State of Play: Chips / DAVOS 2025

Summary

This panel discussion focused on the future of semiconductor chips and their role in powering emerging technologies, particularly artificial intelligence (AI). The panelists, including industry leaders and a UN representative, explored the challenges and opportunities in chip development and manufacturing.

A key theme was the increasing demand for advanced chips to support AI and other computational needs. The speakers highlighted the importance of balancing performance with energy efficiency, as power consumption is becoming a major constraint. They discussed the trend of more companies entering chip design while fewer can manufacture them due to high costs and technical complexity.

The panelists emphasized the need for innovation in chip architecture and manufacturing to address power and performance challenges. They also stressed the importance of industry collaboration and interoperability standards to unlock the full potential of AI and connected devices.

The discussion touched on geopolitical aspects, including concerns about supply chain concentration and national efforts to boost domestic chip production. However, the panelists agreed that technology challenges remain the top concern for the industry.

The speakers explored the transformative impact of chips on various industries, from healthcare to hospitality, and the increasing importance of data utilization. They also discussed the need for better institutional capacity to address the rapid pace of technological change.

Overall, the panel highlighted the critical role of semiconductor chips in shaping future technologies and economies, while emphasizing the need for collaborative efforts to overcome technical, environmental, and geopolitical challenges in the industry.

Keypoints

Major discussion points:

– The future of chips and semiconductors, including AI-driven demand and innovation

– Challenges around power consumption, energy efficiency, and sustainability as chip demand grows

– The need for industry collaboration and interoperability in a hyperconnected world

– Balancing performance, privacy, and protocols in chip design and implementation

– Manufacturing capacity constraints and the concentration of advanced chip production

Overall purpose:

The goal of this panel discussion was to explore how the semiconductor industry will evolve to meet growing demand for faster, smarter, and more efficient chips, especially for AI and other emerging technologies. The panelists aimed to highlight key challenges and opportunities in chip design, manufacturing, and implementation.

Tone:

The tone was primarily informative and analytical, with panelists offering expert insights into industry trends and technical challenges. There was a sense of cautious optimism about technological progress, balanced with concern about energy and manufacturing constraints. The tone became slightly more urgent when discussing power consumption issues, but overall remained measured and professional throughout.

Speakers

– Yin Fan: Moderator from Yicai Media Group

– Rodrigo Liang: Co-founder and CEO of Simba Nova

– Amandeep Singh Gill: United Nations Secretary General’s Envoy on Technology

– Christina Kosmowski: CEO of Logic Monitor

Additional speakers:

– Armandeep Singh Gill: United Nations Secretary General’s Envoy on Technology (likely same person as Amandeep, with name variation)

– Mr. Zhang: Audience member who asked a question

– Ed: Audience member mentioned but did not speak

The Future of Semiconductor Chips: Powering AI and Emerging Technologies

This panel discussion, moderated by Yin Fan from Yicai Media Group, brought together industry leaders and a UN representative to explore the future of semiconductor chips and their role in powering emerging technologies, particularly artificial intelligence (AI). The panellists included Rodrigo Liang, Co-founder and CEO of SambaNova; Amandeep Singh Gill, United Nations Secretary General’s Envoy on Technology; and Christina Kosmowski, CEO of Logic Monitor.

Key Themes and Discussions

1. AI-Driven Demand and Innovation

The panellists agreed that AI will be a major driver of future chip demand and applications. Rodrigo Liang emphasised AI’s central role, while Amandeep Singh Gill and Christina Kosmowski highlighted the broader impact of chips in enabling hyper-connected environments across industries and powering connected devices in everyday life.

Gill expressed excitement about emerging technologies, including neuromorphic chips for new ways of training AI models and quantum chips for more powerful computing. This broadened the scope of the discussion to include potential future innovations beyond current AI chip architectures.

2. Power Consumption and Energy Efficiency

A critical challenge highlighted by all speakers was the increasing power consumption of advanced chips, particularly for AI applications. Rodrigo Liang emphasised this point, stating, “We’re running out of power. That’s a worldwide issue. If you’re going to scale AI, if you’re going to make AI a proliferated technology that everybody has access to, we got to solve the power issue.”

The panellists agreed on the need to balance performance demands with energy constraints, recognising that efficiency gains are often offset by rapidly increasing demand. This discussion underscored the importance of developing more power-efficient chip architectures and innovative solutions to support the growing demand for AI computation.

