Who Should Buy ARM?¶
There are rumors about NVIDIA in advanced talks of buying ARM. Softbank has been in need of cash injection for some time, and ARM itself is a good asset.
The question to ponder, more interestingly, is how the synergy would work out for ARM and its potential buyers.
NVIDIA has some presences in mobile. From its ill-fated Denver CPU to the modest success with Tegra X1 in Shield and Nintendo Switch, for one, it is not lack of trying. So far, NVIDIA has established itself as an interesting player in the mobile space, while not gaining meaningful market share in the otherwise vast mobile market.
On the other hand, NVIDIA’s data center business has enjoyed a strong growth trajectory in the past a few years. It has established itself, rightfully, as the go-to choice for any high-performance computing needs.
It is likely that the acquisition of ARM won’t be for the mobile market share. With a strong CPU design portfolio like ARM, NVIDIA can have more flexibility in its high-performance computing package design. Right now, NVIDIA’s flag-ship design is DGX A100, with its CPU being AMD Epyc. The past few years, in the high-performance computing world, the role of the CPU is transitioning from computing powerhouse, to a computation coordinator. This is particularly obvious from the accelerated development pace of PCIe 4 and PCIe 5 standard.
That won’t be nearly enough for Jensen Huang. Mobile-based solutions such as NVIDIA Xavier have been based on the shared memory design for quite some time. While there are challenges particularly on cache coherency model for such design, a CPU-GPU shared memory design would alleviate many pains for software developers to unlock the computation power of such a machine. The recent deep-learning models have been hinting a future of at least two orders of magnitude more parameters. To accommodate these new trends, a bigger departure from current PCIe based interconnect and form factor for accelerated computing co-processor is needed badly.
One thing interesting about Jensen is that for the past two decades, he always has the right insight. However, his endeavors were expensive for the company to pursue, with an often stretched credit line.
Apple has far more cash reserves for such an expensive purchase. But an ARM purchase would be a defensive move. Apple’s roadmap, while heavily ARM-based, doesn’t require radical departure from what ARM’s currently doing. The cost of inaction to Apple, would be a new adversary owner of ARM. From the past track record, NVIDIA's ARM would likely not prioritize any Apple’s request.
Apple had made defensive purchases in the past, and has the habit of hoarding good technology companies for many years until its technology has matured. Apple’s ARM purchase would be boring, but a strategically important one for them.
A Google’s ARM would be a distracted one. There are many product lines within Google that would benefit from an ARM purchase, from its mobile phone, Chromebook to its data center. The distraction and the likely politics interplay will likely terminate these endeavors prematurely. This will be an OK turnout for its competitors such as Apple. A Google’s ARM won’t be anywhere near as adversarial to them as NVIDIA's purchase.
A good asset like ARM would have many potential buyers. Many of them I didn’t cover here but equally likely. To me, as an interested observer, the most exciting one would be NVIDIA by far.