2021 Prediction Came True... The 'Intel Killer' Was Nvidia? [Tech Talk]

NVIDIA Absorbs Intel's Revenue
Data Center CPU Sales Also Decline

Intel and Nvidia. The two giants of the data center market. However, one makes central processing units (CPUs) while the other leads in graphics processing units (GPUs). How did these two companies, seemingly operating in separate domains, end up threatening each other's market share?

Intel Down 10%, Nvidia Up 400%

Pat Gelsinger, Intel CEO (left), and Jensen Huang, NVIDIA CEO [Image source: Intel, NVIDIA]

In fact, the 'prophecy' that Nvidia would threaten Intel came from within Intel itself. Dylan Martin, editor-in-chief of the U.S. technology media outlet 'CRN,' recently revealed that in 2021, Intel CEO Pat Gelsinger expressed concern to him, saying, "It won't be ARM but Nvidia that will become Intel's rival in the data center market."

About two years later, in the fourth quarter of last year, the fortunes of the two companies diverged sharply. Intel's data center market revenue was about $4 billion, down 10% year-over-year, while Nvidia recorded a 400% surge to $18.4 billion.

In the data center market, Intel produces CPUs, and Nvidia produces GPUs. At first glance, even though both serve data centers, they seem to operate in different areas. However, CEO Gelsinger's gloomy yet sharp prediction was very accurate.

How Did Nvidia Take Over Intel's Share?

Generally, the CPU handles the computer's tasks, while the GPU acts as a booster that 'accelerates' specific functions (graphics processing, AI, etc.) alongside it. The photo shows an H100 GPU board combined with the Grace CPU.
[Image source=NVIDIA]

Recently, the dominant trend in data centers is undoubtedly artificial intelligence (AI). When building AI data centers, the two most important chips are CPUs and GPUs. While GPUs are commonly known as AI chips, more precisely, they are 'accelerators' that amplify AI task processing speed. Meanwhile, most computer tasks are handled by CPUs. Roughly speaking, the CPU is the motor of AI, and the GPU acts as a booster that makes that motor run much faster.

The problem lies in the configuration of data centers. Unlike a typical PC, which has one CPU and one GPU, AI workloads usually allocate 2 to 4 GPUs per CPU. This means IT companies building AI data centers allocate a much larger budget to GPUs than to CPUs.

For example, if company A has a data center construction budget of 100, in the past, the budget might have been split 50:50 between CPUs and GPUs, but now it can skew as much as 20:80.

That's not all. As the importance of AI workload acceleration grows, companies are eager to acquire the latest generation GPUs. Meanwhile, the processing speed of the latest generation GPUs continues to increase. The more this happens, the less relative contribution CPUs make within data centers. In other words, when upgrading processors in data centers, replacing only the GPU with the latest specs while adopting cheaper, older generation CPUs does not significantly impact overall processing power.

The Cause of Intel's Losses: Weakening Pricing Power

[Image source=Yonhap News]

This is the most critical problem for Intel, a data center CPU vendor. As the GPU share in data centers grows, customers increasingly seek inexpensive, older generation Intel CPUs rather than the latest ones. If this trend continues, even if Intel maintains its market share, its revenue will keep declining because it gradually loses pricing power.

Pricing power is very important in the semiconductor industry. Vendors like Nvidia, AMD, Intel, and ARM raise prices by updating their technologies to the latest versions, thereby increasing margins. The expanded resources are then invested in new research and development (R&D) and capital expenditures (CAPEX) to outpace competitors.

So, do AMD and ARM, who also make data center CPUs like Intel, face similar issues? AMD is in a better position because it also stands out in the data center GPU market. ARM, on the other hand, is a relatively latecomer compared to the other two, and Nvidia has already adopted ARM's CPU cores to develop its own CPU called 'Grace.' For now, the loose alliance with Nvidia is actually helping ARM grow.

Intel Prepares AI Accelerators: Is There a Chance for a Comeback?

AI accelerator 'Gaudi' developed by Habana Labs, a subsidiary of Intel.
[Image source=Intel]

Of course, Intel is not just passively taking losses. Since CEO Gelsinger took office, Intel has devoted considerable effort to its AI accelerator roadmap. It acquired the Israeli AI accelerator fabless company 'Habana Labs' and has been steadily developing its own accelerators called the Gaudi series.

If the Gaudi chips can establish a three-way competition with Nvidia and AMD in the data center GPU market, Intel's current 'bleeding' in the data center CPU market could stop.

By providing high-performance Intel CPUs to power the Gaudi accelerator chips, Intel could reclaim lost market share. However, given Nvidia's strong position riding the massive wave of the AI boom, Intel is likely to endure a challenging period for the time being.