3. Manufacturing Challenges and Industry Collaboration

The panel explored the evolving landscape of chip design and manufacturing. Amandeep Singh Gill noted that while chip design is becoming easier and more democratized, manufacturing is growing more difficult due to high costs and technical complexity. This has led to fewer players being able to manufacture advanced chips, as pointed out by Rodrigo Liang.

All speakers emphasised the importance of industry collaboration in chip innovation. Christina Kosmowski highlighted the need for cooperation, while Gill advocated for a diverse innovation ecosystem in chip design and manufacturing. The discussion touched on the role of government incentives, with Gill suggesting that while important, they have limited impact in shifting the manufacturing landscape.

4. Balance Between Mature and Advanced Chips

Rodrigo Liang emphasized the importance of balancing the production of mature and advanced chips. He noted that while advanced chips grab headlines, mature chips still play a crucial role in many applications and industries. This balance is essential for meeting diverse market needs and ensuring a stable supply chain.

5. Observability and Monitoring

Christina Kosmowski highlighted the increasing importance of observability and monitoring in the context of chip performance and business impact. She emphasized how chips are integral to everyday experiences, from hotel check-ins to fast food ordering, and stressed the need for effective monitoring to ensure optimal performance and business outcomes.

6. Geopolitical and Policy Considerations

The panel addressed geopolitical aspects of the chip industry, including concerns about supply chain concentration and national efforts to boost domestic chip production. Amandeep Singh Gill stressed the need for collaborative leadership and digital cooperation globally, while also acknowledging the importance of supply chain security and national investments.

Rodrigo Liang argued that technology concerns remain the top priority over industrial policy and national rivalry, highlighting the industry’s focus on overcoming technical challenges. The discussion also touched on the three main concerns in the chip-making industry: industrial policy, national rivalry, and technology.

7. Institutional Capacity and Governance

Amandeep Singh Gill raised a thought-provoking point about the gap between technological advancement and institutional readiness. He emphasized the need for international cooperation and updated regulatory frameworks to keep pace with rapid technological changes.

8. Future Trends and Applications

The panellists explored the transformative impact of chips on various industries. Christina Kosmowski emphasised the increasing importance of data utilisation in our deeply connected world. Rodrigo Liang provided insight into potential market shifts, noting that inferencing is likely to become a significantly larger use case than training for AI chips.

9. Access to Advanced Manufacturing Capacity

In response to an audience question, Rodrigo Liang discussed the challenges of accessing advanced manufacturing capacity. He highlighted the complexity and cost involved in advanced chip manufacturing, explaining why only a few companies can produce cutting-edge chips.

Conclusion

The panel discussion underscored the critical role of semiconductor chips in shaping future technologies and economies. It highlighted the need for collaborative efforts to overcome technical, environmental, and geopolitical challenges in the industry. Key takeaways include the need to focus on developing more power-efficient chip architectures, increasing global collaboration in chip innovation, and updating institutional and policy frameworks to better handle the rapid pace of AI and chip technology advancement.

Unresolved issues remain, such as effectively balancing the concentration of advanced chip manufacturing with supply chain security needs, and closing the gap between efficiency improvements and rapidly increasing demand. The discussion pointed towards potential compromises, including balancing national investments in chip manufacturing with acceptance of some manufacturing concentration, and focusing on both improving chip efficiency and managing demand growth to address energy constraints.

As the semiconductor industry continues to evolve, addressing these challenges and opportunities will be crucial for unlocking the full potential of AI and connected devices in our increasingly interconnected world. As Yin Fan concluded, it’s important to ensure that chips do not become a flashpoint in a 21st century Cold War, emphasizing the need for global cooperation in this critical technology sector.

Yin Fan: Hello everyone, my name is Yifan from Yicai Media Group, the financial arm of Shanghai Media Group. I’m very delighted to moderate today’s discussion. So first of all, let’s give a warm welcome to our all-star panelists today, Armandeep Singh Gill, United Nations Secretary General’s Envoy on Technology, Kristina Kosmowski, CEO of Logic Monitor, and Rodrigo Liang, co-founder and CEO of Simba Nova. Thank you all. And we have, today we have one distinguished UN officials, two CEOs from American chief companies, and of course, in true devil’s fashion, the organizer has made sure we have a Chinese moderator for the chief-focused discussion. How fitting. So this panel is now being live-streamed. A warm welcome to our online audience. If you are sharing about us on social media, feel free to use the hashtag WEF25. Now let’s get started. Today’s topic is chips. You know, when chips were invented in 1958, the first significant market for them was inside nuclear missiles. But today, about one trillion chips are made every year. Ever more devices and machines contain ever more chips, and the new types of computations are booming, including AI and data crunching. Demand will soar further as more industrial machines are connected and fitted with sensors. Today’s discussion will focus on how industries will evolve to meet the escalating demand for technologies that are not only faster, smarter, but also productive, secure and infinitely scalable. So I want to start with big, big pictures. First question for three of you, what does the future chips, what does the future chips are powering look like?

