Few technologies have generated such a full-throttle frenzy of high-stakes investment as AI. Tech giants are going all-in as they race to create artificial general intelligence (or to reach major checkpoints on the way there). Meanwhile, companies in other sectors of the economy are integrating AI in their systems as fast as they can, to improve efficiency, optimize customer interactions, and attract interest from investors.
Everyone, it seems, is suddenly in a panic about how to lead in this highly promising new tech space — or at the very least, how not to get left behind.
But there is one group of companies that seemingly has nothing to fear, and indeed has already won. The manufacturers of purpose-built AI semiconductor chips will benefit hugely from the ongoing AI competition, regardless of who comes out ahead. All the key technology players, at least for the time being, will need those chips to fuel their efforts. Chip manufacturers therefore need only produce a consistent supply of semiconductor chips, each iteration a bit cheaper or more efficient than the last, and their market for the foreseeable future is assured.
But what’s so special about these chips, compared to all the other computer chips made before them? Why are they still so scarce, despite unprecedented demand? And what effect will the AI competition have on the global economy moving forward?
How AI chips are reshaping the tech industry
Many people may think that a ‘computer program’ is written line by line by coders sitting at rows of desks typing on their computers. Some still are, but more sophisticated programs use much more technical and advanced methods than they used to. Modern AI undergoes rapid training through the analysis of huge data sets. From there, pattern detection tools distill the lessons they learn into neat data packages, which are then encoded into algorithms that guide future behavior. Each step in this process requires very specific types of calculations and data transfers, which are facilitated by a specially designed AI chip.
Standard computer chips are made to be versatile, because the user might require a wide variety of tasks to be completed. AI chips are streamlined to perform the specific tasks which are native to the world of AI training.
The difference is night and day. According to a 2020 paper on the topic, highly specialized chips are “essential for cost-effectively implementing AI at scale; trying to deliver the same AI application using older AI chips or general-purpose chips can cost tens to thousands of times more.” The technology has only improved since then, with new design upgrades continuing to appear regularly.
The rise of self-driving cars, image generators, and LLMs like ChatGPT, not to mention software for improving business operations internally, has caused a demand spike for AI chips on a scale that has few parallels in history. Valued at a respectable $23 billion a year ago, the market for AI semiconductor chips is expected to leap to $150 billion by the end of the decade.
Global competition
At present, the leading quality, highest-volume AI chips are produced in Taiwan, South Korea, and the US. Chip fabrication factories (“fabs”) in these countries are significantly ahead of their Chinese counterparts, and the US government has set up targeted trade barriers in hopes of keeping it that way for as long as possible.
TMSC in Taiwan, and Nvidia in the US, are currently selling advanced chips as fast as their fabs can make them, and have an essentially guaranteed customer base for the foreseeable future, while other companies like Intel, Apple, AMD, and Huawei are investing heavily in an effort to keep up.
But the design and large-scale manufacture of these chips is an extremely challenging task. A modern fab, for example, costs $10-20 billion or more, due to the advanced machinery necessary to build such high-density chips. And each fab requires a very highly skilled workforce to maintain quality output.
An accelerant for future growth
Among chip manufacturers, AI is vital not only to their sales departments, but also to their design and production teams. Such is the power of this technology that it can even help manufacturers design and build better versions of itself, and do so more efficiently than human ingenuity alone can accomplish. Presumably, these better chips will lead to even more advanced AI in the future, which in turn can produce even greater innovations on the manufacturing floor, and so on.
Wherever our own minds (and energy) reach their limits, AI can take over to reach even greater heights. Although the chips that serve as a foundation for AI are highly specialized, AI itself is not. Its value as a supertool extends into areas as disparate as health care, finance, architecture, creative writing, and anything else we value as a society. A fortune therefore awaits any business that comes close to perfecting it, as well as any company that produces chips for all the companies that will try.
At present, the manufacturing process for those chips is so difficult, and the demand so huge, that companies like Nvidia and TSMC can find eager customers around every corner. Whether the efforts of less experienced AI chip manufacturers will be as successful is anyone’s guess, but any questions surrounding the utility of AI technology have already been definitively answered. AI is here to stay, and we will all soon be using it routinely in every aspect of our lives.
