Nvidia-Arm Deal: Does It Make Technological Sense?

Article By : Junko Yoshida

The proposed Nvidia-Arm deal is no longer an idle speculation but appears real. So the next question is: does it make technological sense? No, it does not.

That Nvidia intends to buy Arm from Softbank now appears to be more than just idle speculation.  Nvidia initially approached SoftBank with a proposal to purchase Arm for more than $32 billion in a cash-and-stock deal, according to two news sources.

Bloomberg and the Financial Times independently confirmed the courtship.

There’s no guarantee that the discussions will result in a sale, but just raising the possibility spurred sources in the tech sector to reach out to EE Times to tell us how little sense the deal makes.

Put simply, the Nvidia-Arm deal is a tough sell. Even tougher is coming up with any semblance of positive narratives for this M&A. Forget about “synergy” or “win, win.”

The merger would be a disaster.

In looking at the potential downside, Mike Demler, senior analyst at The Linley Group, said “key people might defect. More Arm customers might look to RISC-V,” referring to an open standard instruction set architecture based on RISC principles. Our colleague Bolaji Ojo called Nvidia-Arm move “reckless.” Kevin Krewell, principal analyst at Tirias Research, in our interview, called it “the most audacious move by Jensen Huang as CEO of Nvidia.”

Nvidia Buying Arm Would be Beyond Stupid

“There are just so many roadblocks,” pointed out The Linley Group’s Demler. Think, for example, of “government approvals, customer reactions.” Given the history – China’s fatal stall on the Qualcomm-NXP deal in 2018, Krewell at Tirias Research, said, “There is a very serious challenge [to] clearing the regulatory hurdle.”

Krewell also added, “It would also put Nvidia directly in the middle of the fight between the U.S. government and Huawei, and other Chinese Arm licensees.” Who needs that?

The loudest alarm might be triggered — if the deal goes through — by the uncertainty Nvidia brings to the entire Arm ecosystem. Arm licensees — among them Apple, Qualcomm, Broadcom, NXP, STMicroelectronics and MediaTek — “would be concerned,” said Krewell, as Nvidia competes with them. He added, users of Arm’s microcontroller (Cortex-M) “may feel Nvidia would not invest,” thus resulting in the loss of their business in the future.

Demler agreed. “How the companies could alleviate that concern is the biggest issue.” If Nvidia is serious about maintaining Arm’s IP business, “There would need to be an arm’s-length distancing of the CPU licensing business.”

How to justify the deal?
But if you are in Nvidia’s shoes, asked to justify the proposed acquisition, how would you explain it?

Potentially, there are two answers:

  • The deal is driven by potential benefits for Nvidia to own both GPU and CPU platforms and their ecosystems.
  • Nvidia plans to diversify its business model, now with an IP licensing component.

Late last week, Demler tweeted, “As much as IP licensing is a poor fit for Nvidia, imagine this. GPU-compute proliferation on CUDA platform, coupled with Arm CPUs = serious competitor to x86.”

Asked if he meant that “there will be a war” against Intel, Demler said, “No.”  In his opinion, there is “nothing new to the competitive battle between Arm and Intel CPUs.” That’s already well known. Demler was thinking out loud the potential benefits of combining Arm CPUs with Nvidia’s GPUs.

“You’d have the leading power-efficient CPU platform with the leading platform for GPU compute and AI training. Jensen (Huang) often talks about the advances their GPUs have made compared to Moore’s Law, and how they’ve fostered a new computing paradigm. And after years trying, Arm licensees are finally making some inroads into traditional x86 strongholds, such as high-performance computer (HPC) and PCs.”

He concluded, “So putting all competitive, legal, and technical obstacles aside for a moment, from a strategic point of view, Arm could help Nvidia create a more powerful computing platform based on the combination of their technologies, and Arm’s expertise in IP licensing would help proliferate that platform.”

Similarly, Krewell noted, if consummated, the deal “would give Nvidia control over the most pervasive instruction set on the planet. It would allow Nvidia to drive its AI cores and CUDA ecosystem into the Arm ecosystem.”

Harder to grasp, however, is how seriously Nvidia is thinking about making IP licensing one of its business pillars?

In theory, Krewell believes the acquisition of Arm “could also diversify Nvidia with an IP business, much like Qualcomm is structured.” Qualcomm has QCT (Qualcomm CDMA Technologies) semiconductor business, while also running the QTL (Qualcomm Technology Licensing) licensing business.

Demler explained that IP licensing is a direction Nvidia once considered, back when. He cited Nvidia’s 2013 announcement stating its intention to license Kepler GPU cores.

In that 2013 blog, Nvidia wrote:

… Not so long ago, we only made and sold GPU chips, albeit the world’s fastest ones. Five years ago, we introduced Tegra, a system on a chip. More recently, GRID – a complete system that streams cloud games and other graphics-rich content — as well as the SHIELD gaming portable have been unveiled.

But it’s not practical to build silicon or systems to address every part of the expanding market. Adopting a new business approach will allow us to address the universe of devices.

So, our next step is to license our GPU cores and visual computing patent portfolio to device manufacturers to serve the needs of a large piece of the market.

Back then, Demler said, “I had said that Nvidia was underestimating how difficult it is to repurpose cores designed for internal use so that they would be usable by external customers. The fact they never actually licensed any cores is a testament to that, although they licensed an earlier GPU core to Sony for the Playstation 3.”

Demler added, “Since 2013, Nvidia has licensed NVDLA, so there is precedent for them wanting to expand the footprint of their platforms.” NVDLA, the Nvidia Deep Learning Accelerator, is a free, open architecture that promotes a standard way to design deep learning inference accelerators.

Remember MIPS?
Operating a successful IP licensing business is a lot harder than it seems. It needs a big anchor customer or two, while also devoting time and effort on building a broader ecosystem around the core. Arm by far is one of the most successful IP companies in the semiconductor industry.

Earlier this year, Imagination Technologies sold MIPS to Wave Computing.

MIPS, which started out as a computer company (MIPS Computer Systems) and processor designer, went IPO in 1988. Its architecture quickly got attention from Silicon Graphics (later renamed as SGI), which acquired MIPS in 1992. SGI justified the deal as a means to secure the supply of future generation of MIPS microprocessors. It was also driven by “the need for SGI to compete with Sun/SPARC and DEVC/Alpha,” explained Krewell.

SGI was MIPS’ big customer, but it also became its parent company.

Did it help? Not really.

MIPS’ fate took a complete U-turn in 1998 when SGI decided to migrate its workstations to Intel’s Itanium architecture. SGI first spun off MIPS as an IP licensing firm, and completely spun it out in 2000, by distributing all its interest as dividends to stockholders.

The rest is the history.

What’s next?
Krewell noted, “While we were just getting used to Nvidia as an enterprise and AI company, this [acquisition of Arm] would make Nvidia the largest IP company.”

Whether Arm can survive under Nvidia’s stewardship is a relevant question. Krewell suggested, “I would like to see an Arm IPO as an independent company again. Alternatively, a consortium of companies [could] invest in Arm and give all the stakeholders a seat at the table. Capital is relatively cheap right now, raising the money to buy Arm should be possible.”

Subscribe to Newsletter

Leave a comment