Amandeep Singh Gill: The opportunity is huge, the use cases are proliferating and you mentioned AI. So there’s just the scale of the demand that has to be met, so it’s a short term challenge for the industry and then there’s innovation in terms of what kind of chips are going into these devices, these new data centers, these new AI factories. I’m very excited personally about neuromorphic chips, so new ways of training AI models using architectures that have not been successfully used earlier. And then there is the whole field of quantum, where more powerful quantum chips are on their way. We are not at a place where you have significant commercial application yet, but in the next few years that could change, so huge, huge opportunity across many areas. Rodrigo, what’s the future look like?

Rodrigo Liang: Well, I think it’s going to be AI driven first and foremost, I think to the extent that we’ve seen decades where traditional CPU computing has been dominant, I think we’re seeing a rise where artificial intelligence will be at the center of the semiconductor landscape going forward. I think it’s going to be about power, it’s going to be about efficiency, actually the deploying these chips across a broad range of applications from your data center all the way to the edge. And so you’re gonna be very focused on making sure you deliver the computational capacity at an efficient, power efficient way. AI-driven, do we have enough chips supply to the AI-driven world today? Well, I mean, we’ve lived this last couple of years. There’s seemingly an insatiable demand. And here’s an interesting thing. The supply issues that we’ve experienced over the last two, three years, that was all for training models. And for the most of the world, we believe that inferencing is gonna be a significant larger use case than training. And so I think we’ve gotta be prepared for the fact that you’re gonna need a lot more chips as the world turns on inferencing, they turn on the use of AI, not just to train the models, but to leverage those models and create workflows. And you’re gonna engage an 8 billion person population around the planet using the technology. So I think we’ve gotta be prepared to be efficient.

Yin Fan: Christina, what’s the future look like?

Christina Kosmowski: Yeah, definitely, I’ll echo the points already said. But I think first and foremost, we’re in a deeply connected world. And so if you think about every experience you have in your life, from your healthcare to your entertainment, and so this connection of bringing the physical and digital world together just is becoming more and more important. And so the ability to not only power these data centers for AI, but also the MRI machine or CAT scan machine as you go into your doctor or your surgery centers, or when you’re ordering French fries from your local fast food restaurant and all of that. And so this connected world is really important. And then of course, AI and the edge compute that we’re just talking about is critical, is that we’re able to have the visibility into the performance of these data centers that are fueling AI, and that we’ve got the visibility into it because we don’t have infinite resources. And there are impacts on cost and sustainability that need to be taken in account. as well, and having that visibility,

Yin Fan: having that ability to be efficient across all of that is important. Yes, Christina, very interestingly, other than AI, you also mentioned some daily life. So would you please give an overview of your company and what are you doing, and what are some overlooked ways that chips are already transforming our everyday interactions? Yeah, absolutely. Well, so Logic Monitor,

Christina Kosmowski: we’re a hybrid observability platform, so what does that mean? So we monitor both physical and digital services, and so typically that is in a data center, so we monitor the on-premise data centers or the cloud that’s powered by AWS or Azure, and then we can monitor anything with an IP address, and so as I was mentioning, you know, lots of different companies with these connected devices we also monitor. So the first thing we’re looking for is the performance. So as we’re relying more on data centers to fuel AI, which are fueling all of these workflows, or you’re relying on these physical connected devices in your everyday life, you need to ensure that they’re up and running, and so if they’re not up and running or there’s a performance degradation, we’re all in trouble, right? But then there’s all this innovation we’re producing, and without the visibility into the costs and the sustainability impacts, it’s gonna be very difficult to predict what the requirements are and to be sustainable. So even any organization doesn’t have an infinite amount of money to spend on this, and so we provide that visibility to say these workloads are best served in the cloud, these workloads are best served in a physical data center that you might own, how you balance against that and how you think about those components. So I do not want to, Rodrigo, say too much on your company, but you know, I asked the Chatterbot about Sembanova, and the Chatterbot told me Sembanova is one of the most promising AI chip startups. competing against NVIDIA, Google, and AMD in the AI hardware race.

