Skip to main content

4 Interesting Things Nvidia Just Shared About its Plans for ARM

Among other things, Nvidia plans to dial up ARM's server CPU R&D and promote its GPU and AI IP to ARM clients.
  • Author:
  • Publish date:

During a series of press and analyst calls, Nvidia  (NVDA) - Get Free Report CEO Jensen Huang shared some details about how his company aims to make its planned acquisition of ARM pay off.

To recap: Nvidia is paying between $35 billion and $40 billion -- the exact sum will depend on whether $5 billion worth of earn-out payments to ARM owner SoftBank are made -- to buy ARM. ARM controls what’s by far the world’s most widely-used CPU instruction set architecture (ISA), while licensing its ISA, CPU core designs and other IP to most of the world’s chip developers.

Nvidia asserts the cash-and-stock deal will be immediately accretive to non-GAAP EPS at the time of closing, and forecasts (amid expectations of close regulatory scrutiny) that the deal will close within 18 months. Nvidia shares closed up 5.8% to $514.89 in Monday trading.

Here are a few noteworthy things that Huang has shared about Nvidia’s plans for ARM since the deal was announced on Sunday afternoon.

1. Bringing Nvidia’s GPU and AI IP to ARM-Powered Chips Is a Top Priority

On multiple occasions, Huang emphasized that Nvidia sees a large opportunity to license its GPU, AI co-processor and video encoder/decoder technologies to chip developers creating ARM-powered silicon.

“Last year, ARM [licensees] sold 22 billion chips,” Huang pointed out. “We believe all of those chips in the future are going to include AI computing. We believe all of those chips in the future will [support] accelerated computing and we don't have to build all of them. We would love to extend our architecture to all of them.”

The growth of ARM-powered silicon over the years. Source: Nvidia.

The growth of ARM-powered silicon over the years. Source: Nvidia.

Those 22 billion chips (22.8 billion, to be precise) include the processors powering the overwhelming majority of the world’s smartphones and tablets, as well as processors powering things such as AR/VR headsets, infotainment systems, smartwatches and smart speakers/displays. All of these products could potentially make use of Nvidia’s GPU and/or AI co-processor IP.

2. Nvidia Plans to Continue Licensing ARM’s Mali GPU Designs

When asked if (in light of Nvidia’s plans to license its own GPU IP) Nvidia plans to stop supporting ARM’s Mali GPU designs, which have seen meaningful traction in the mobile processor space, Huang said that won’t happen, while insisting that Mali and Nvidia’s GPU IP complement each other.

“One of them is a computing platform with a rich ecosystem of software tools and a lot of third-party developers,” Huang said, referring to Nvidia’s own GPU architectures. “There are people who are very satisfied and very happy with Mali, we will continue to offer that. “For people who would like to have a richer ecosystem around AI computing, accelerated computing, and all of the games and tools...that are built on top of our ecosystem, our GPUs will be quite wonderful for that.”

Huang also said that Nvidia plans to continue developing chips relying on the open-source RISC-V (pronounced risk-five) CPU instruction set, which among other things is used by Nvidia to develop microcontrollers for its GPUs.

3. Nvidia Insists it Doesn’t Plan Any Changes to ARM’s Business Model

Nearly every major chip developer on the planet is an ARM licensee. Even Intel  (INTC) - Get Free Report, though getting most of its revenue from CPUs relying on the rival x86 ISA, uses ARM CPU cores within FPGAs and ADAS vision processors.

Given that certain Nvidia products compete against offerings from ARM licensees such as Intel, Qualcomm  (QCOM) - Get Free Report and Xilinx  (XLNX) - Get Free Report, many industry observers are nervous about the conflicts of interest an Nvidia-ARM deal could pose. Some, such as ARM co-founder Hermann Hauser, want a guarantee that Nvidia won’t get any preferential treatment from ARM post-acquisition.

How ARM's business model works. Source: Nvidia.

How ARM's business model works. Source: Nvidia.

Sensitive to such concerns, Huang promised to maintain ARM’s current business model, in which the company licenses its IP to all comers while typically collecting up-front fees and per-chip royalties. He also promised Nvidia will retain ARM’s name and brand, keep the company headquartered in the U.K. and “protect the confidentiality” of any work that ARM does with other chip developers.

4. Nvidia Plans to ‘Turbocharge’ ARM’s Server CPU Efforts

Though the lion’s share of the world’s server CPU sales still involve Intel and AMD’s  (AMD) - Get Free Report x86 CPUs, ARM-powered server CPUs from the likes of Marvell Technology  (MRVL) - Get Free Report, AWS and private Ampere Computing have begun making some headway, particularly among Internet/cloud service providers. And there’s a growing belief that -- since it will result in many developers coding on ARM-powered machines -- Apple’s  (AAPL) - Get Free Report plans to migrate its Mac lineup to ARM-powered CPUs will help drive greater ARM server CPU usage.

Huang, whose company gets over 40% of its revenue from data center products following the recent Mellanox Technologies acquisition, stressed that Nvidia will build on ARM’s already-sizable investments in its Neoverse platform for server CPUs and other “infrastructure” processors. He also signaled Nvidia plans to use its server GPUs, Mellanox’s data processing units (DPUs) and ARM’s server CPU IP to develop end-to-end silicon and software platforms for servers.

“In the data center, NVIDIA will turbocharge ARM's R&D and meet cloud computing customers' demand for a higher-velocity ARM server CPU roadmap,” Huang said. “We are a computing platform company that is able to innovate from the chips to the systems, system software, all the way to the application stacks on top.”

Huang also left the door open to Nvidia developing and selling its own ARM server CPUs, as opposed to simply licensing ARM’s IP to third-party CPU developers. However, he also suggested that whether or not Nvidia develops its own server CPUs matters far less than creating an end-to-end platform.