s PC business (Graphics) has steadily grown, approaching the level last seen in 1Q22 before macroeconomic factors and a bursting cryptomining bubble deflated GPU revenue. The company has since employed Apple’s iPhone strategy, raising prices above $1,000 for a product that once cost a few hundred dollars. As with PC CPUs, AMD offers a competitive product but has been unable to take share from the GPU leader. Intel’s most recent PC GPU attempt may be its best yet but hasn’t affected the market.
Embedded
Taking Intel’s NEX, AMD’s Embedded, and Microchip’s Mixed-Signal MCU revenue as proxies for the overall embedded market, we can see in Figure 3 that it’s in a cyclical downturn. Like other embedded-processor suppliers during the pandemic, Microchip required customers to make noncancellable, nonrefundable (NCNR) orders to help manage the shortage and discourage overordering. These NCNR policies might have constrained bookings and avoided disruptive cancellations (as Henry Nicholas, former CEO of Broadcom once said, “We don’t rent chips.”), but a cyclical downturn was inevitable.
Intel’s NEX includes Ethernet components, which have propped up that business. AMD, by contrast, counts FPGAs in its embedded sales. These chips include the popular Zynq family and its successors, which are as much processor as programmable logic. These, along with embedded x86 chips, combine to make up the majority of AMD’s revenue in this category, which has recently tracked Microchip MCU’s downward trajectory.
Caveats
Note that we’re examining only companies’ publicly disclosed revenue. We don’t doubt their veracity, but reporting-unit composition is murky. Moreover, it differs among vendors. For example, AMD reports FPGAs as embedded sales, whereas Intel puts them in the data-center bucket. Per above, Nvidia has substantial system revenue, whereas the other vendors almost exclusively sell standalone semiconductors. Also, estimates for 1Q24 are tenuous, based on the earnings-call guidance.
Bottom Line
The processor business is not healthy. The embedded market is contracting. The PC segment is resuming growth but is still below pandemic-era highs. The data-center market is doing well, but Nvidia is capturing an outsized piece. Rivals might find themselves finally capturing data-center GPU/NPU share from Team Green just as the AI bubble bursts.