Rodrigo Liang: Is that true? Yeah, well, it’s kind for them to say, but we’re, yeah, we build chips on, we say we’re a full-stack company, and what do I mean by that? On one end, we build chips that compete with NVIDIA. On the other end, we build a platform that allows you to leverage open-source models and customize them for your own private needs. And so, and we put it all together so it takes away this big, big gap in most companies and countries where they have to be part of the AI ecosystem, but they don’t have the talent. And so, so on the, yeah, on one end, we have chips, and we’re focused ultimately about performance and power. That’s what we’re doing there. We’re targeting 10X performance at 1X the power because for any given community, there’s gonna be a need, a need for being able to provide a certain capacity. If I can make it faster, I can reduce the number of servers and the reduced number of chips required. And then if I take every single one of those chips and drive the power down by 10X, that allows me to enable a lot of more use in places where perhaps you don’t have sufficient power to power it. And then on the other end, what we did with models is really focus on privacy and security. Many countries have this very important value around sovereignty, bringing AI that’s sovereign. I can control it. I wanna have it. And so we give a platform that allows you to then drive that sovereignty and control the models that you use. Yeah, Rodrigo, you said that you are a full stack of full stack of chip company. So can you briefly give an overview of the full stack of chip industry and what has changed? How is it changing? Yeah, I mean, the industry, you always start with the chips to run these models. And then you have an entire software stack that’s required in order to operate those models. and then you have the models themselves. And we’ve seen an incredible advance with open AI and companies like that building these models. And now you’re starting to see the layers above it, right? You’re starting to see people creating agents that are distilled from these models and just do a lot of different tasks. Now you have the operational folks really monitoring and then evaluating and being able to do these things. So it’s a very sophisticated stack that we’re all of us are contributing to, Simbanova being built up to a certain level. And then we partner with all the ecosystem to build these additional things that are required if you’re gonna be an AI first world, you’re gonna need all these pieces working together.

Yin Fan: So, Amandeep, Simbanova is just AI chips design company, is a startup, but now he’s competing with some established giants, AMD, Google, NVIDIA. So some people say that maybe this is a trend. Design chips is getting more easier, but manufacturing them is getting harder. Do you agree? Yes, indeed. But we need to kind of have a more diverse innovation ecosystem where startups and medium-sized companies are disrupting the incumbents with new ways of not just designing chips,

Amandeep Singh Gill: but also manufacturing them. And there are efforts underway in Japan, China, other places to do that. I think what we need, given the context, there is the market context, but there is also a geopolitical context, concerns about supply chains, concentration of manufacturing in a few places which are geopolitically sensitive export controls, just have had some developments on that recently. So I think that’s upping the stakes in terms of what a diverse innovation ecosystem needs. looks like, but I don’t think without such a system, we can really unlock that huge potential that we’ve spoken about earlier. So it’s important that in the next few years, we focus on the critical enablers of an AI economy. Compute, both on the training side and the inference side, is an essential part of those critical enablers. But there’s also data, there’s also the right talent. Globally, the talent is again concentrated in a few places, and we need to have more diverse talent coming together from fields such as agriculture, medicine, and other places. And that talent would bring more use cases into the AI economy. So that is the major focus for us at the UN in terms of what enables a more democratic, more sustainable development goals-oriented

Yin Fan: AI economy of the future. That’s the view from the United Nations. Rodrigo, if I ask you from the perspective of AI chips designer, do you agree, is that a trend that more and more companies are involved in, can involve in the chips design, but less and less company are able to manufacturing them?

Rodrigo Liang: I think that’s probably true. I do think that we’ve seen a revival of the semiconductor design industry, right? I mean, we saw that in the 90s, there were many, many chip design houses, and then those got consolidated over those years. And now, and then we went for probably 15 years without much investment in semiconductors globally. With AI, suddenly now you’re starting to see more and more people make those investments. It’s not a small investment. Semiconductor design is a large investment, but it’s still smaller than the manufacturing investment that’s required to build state-of-the-art fabs. And so I think just the economics of it is gonna make the number of players that are doing the manufacturing and the foundries of semiconductors to be fewer than the number of players that are playing in chip design. Is that a challenge in the future? Well, I think so. I think you’re gonna see few players that are gonna have extreme high capacity to drive that innovation because the demand’s gonna be there. Whether that continue to keep up with all the demand, we’ll see, right? But I do think that the processes that are required to actually design a lot of chips, and you have a lot of innovation that’s going into it to kind of make sure that from a technology side you can do it. And like we mentioned earlier, I think that there are a lot of new investments at a national level in different places where it’s trying to also bring alternatives that allow you to actually satisfy that demand.

