Free Download Video

How Nvidia’s Arm acquisition will drive AI to every edge

Nvidia is sitting pretty in AI (artificial intelligence) right now. For the next few years, most AI systems will continue to be trained on Nvidia GPUs and specialized hardware and cloud services that incorporate these processors.

However, Nvidia has been frustrated in its attempts to become a dominant provider of AI chips for deployment into smartphones, embedded systems, and other edge devices. To address that strategic gap, Nvidia this past week announced that it is acquiring processor architecture firm Arm Holdings from SoftBank Group and the SoftBank Vision Fund.

Once the acquisition closes in the expected 18 months, Nvidia will retain Arm’s name, brand identity, management team, and base of operations in Cambridge, United Kingdom. It will also expand Arm’s Cambridge-based research and development facility, while establishing an Nvidia research facility, developer training facilities, and startup incubator at the site. Arm will operate as an Nvidia division.

This is truly a landmark deal. Nvidia will almost certainly integrate its GPU technology into the core smartphone, IoT, and embedded chip architectures designed by Arm, thereby driving its AI technologies ubiquitously to edge devices everywhere.

What follows are the principal respects in which Nvidia will benefit from acquiring Arm.

Bolstering Nvidia’s bottom line

Nvidia is picking up Arm for a $40 billion consideration of cash and shares, which makes it far larger than the $7 billion acquisition of Mellanox that closed in April.

Arm comes into Nvidia as a significant cash cow from which its new parent will almost certainly fund ambitious new projects and fill-in acquisitions going forward. Nvidia’s acquisition of Arm is expected to be accretive to the acquirer’s bottom line. Once the deal closes, Arm will start contributing its considerable profits to its new parent’s own net income immediately. Nvidia’s income statement will gain millions of dollars in Arm’s annual licensing fees and billions of dollars in royalty fees.

The transaction, which is subject to the usual closing conditions and regulatory approvals, represents a middling return on SoftBank’s $32 billion outlay when it took Arm private in 2016. Nvidia’s bargain on the deal stems largely from the fact that SoftBank had run into a cash crunch after losing billions of dollars due to the pandemic and bad bets on Uber and WeWork. SoftBank will acquire an ownership stake in Nvidia of no more than 10 percent, effectively making it one of Nvidia’s largest shareholders.

Giving Nvidia a new competitive lever

Nvidia’s Arm acquisition comes at a time when Intel’s next generation of chips has encountered major delays. Nvidia will be able to leverage the Arm acquisition to contend with Intel across a wide range of mobile, edge, embedded, gaming, and IoT end points. Arm provides the basic architectures for the low-power central processor chips within smartphones and tablets from such licensees as Apple, Samsung, and Qualcomm.

Through licensee Apple, Arm-based processors will replace Intel processors in the next generation of Mac computers. The deal will give Nvidia a better shot at displacing Imagination Technologies as Apple’s GPU supplier for its iOS devices. And it will enable Nvidia to serve chip makers who are adapting Arm designs to work in servers and PCs, which have long been Intel’s stronghold.

Boosting Nvidia’s processor market position

Nvidia is rapidly becoming the dominant vendor of processors for cloud-to-edge deployments of AI. Even before the Arm acquisition, Nvidia was racking up record sales, saw its share price double this year, and overtook Intel as the most valuable U.S. semiconductor company. At the same time, Intel continued to stumble in its efforts to bring to market a credible GPU alternative to Nvidia’s flagship offerings.

Nvidia’s Arm acquisition will boldly advance its position in the smartphone market, as well as in the markets for embedded, IoT, and other edge devices, which are all segments from which Nvidia has largely been absent. By contrast, Arm boasts a near-monopoly in providing IP (intellectual property) for mobile device chip architectures. Arm currently licenses designs for third-party microprocessors that power approximately 90 percent of the world’s smartphones and in many other types of edge and mobile devices. Arm’s energy-efficient designs have been used to create 160 billion chips that are manufactured and sold by more than 1,000 licensees.

Diversifying Nvidia’s solution and technology portfolio

Nvidia is acquiring a firm with a complementary technology portfolio, business model, and go-to-market approach.

Nvidia doesn’t design CPUs, which are the core of Arm’s chip IP. Nvidia doesn’t license IP to semiconductor companies, which is Arm’s principal business model. And Nvidia doesn’t compete in the mobility market, which is where Arm’s primary licensees operate.

