Tea leaf reading regarding Chinese regulators’ decision on Nvidia’s intended takeover of Arm may have taken a negative — or at least a time consuming — turn. A report by IT publication The Information states that regulators in China have not yet begun a formal review of the deal, which was announced last September.
“Regulators have had informal discussions with both companies, but nothing concrete has emerged,” stated an article in today’s Data Centre Dynamics on the development. “Regulatory approval in the country was expected to be difficult due to the ongoing trade war with the US, and the fact that if Arm was owned by an American company, it could be subject to … export limitations.”
The $40 billion acquisition would be a significant move by Nvidia in rounding out its microprocessor product portfolio. The Santa Clara, CA-based company, having achieved global leadership in GPUs, wants to add Arm CPUs to an offering that, combined with Nvidia’s 2020 acquisition of high performance networking company Mellanox, could enhance Nvidia’s growing presence in data center servers, HPC and supercomputing.
When the Nvidia-Arm deal was announced nearly a year ago, Nvidia said it expected regulatory approval within 18 months. Now, with the news out of China, that timeline may need to be revised. The Data Centre Dynamics story also reported that the deal “is also behind schedule in Europe – but because Nvidia has yet to submit its paperwork to the European Commission,” according to a story in The Telegraph . “That body has now recessed for a summer break, and is unlikely to look at the acquisition before September at the earliest. A decision could then take more than six months,” stated the publication.
The news regarding Chinese regulators comes amidst recent attacks by the PRC on some of China’s largest technology companies. This morning, Tencent Holdings stock fell after a state-owned Chinese newspaper attacked online gaming as “opium for the mind,” “fueling investor concerns that the companies’ popular games could be swept up into a broader regulatory crackdown,” reported The Wall Street Journal .
“In recent months, China has intensified scrutiny of big technology companies over issues such as data security, monopolistic behavior and financial stability, sparking a steep selloff in the shares of companies like Tencent and Alibaba Group Holding Ltd.,” the Journal stated.
All this comes during intensifying trade antagonism between China and the U.S., much of it focused on technology. Nvidia’s ability to navigate this difficult and unpredictable trade and geopolitical landscape will be a major test of its communications and government relations skills.
Interesting, too, are the evolving product portfolio strategies of the major chip companies. Having seen over the past decade near blanket dominance in HPC-class servers of Intel CPUs encroached upon and augmented by GPU and FPGA accelerators – what HPC industry analyst Addison Snell, CEO of Intersect360 Research, has called “technology disaggregation” – we now may be seeing an attempt at re-aggregation by the big chip companies. That is, GPUs, FPGAs, Arm and custom AI chips are increasingly used in tandem with CPUs for advanced computing workloads. But instead of combinations of chips from combinations of vendors, those vendors are moving to become single-source providers of multiple architectures.
Cases in point: the world’s second and third most powerful supercomputers, Summit and Sierra at Oak Ridge National Laboratory and Lawrence Livermore National Laboratory, respectively (both installed in 2018), are powered by IBM CPUs and Nvidia GPUs. But next-gen, state-of-supercomputing exascale systems will use chips from single vendors. Frontier, scheduled to be delivered to Oak Ridge National Laboratory later this year, will be powered by AMD CPUs and GPUs. Aurora, slated for Argonne National Laboratory, will be equipped with Intel “Sapphire Rapids” Xeon CPUs and Intel’s “Ponte Vecchio” GPUs.
Thus, the Nvidia-Arm move would be in keeping with the broadening chip choices of AMD and Intel, both of which are building out CPU-GPU product strategies (Intel also has FPGAs and AMD will likely soon have them, assuming its acquisition of Xilinx is approved). The result could be performance advantages from tighter multiprocessor integration. And it could mean expanded market reach and tighter account control for the chip vendors whose customers rely on only them for of their chip needs.