Yin Fan: Christina, how do you think about cost optimization and the resilience for the development of future technologies?

Christina Kosmowski: Yeah, I think if you think about the supply chain and the stack, right? So you have these chips that are fueling these data centers, but then once you’re using them, you’ve gotta look at what are the cost implications? How can I be more efficient as I’m using them and balance those loads? So one of the things we really focus on is saying, hey, can, does this server need to be running 24 by seven? Or does this actual GPU need to be, do we need to collect that data at this point because it’s not maybe always as relevant to all of the different workloads and use cases that you’re using? And so you can start to do this edge computing where you could actually balance that and be smart about what you’re actually ingesting because everyone’s ingesting trillions of records nowaday and that volume’s going up. And so that’s not sustainable, right? So how we can really balance that. And then I think on this democratization of innovation, I think that’s really interesting too, is the so what, and so what is really important. which is, we’re building all of these data centers and chips, and for what? What is actually the business impact that it’s driving? And so being able to connect the performance of these chips, of these data centers, of this infrastructure, to your business outcomes, and being able to provide that visibility to strategic business executives is really, really critical. And so that’s another thing that we really focus on is connecting those dots to, hey, this is not only, how do we optimize this performance on cost and sustainability, but how do we connect that performance to your business and the outcomes and allow these IT people who were traditionally very technical, back office folks, are turning into being much more strategic business players now in their organizations

Yin Fan: than they’ve ever been. You mentioned innovation in this industry. So how much, in terms of the innovation, how much importance would you give to the industry collaboration?

Christina Kosmowski: I mean, it’s critical, right? As we’re all talking about different points in this chain around AI. And so it’s complicated, and we need to be able to solve this together. And the world, as I mentioned, is just getting more connected, not just from policies, but also physical devices and how we’re connecting into these digital worlds. And so it’s crucial that we’re all collaborating together.

Yin Fan: Rodrigo, you mentioned demand. Actually, 90%, actually over 90% of semiconductors manufacturing capacity works with material chips. But there are some computations that the only powerful chips can tackle. For example, if cars go electric and they need more advanced chips. So the question for the demand is, do you think, will the world in the future need more advanced chips? or just mature chips? Well, I think mature chips are always gonna be, by definition, the mature chips are gonna have more volume just because that’s kinda, you’re at a stage of that production.

Rodrigo Liang: I do think that the world’s gonna be very focused on the advanced chips because everything cascades from there, right? And so if you think about kinda the top level technology nodes that drive kinda the advances for semiconductor, that over time cascades into the maturity of much more efficient, much lower cost, much more economical devices. And so I think you’re gonna see a world that that’s gonna continue. It’s been around for, that’s been the trend for the last three decades. I think it’s gonna continue. And I think you’re gonna see from volume of chips, of course, you’re gonna see the maturity be down at the bottom as far as kinda things that you can pervasively deploy. From dollars, I still believe that the advanced nodes are gonna drive that because that’s where your highest performance is. That’s all your mission critical things are at. That’s gonna be the highest computational needs where it’s hard to get. And when you don’t have alternatives, then the cost drives and the cost goes up because that’s, you need it and there are very few players in that space. What’s your take on this question? We have the word, yeah.

Amandeep Singh Gill: I think it’s a three-dimensional issue. So there’s power and performance. Rodrigo talked about that. We also touched upon privacy. So a lot of people, if they are going to have a principal agent on their phone perform certain tasks that are sensitive, the computation has to be done in a privacy-preserving manner. So that’s going to be critical in terms of future demand. But there’s a third dimension. I think, Christina, you referenced that. And that is the connectedness of hardware devices. It’s the protocols, so the power and the performance, the privacy. preservation, and the protocols. It’s a little bit like the original vision for the Internet. You know, what kind of an AI innovation ecosystem we want, that’s equivalent of that vision of the Internet, where you add value, create value in various ways, but things are connected through common protocols. I worry that, you know, the hardware effort is going to land us in silos, where different bits of hardware don’t speak with each other, and we land up with standards. And the real power from general-purpose technologies like AI, as many economic historians have seen, is unleashed when you have those kind of standards, and there is horizontal diffusion of that technology in two more use cases. So I think as we go down the path of addressing the demand

Yin Fan: and innovating in the chip sector, we have to keep those three dimensions in mind. Amandeep, you mentioned the connectedness. And how do you envision the role of semiconductor chips in shaping hyper-connected environments in the next five to ten years?