Also, Nvidia doesn’t own any chip fabrication plants, but outsources production of its chip designs to specialized foundries. Arm, by contrast, doesn’t outsource its chip designs at all, but instead licenses its IP to other vendors that fabricate them, either in their own facilities or outsourced.

Once the acquisition is finalized, Nvidia plans to build an Arm-powered supercomputer to support AI R&D at Arm’s Cambridge location. Nvidia also plans to expand Arm’s IP licensing portfolio with Nvidia technologies, especially the latter’s market-leading GPUs.

The converged firms will be able to address cloud-to-edge opportunities that combine Nvidia’s AI solutions with Arm’s vast array of licensees. Even before this latest deal, SoftBank had driven Arm’s diversification into new opportunities to license its IP into partnerships in the data center, automotive, IoT, and network processing markets. Arm had already announced that it was designing its Pelion software IP for a growing range of low-power, high-performance AI apps running on edge devices.

It’s unclear where Arm’s AI investments will land in the converged firm’s strategies and solution portfolio. To support its ambitions in the AI arena, Arm already leverages IP from its recent acquisitions of Stream Technologies and Treasure Data. These technologies support ingestion, storage, and management of data to be used in building and training machine learning (ML) models that can execute transparently across CPUs, GPUs, and neural network processing units. Arm has also been investing in tools that allow ML models to be dynamically updated on edge devices and also to support secure, distributed, cross-node ML computations.

It’s important to note that Arm’s processor designs also serve as the basis for Amazon Web Services’ Graviton2 processors and for Fujitsu’s A64FX processors that are used in the latter’s Fugaku supercomputer. It remains to be seen whether these Arm licensees—whose respective AI solution portfolios directly compete with Nvidia—will adopt any add-on AI core tech.

Nevertheless, even if such licensees balk at adopting Nvidia’s AI, we can expect the combined Nvidia-Arm to offer more of its own chips (GPUs, CPUs, etc.) to power AI in “big-core” data center, high-performance computing, and supercomputer deployments. We should also expect Arm to aggressively offer such IP to its vast ecosystem of licensees.

Locking down a strategic supplier for Nvidia

Nvidia, in acquiring Arm, will be securing its future access to Arm’s processor technology, while keeping it out of the hands of competitors. If this deal is approved, many semiconductor companies that would otherwise have simply been Nvidia’s rivals will also become its customers.

Nevertheless, Nvidia has announced its intention to continue Arm’s open licensing ecosystem, a pledge that will be absolutely necessary in order to secure the necessary regulatory approvals. Nvidia has pledged to continue Arm’s policy of customer neutrality, licensing IP to companies who may compete with Nvidia in GPU, AI, and other product segments.

Extending Nvidia’s market reach and scale

Last but not least, Nvidia is acquiring a vendor with a much larger market reach and scale than its own.

For starters, the deal will expand Nvidia’s reach in the development community from the current 2 million to more than 15 million. More significantly, Nvidia shipped about 100 million chips in the past year, while Arm’s more than 1,000 technology partner licensees shipped more than 22 billion (with a “b”) chips last year and more than 180 billion to date.

Facing opposition

Nvidia’s pledge to continue Arm’s open licensing and customer neutrality will be a key element for running the gauntlet of regulatory challenges that are sure to ensue.

China looked long and hard at the recently approved Mellanox acquisition and may prove difficult to placate, especially considering strains in its geopolitical posture vis-à-vis the United States.

Nvidia’s assurance that Arm’s site and team will remain intact in England will prove essential in securing that country’s approval. However, political forces at work in the United Kingdom might cause the regulatory approvals to be difficult to secure.

Seemingly to prevent the matter from being a regulatory distraction during the approval process, Nvidia’s Arm acquisition will not include Arm‘s two IoT Services Group software businesses. SoftBank had previous announced a plan to spin off the two businesses into new SoftBank-owned stand-alone entities. But nearly three weeks ago, Arm said it was halting the spin-off plans.

The biggest potential challenge for a combined Nvidia-Arm, even after the deal clears all regulatory hurdles, will be whether Arm licensees in the microprocessor industry will be comfortable sourcing this key technology from a competitor. Arm’s reputation as the “Switzerland of the semiconductor industry” is at stake. Nvidia rivals—such as AMD—may seek out alternative sources if they perceive that the deal gives CEO Jensen Huang’s firm an unfair advantage in the battle to bring AI to edge devices.

Related posts

The Other CFO: The Rise of the Cloud Economist


IonQ claims it has built the most powerful quantum computer yet


How to use the ProblemDetails middleware in ASP.NET Core