Competition and Markets Authority Raises Questions of NVIDIA Purchase of Arm

On the 20th of August 2021, the UK Government announced that the Competition and Markets Authority (CMA) had completed its Phase 1 review and found that NVIDIA’s proposed purchase of Arm raised competition concerns.

Arm is a UK headquartered company which supplies semiconductor intellectual property (IP) based on a specific instruction set architecture (ISA). This IP is used by semiconductor suppliers and systems-on-chip (SoC) designers to produce semiconductor chips and related products for a huge variety of applications from mobile phone handsets, computers, televisions, smart watches, and automobiles to networking equipment. The range of equipment that relies on Arm IP is staggeringly large. Arm does not manufacture these processors, instead it provides its IP for a license fee, allowing its licensees to develop their own products by customizing the ISA or by manufacturing a product using the blueprints supplied. It is a very successful business model, although many would argue that using proprietary IP to manufacture processors and selling the finished product, which is what Intel and AMD do, is significantly more profitable. Arm is currently owned by SoftBank, a Japanese multinational conglomerate holding company.

NVIDIA is a US based company that supplies semiconductors including central processing units (CPU), graphical processing units (GPU), data processing units (DPU), and network interconnect products to customers globally. NVIDIA sells to original equipment manufacturers (OEMs), systems integrators, and solution builders, as well as directly into to the consumer market. NVIDIA maintains a rich ecosystem around its products that has attracted a solid following amongst developers and users because it easily facilitates powerful customization and interaction with its technology. The proposed purchase of Arm by NVIDIA is valued at US$40 billion.

The main concerns raised by the CMA are centered around the ability of the merged entity to restrict access to Arm IP that would lead to a substantial lessening of competition. In its Phase I review, the CMA found that the merged entity would have the ability and incentive to do so, and that the incentive was likely to outweigh the costs.

The CMA listed specific components and global sectors that gave it cause for concern if competition were restricted to the technology supply, namely, the supply of CPUs, network-interface controllers (SmartNICs), and GPUs for data center servers, and the supply of SoCs for high performance internet-of-things applications, automotive applications, and gaming consoles. That list covers a broad range of technologies, which, if they were to be adversely affected, would negatively affect the global economy significantly. We only have to look at the way the Auto industry has been affected by a shortage in semiconductor supply; delivery times for many vehicles has grown by many months as a result and has hit auto industry sales figures and profits. It has significantly driven up the demand for, and value of, secondhand vehicles. So, there is justification for the CMA to investigate this deal, and neither NVIDIA nor Arm would have expected otherwise; the SoftBank purchase of Arm in 2016 underwent a similar process. However, the dynamics of that deal were subtly different in that SoftBank was not in a position to benefit from negatively affecting the competitiveness of its licensees, rather to the contrary, it would benefit more by improving their competitive edge.

The CMA expressed a concern that the merged entity could harm NVIDIAs rivals, by refusing to supply, increasing the price, or worsening the quality of Arm semiconductor IP available to them. The CMA also expressed a concern that the merged entity has the ability and incentive to adversely affect competition in the supply of CPU and SmartNIC components market.

Perhaps more telling was the observation that the merged entity could restrict or degrade the interoperability between all three components and modify the interoperability between CPU, GPU, and SmartNICs to enhance NVIDIA’s products and undermine the operability of rival’s products.

Looking at current processing trends, we are seeing the convergence of artificial intelligence (AI), machine learning (ML), and traditional compute workloads. Increasingly we are seeing multiple processing units being deployed alongside each other to address these workloads. Current semiconductor trends are demonstrating that a large amount of investment in research and development (R&D) is taking place to optimize data flow and system calls, interoperability, between of all the major components highlighted by the CMA in their report.

In August 2021, Intel announced its Arc GPU product that will be offered alongside its CPU. At its August global architecture day, it claimed that its new architectures blended the best transistors for a given engine, connected them through advanced packaging, integrated high-bandwidth, low-power caches, and equipped them with high-capacity, high-bandwidth memories and low-latency scalable interconnects for hybrid computing clusters in a package, while also ensuring that all software accelerates seamlessly. ABI Research believes that Intel is investing considerable research efforts to optimize the interoperability and effectiveness of all their CPU, GPU, and DPU, with the aim of producing a highly advanced heterogeneous compute solution in line with NVIDIA’s Grace CPU.

NVIDIA’s Grace processors were released in April 2021. NVIDIA claims “Using licensed Arm IP, NVIDIA has designed Grace as a CPU specifically for giant-scale AI and HPC. Coupled with the GPU and DPU, Grace gives us the third foundational technology for computing, and the ability to re-architect the data center to advance AI. NVIDIA is now a three-chip company.”

Both above announcements clearly demonstrate that component interoperability is a significant area of research and has been for many years. It is hoped that advances made in this field will help the industry keep pace with the explosion in demand that AI and ML workloads have created. Here the CMA has identified one of the core areas of innovation focus that all semiconductor producers, NVIDIA, and its rivals, have been concentrating on in recent years, optimizing the flow of data and compute tasks between the various processing components to unlock significant performance gains. If the CMA hopes to prevent the merged entity from degrading interoperability between components for Arm’s current licensees, it will have to do so very carefully if it is to avoid shutting down one of the most productive areas of current processor research, an area that many would argue is absolutely essential to technology progression.

Furthermore, it would be self-defeating for NVIDIA to limit interoperability, withhold technology or disadvantage Arm’s licensees in any other way for a number of reasons.  First, NVIDIA has publicly stated its intent to sell AI and GPU IP through Arm’s channels.  Second, NVIDIA sees its largest opportunity in the x86 CPU space (servers, PCs) and needs the licensees to help build out the ecosystem.  Third, NVIDIA is counting on IP revenue from existing Arm contracts, many of which have very long terms. These are three reasons are why NVIDIA is unlikely to disadvantage Arm’s licensees post-acquisition.

ABI Research believes that healthy competition breeds innovation and is sure that no matter the outcome of NVIDIA’s proposed purchase of Arm, the two entities will continue to develop inspirational technology that will push their competitors and drive their licensees forward at pace.