Amandeep Singh Gill: Critical. You mentioned cars, for example. And, Christina, you mentioned everyday use, from toasters to fridges, but also medical devices. There’s huge potential in everyday use. It’s not all ending up in ballistic missiles. So there’ll be some critical high-end areas where there are national security considerations, other proprietary considerations, but most of it is going to be in everyday mundane stuff. So I think this is where chips, you mentioned the trillion level of production today, but it’s going to go into multiple trillions, and this is going to be ubiquitous. So we have to prepare for that future, the right use cases, the right protocols, the right kind of innovation ecosystems, so that everyone can thrive in the future AI economy. Christina, are there specific industries or applications?

Yin Fan: that you believe will drive next wave of innovation?

Christina Kosmowski: Yeah, I mean, first of all, I couldn’t agree more on this connectedness component and bringing this visibility together like in one single pane of glass is critical, right? Because there’s so much data coming and it’s all in all these different places and how do we start to unify that? And then how do we even reduce the noise of all of the data that’s coming in? And so I think there’s a wide range of industries as I was mentioning, a lot of people here were all checked into a hotel, right? And so if you check in and you go in and you’re looking for your reservation that was made on a global worldwide web, I came from San Francisco, I made my reservation in San Francisco and then I come here to Davos and I go and physically check in and I go to my room and I can’t check in. Where did that problem occur, right? Why can I not check in? Did it occur at the corporate global network? Did it occur in the actual key card? Did it occur at that local specific hotel instance? And so this is this interconnectivity that we’re talking about and being able to pinpoint like where those issues are coming from so that you can solve them because what’s the impact of that? I’m angry, I may not come back to that particular hotel again, I’ve got there’s a customer loyalty problem, they could lose revenue and so there’s real tangible business impacts. And then again, we were using the just going to order your food and there’s the order machine, there’s the actual payment processing, there’s the physical ice cream machine and the French fry fryer and all of those devices again are interconnected and they’re all, the protocols are all coming at different points and that really affects people’s core experiences

Yin Fan: and companies core businesses. In the hyperconnected world, do you worry anything? like privacy, security? Oh, I mean, of course. I mean, you can’t, without privacy and security,

Christina Kosmowski: I think you’ve got nothing, right? And so, it’s critical that that is first and foremost, as we’re building these connectivities, is ensuring that you’ve got a secure, trusted platform and that you’re getting the data at the source, I think is really important as well, is that it’s the data that you’re getting that context, you’re getting the data directly from these services or physical devices. Andrew, same question to you. Are there any specific industries or applications that you think will drive the next wave of innovation? Well, I mean, I think we touched on it.

Yin Fan: I think data is at the center of everything that we do.

Rodrigo Liang: And so, I think you’re gonna be trying to think about how to leverage it. Most of us here are sitting on a lot of data that we don’t know what to do with. Or we don’t know what it says, we don’t know what to do with it. And so, I think you’re gonna have an entire economy that’s gonna be created to try to leverage that data because the technology is available today for us to start building on top of that. And so, what would it look like if you’re able to unlock the true value of the data you already have and continue to actually leverage that to actually create services that are interconnected with other ecosystems? And so, what you know, what does your partner know, what do other players know, and can you exchange that in terms of services where just exchanging data, which we do today, right? In the internet, we already do today. We’re not getting the most value out of it. But what if I actually started exchanging in terms of services that are leveraged on that data? And you’re gonna get much more value, much more usable information that’s exchanged between parties, which I think is going to be something that is gonna continue to require more innovation.

Yin Fan: Same question, Amandeep.

