It is a thrilling time to be a semiconductor company. Uses for chips keep proliferating, from connected cars to connected home speakers, and at the same time, the process of making chips gets harder and harder, which makes the things more precious.
That should be music to the ears of holders of shares of two of the best companies in the business, Nvidia and Advanced Micro Devices.
As the only two freestanding developers of computer chips that can process today's mania for artificial intelligence, the graphics processing unit, or GPU, chip, these two are in the catbird seat. There is endless demand for the things, as more and more problems in computing are re-fashioned as GPU problems.
It's been a rough fall for both stocks just like the rest of the market. Nvidia is down about 22% this year, and 45% in just the last three months. Advanced Micro Devices, which struggled early in the year, managed to double in the past 12 months, but its shares were cut by a third in that same three-month period leading up to today.
Both stocks are pricey, but a lot less than they were not long ago. Nvidia is the more reasonable bet at this point, trading at 23 times next year's estimated $6.59 per share in net income, versus 31 times for AMD.
The trends that define these two companies are deeper than either current stock market uncertainties or the other problem that constantly dogs their shares, the fact that they make money off of the precarious fluctuations of the crypto-currency market.
AI, the problem of how to compute problems via manipulations of vectors, has proven a boon to the GPU business. Nvidia owns what some estimate is about 80% of the market to "train" neural networks.
That process involves large numbers of parallel computing operations that multiply vectors and matrices together to find a hidden organization of connections between data points. It could be the features of a picture that collectively represent an object, as in computer vision. It could be the path through a series of steps to solve the strategy game Go.
All these treasure hunts have demanded heaps and heaps of Nvidia GPUs over the last several years. Revenue at Nvidia is expected to rise 22% this fiscal year ending in January, up from a 25% increase the year prior, a sharp increase from the outright declines and sluggish growth in many years leading up to 2017. Nvidia is on a roll.
AMD has nowhere near the same prominence in AI that Nvidia has, but as the only other major vendor of a GPU chip, that simply spells opportunity for the company to be the second source. Last month, AMD announced what it bills as the world's first GPU chips made at the smallest feature size available for chips today, seven billionths of a meter.
AMD is a fraction of Nvidia's size, at $20 billion in market capitalization versus $91 billion. The fact that it is little AMD making this announcement is a momentous come-from-behind achievement for a company near death four years ago.
The big threat to both is that the proliferation of AI keeps leading to more startup companies that could totally change the computing landscape and obsolete the GPU. Graphcore, a Bristol, U.K. startup company that has a fascinating alternative to the GPU for AI computing, yesterday announced it raised an additional $200 million from venture capitalists, with this round including contributions from Microsoft and from BMW. With $300 million raised to date, Graphcore has real money, not play money, that ensures its chips will get a shot at taking on the GPU.
There are dozens of such startups. And the lesson of Intel should give pause. Intel spent $14.5 billion last year to acquire Mobileye, a promising competitor in the race to use AI for self-driving cars. With the numerous startups emerging for AI, it's not clear that such a large purchase will ever pay for itself. It's hard, in other words, for the established chip vendors to keep pace with the change that is being prompted by AI.
Just as the Intel microprocessor was eclipsed in recent years by alternatives such as ARM Holdings chips, which now dominate mobile computing, and by the GPU, which dominates AI training, the GPU could also find itself drowning in alternatives.
In times of epic change, it's good both Nvidia and AMD have leadership that is the best in the business. Jensen Huang of Nvidia is a visionary who has been guiding the company to this moment for decades. Lisa Su is one of the most canny in the industry, and brought AMD back from the brink of disaster to fighting form.
Astute management may help both firms to continue to reap the rewards of a raging market for semis as AI proliferates.