Artificial intelligence is booming.
Tools like ChatGPT are improving at an alarming rate as companies race to plug into new areas of the economy.
However, the surge in demand for AI computing power faces major constraints. A graphics processing unit (GPU) required to train and deploy these models.
Professor of History at Tufts University and author of Chip War: The Fight for the World’s Most Critical Technology.
Marketplace’s Megan McCarty Carino told Miller about the potential shortage of chips to power the AI boom.
Below is an edited transcript of their conversation.
Chris Miller: The entire chip industry is facing a slowdown. The economy is slowing, people are buying fewer smartphones, and businesses are spending less money on updating their data centers. However, there is a real boom in certain types of chips used for AI, and some shortages are already visible. And the demand for these types of chips seems to only grow.
Megan McCarty Carino: And what about risks? You know, I’ve seen many other cases of technology where the supply chain for this is very concentrated.
mirror: Well, there are certainly risks. But in the short term, there aren’t many other options for most companies. The number of GPU designers and manufacturers is fairly limited. Also, not all GPUs are the same. They have different technical specifications, so if you’re used to one type, it’s not always easy to switch to another.
McCarty Carino: what about CHIPS [and Science] Act, will these hundreds of billions of dollars increase domestic chip development and production? Will it help here?
mirror: Most of the CHIPS Act funding is not for a specific type of chip. I think it’s likely that CHIPS Act funding will be deployed in facilities that can manufacture GPUs and other types of chips used in generative AI, but I don’t think we’ll see any government or other agency either. Companies applying for funding are inevitably focused on generative AI as their primary use case. Because in reality, we need different types of chips. And this is one example that deserves attention. But across the economy, there are varying demands for increased chip manufacturing capacity.
McCarty Carino: What are the challenges in transitioning to more domestic manufacturing of these types of semiconductors?
mirror: The kinds of chips used to train AI systems, including these advanced GPUs, are generally manufactured using existing state-of-the-art manufacturing processes. And only a few companies in the world have the necessary capabilities to manufacture these chips.We are looking at just a few companies [specifically] their decisions about where to build new facilities; And they are basing their decisions partly on cost dynamics, looking at incentives, taxation and workforce dynamics. Can they hire all the workers they need? [South] South Korea is looking not only for talent, but also for support from the government. As such, the United States must work hard to attract domestic semiconductor manufacturing investment.
McCarty Carino: Recently, there has also been a move in line with the CHIPS Act in Europe to step up production of this type of technology. Basically, what do you see as the big challenges to making this a sustainable transition for countries like the US and Europe?
mirror: Well, I think there are different dynamics in America, Europe and Japan. For example, the United States has traditionally been a leader in many types of chip manufacturing technology. In addition to designing advanced chips like GPUs, the United States also has a significant amount of advanced chip manufacturing production, which has been around for some time. Europe and Japan are not very advanced in chip manufacturing today. That makes it difficult for these countries to try to scale at the absolute leading edge. And I think that’s why we’re already seeing Japan, for example, focus on chips that are a little behind, rather than cutting-edge chips. And when various European nations start drawing up plans for the chip industry, they won’t end up focusing on slightly less advanced chips with broader applications than the cutting edge. I think.
McCarty Carino: Overall, how do you see the AI boom impacting domestic chip manufacturing?
mirror: Well, it’s ultimately good news for the semiconductor industry as a whole. Demand for semiconductors will increase. Computing power is at the forefront and center of the transition in the model of how many companies plan to develop new products. And this shows continued US strength, as US companies designed the chips that underlie many of these AI systems. It is also a US company that designs the AI system itself. In short, this is a success story for the US tech industry.
McCarty Carino: What are you looking at when this money spreads around the world and you start to see some guidelines coming out of the Department of Commerce on how this money will be spent?
mirror: I think the key question going forward will be how businesses will respond now that they know how the government plans to use much of the CHIPS funds. And how will the chip maker’s customers react? So the main factors going forward are Nvidia, Apple, [Advanced Micro Devices], where would they want to buy semiconductors from? America? From Taiwan? From South Korea? From another region? And while that’s what we’ve seen some early indications from the company, it’s still unclear at this point how it plans to source the chips it needs over the next few years.
On our partner show Marketplace, Kai Ryssdal interviewed Commerce Secretary Gina Raimondo last week about how she views the deployment of the CHIPS Act Fund as a matter of national security, not just an economic policy. I talked about dolphins.
And last week, Nvidia announced a new innovation in the AI space — not hardware, but software called NeMo Guardrails. This could prevent chatbots from hallucinating or saying inappropriate things. The first large language model says:
But what about the second large language model? Who watches over the guards?