Amandeep Singh Gill: I think the innovation is likely to be driven by one consideration around the cost of both training and inference. cost in terms of not just dollars, but also the energy and material consumption of what’s going into the economy. So that’s going to be a major driver. The second factor is going to be this issue of concentration versus distributed approaches. So sovereign approaches, more distributed approaches, that’s also going to drive a lot of the innovation. And finally, it is when this power inherent in AI meets the use cases in health, the green transition, agriculture, scientific discovery, where you will have specific requirements placed on the data sets, the quality, the diversity of those data sets, but also on the hardware that’s brought to bear on those data sets. So those considerations are going to drive the innovation in the chip sector in the coming years. You know, Amandeep said a very important thing,

Yin Fan: cost, and as you know, the question is to you, cost, he said that it’s not only means dollar, but also energy. So the question is, how can chips balance performance demands with energy constraints? I think that’s probably the most under-highlighted aspect

Rodrigo Liang: of this latest AI drive is we’re running out of power. That’s a worldwide issue. It’s, if you’re going to scale AI, if you’re going to make AI a proliferated technology that everybody has access to, we got to solve the power issue. And I think that that’s the stark reality that the technology is there, right? The technology is there to allow AI to be pervasive and commoditized for everybody to use, but you’re soon running into a problem where power is not available to everybody, right? People are out there building gigawatt data centers and nuclear power plants to support this, and yet not every country can do this, not every company can do this, and certainly most private individuals are going to struggle to continue to scale that. And so I think as a technology company, we’re very focused. on that, how do we actually continue on this journey that we’re all on and build better and better AI technologies that allow us to actually get the benefits, but you gotta do it in a way that’s scalable and power is the number one limiter and we’ll all have to get together to figure out how to do this and it means innovating at the core of the technology, build faster chips and lower power chips, all the way to the data centers and then the energy sources in order to actually create a much more efficient ecosystem that allows us to actually take the technology on this journey without breaking the grid that we have. Christina, same thing. Yeah, I was gonna say.

Yin Fan: The power thing.

Christina Kosmowski: I mean, power is obviously a huge issue and so I think, as you said, as you build up the stack, being able to really manage across that and if you think about how roles are evolving and the skills that are required for that, you have these IT folks now that are going to be these stewards of how do you balance those three factors and they’re not only strategically saying what’s the performance impact on our business, but also how do I support all this innovation in this cost effective way and so they’re going to look for tools like how can I balance what data I’m consuming because you don’t need all of this data all the time and so you need these smart kind of edge computing that can be intelligent around what data you’re collecting when and where you’re passing it to and I think that’s really important and I think the people now that are supporting that, their role is changing because they’re gonna be critical in helping think through that. Are they hitting a power wall, Rodrigo? I’m sorry? Are they hitting a power wall in terms

Rodrigo Liang: of the process of manufacturing more advanced chips? Are they hitting a power wall? Yeah, they’re hitting a power, I mean, I think you’re gonna find that. you can’t just build chips and brute force your way into it. You’re gonna have to think about better ways to actually create the technology that gives you the capabilities, but you’re not brute forcing. And so we’ve all lived kind of this, you know, the most recent set of issues around power. We’ve seen these chips, you know, you’re growing, you know, the population of chips that you’re deploying, but every single place you’re having to upgrade your, you know, your data centers in order to bring enough power in. And of course, on the other side, you know, being upgrade your cooling systems to get all that power back out, right? And so there are technologies today, alternatives, both data center on the edge, that are gonna be able to actually drive much, much lower power and still maintain a certain level of performance. In fact, probably a higher level of performance than you have with existing technologies, because you can’t, like you said, you can’t just focus on, can we drive brute force to the manufacturing to drive lower power? You’re gonna have to be significantly more innovative and bring new architectures, different ways of designing this so that you don’t put it only on the factories to find a way to actually squeeze just a little bit more power out of the system, because the level of improvement is not 5%, right? What we need to figure out is how do I get 5X, 10X out of the existing chips so you can continue on this journey that we’re on.

Yin Fan: Let’s skip the technology a little bit back to you, Amandeep. So in this hyper-connected world in the future, how should we think about leadership in this space?

Amandeep Singh Gill: I think collaborative leadership. So there is going to be competition, economic competition, I hope mostly, but some geopolitical competition, but we need to have collaborative leadership. That’s been our focus at the UN the past year when we had the Summit of the Future and some significant decisions on artificial intelligence for humanity, all of humanity, regular scientific assessments on the capability. and the evidence for the opportunities and the risks, a policy dialogue for interoperability, mutual learning, and a massive global effort on capacity building so that benefits can be shared, everyone can access these technologies. So collaboration, digital cooperation as we call it, especially connected innovation ecosystems. I think this is truly a general purpose technology. The hardware side is one thing, but there’s also data and talent and the use cases. So how do we bring all these things together in collaborative spaces around the world using market mechanisms, using the private sector startups in particular as a driver. And the sustainability piece, I think we would also need to, early days with AI science, the engineering has marched ahead, but the science is still catching up. So neuromorphic computing or quantum computing, there are ways in which the existing machine learning paradigm can be disrupted because you end up with trillion competitions for outputting just one token. That’s just the nature of the beast. So you have to throw brute force at it.

