NVIDIA (NVDA +2.71%) just unveiled Grace, its first high-end CPU for data centers. The chip, named after the computer science pioneer Grace Hopper, is an Arm-based chip that will be used to handle demanding AI and high-performance computing (HPC) tasks.
NVIDIA claims Grace can deliver up to 30 times higher aggregate bandwidth than today's fastest servers, and up to 10 times higher performance for applications that process terabytes of data. It says the design will "enable scientists and researchers to train the world's largest models to solve the most complex problems."
NVIDIA's jump from data center GPUs to CPUs isn't surprising, but it could raise red flags at Intel (INTC +4.96%), which still controls nearly 97% of the server CPU market, according to PassMark Software. Let's see why NVIDIA is expanding into Intel's strongest market.
Image source: NVIDIA.
What's NVIDIA's game plan?
NVIDIA is often associated with gaming GPUs, but data centers also use its high-end GPUs to process AI and machine learning tasks.
A CPU processes pieces of data individually, while GPUs process a large array of integers and floating-point numbers simultaneously. Together, CPUs and GPUs process AI tasks much faster than stand-alone CPUs or GPUs.
NVIDIA's DGX Station A100, which it calls an "AI data center in a box," pairs its GPUs with AMD's (AMD +5.24%) Epyc x86 server CPUs. Newer versions of that box could pair NVIDIA's Grace CPUs with its own GPUs instead to showcase the combined processing power of its chips.
In addition to selling new data center GPUs, NVIDIA also acquired the data center network equipment maker Mellanox last April. As a result, its total data center revenue surged 124% to $6.7 billion in fiscal 2021 and accounted for 40% of its top line.
But that's not all -- NVIDIA is also trying to buy Arm from SoftBank (SFTB.Y +2.06%) to fully close the loop.
If NVIDIA buys Arm, it will control the chip architecture that powers most of the world's smartphones. Arm doesn't sell any chips, but it collects licensing fees and royalties from Arm-based chipmakers like Qualcomm and Apple.
NVIDIA pays licensing fees to Arm for its lower-end Tegra CPUs and its new Grace CPUs, but a takeover would eliminate those costs. Arm is still hardly used in data center chips compared to the x86 architecture used by Intel and AMD, but NVIDIA's Grace could finally disrupt that monopoly.
Intel claims to be playing offense
Intel's data center revenue rose 11% to $26.1 billion, or 34% of its top line, last year. However, the segment's sales declined year over year in the third and fourth quarters amid pandemic-related headwinds and tougher competition from AMD's Epyc CPUs.
Image source: Getty Images.
Intel also continued to struggle with an ongoing chip shortage as its internal foundry fell behind Taiwan Semiconductor Manufacturing (TSM +2.83%) in the "process race" to manufacture smaller chips. AMD, which outsourced its chip production to TSMC instead of manufacturing them internally, pulled ahead of Intel with more advanced chips.
Intel's new CEO, Pat Gelsinger, who took over in February, is trying to resolve its R&D and production issues. However, he recently warned that it could still take a "couple of years" to address its current shortages.
Analysts expect these challenges to cause Intel's revenue and earnings to decline 7% and 14%, respectively, this year. They expect NVIDIA's revenue to rise 34% this year with 35% earnings growth.
Gelsinger recently told Fortune that NVIDIA was merely "responding" to Intel's integration of new AI features in its newest CPUs, and that it remained "on the offense, not the defense" in the AI market. Intel also recently launched its first discrete GPUs for data centers to challenge NVIDIA.
That's a bold claim, but NVIDIA's surging data center GPU sales and the potential synergies between Mellanox, Arm, and its Grace CPUs could all spell trouble for Intel. Competitive pressure from AMD in both PCs and data centers could exacerbate the pain.
The bottom line
It's unclear if NVIDIA's Grace CPUs will gain any ground against Intel's industry-standard Xeons in the data center market. However, it represents a bold step in the right direction and sparks fresh hope for Arm-based CPUs to make progress in the data center market again.
Intel claims to be playing offense, but it seems like it's destined to play defense against NVIDIA's ambitious efforts for the foreseeable future.