Yin Fan: But I think at some stage, some disruption is going to come and that would be helpful. Yes, before this panel, I talked to some industry experts. They shared with me that there are three concerns of chip making industry today, which is industrial policy, national rivalry, and technology. So the question for three of you, what are the top concerns among the three and how to rank the three? Let me start with Christina. Well, I’m squarely focused on the technology area. So I’m best served to speak there.

Christina Kosmowski: But I think it is around this interoperability you’re talking about. this hybrid world, right? And I define hybrid as physical and digital, and that’s just growing. I think by the end of 2026, 93% of all organizations will be in this hybrid environment. And so being able to bring all of that together so that we’ll have this unified view is really, really critical. And then being able to sort through all that data that’s coming from that source so that you can summarize that into something that’s actionable for you and your business or your use case that you’re looking at.

Yin Fan: So top concern is technology. Yes. Rigo.

Rodrigo Liang: Yeah, I think so too. I mean, I think the other ones are important as well. I do think that the technology is so disruptive that I think there’s still a lot that we’re trying to learn as far as how to integrate it into our daily workflows. And so you think about kind of the accuracy of these models. How do I bring agents in? How do I bring the security aspects of this in? I train my data into these models. How do I then have traceability of where my data that’s paid into these models, where did that go? And so just from a being able to take the technology and then give the facilities to people who are using it now in an integrated way and an accelerated way, people are doing this so fast. How do we solve all these things in a much more seamless way so that it allows us to actually do what we want to do with the technology and yet make it simple, maybe make it secure and safe, and then ultimately make it power and energy efficient

Yin Fan: so we can scale. Amandeep.

Amandeep Singh Gill: Top concern for me is that our institutional capacity is very limited compared with the challenge we are facing. The technology is moving very fast. Its implications are enormous. But our institutions, economic, social, political, international cooperation later in institutions. do not have the capacity to handle this tech challenge. So we really have to work hard over the next few years to build more international cooperation in a fragmented world, to update our systems. You know, the world of work is going to shift, for example. So what are our institutions, what are the policy instruments that we have? Competition policy, consumer protection, they really need to be updated super fast for the AI economy. So in terms of the discussion here, we, why don’t we stop here and hand the floor, hand the questions to the audience to see if there are any questions from the floor.

Yin Fan: Anyone? Mr. Zhang.

Audience: Curious about the access to advanced manufacturing. Manufacturing, you know, everybody’s competing for the three nano, four nano, coalesce capacity. Only one company can manufacture. You’re competing, small companies, startups, you’re competing with NVIDIA’s, AMD’s, and Microsoft’s, and Amazon, and Google’s, you know, they all want to gain that capacity. I’m just curious, do you find that you have a weak long line, or are you gaining some of this capacity at all, from TSMC?

Rodrigo Liang: We haven’t found that to be, I mean, we found our partners, you know, of course, we build our chips through TSMC as well, and we found our partners to be just incredibly helpful. I think the ecosystem knows that the technology industry is moving fast, and not only do you have to support incumbents, but you also have to support the new players, you know, because this technology is gonna continue to evolve, and so they’ve always been incredibly supportive of new companies, and new startups being able to play, and they have ways to handle that. So we personally have not run into that, and continue to emphasize that. I think supply chain problems. It’s been something that’s been around in semiconductors for a long, long time. And the sophistication of how you manage that, because there’s so many layers across that, is really important. To be able to know that it isn’t put a request in and a few weeks later, chips come out. It’s months, right? And so that ability to plan and plan in advance is incredibly important, especially for the smaller companies. Actually, we have already touched that in our discussion. More and more companies are involved in the chip design business, but less and less companies can manufacture it. So will that be a choke point for the future of this industry? Rodrigo. Well, I think this certainly is a risk. And I think you see countries finding ways to help fund some alternatives here as well. And so, I mean, we’re very focused on the advanced nodes because that’s kind of competitively where things are at today. But there is just a broad range of foundries that build semiconductors for a range of products where you have the memories and you have the mature technology nodes. And so I think you’re gonna have certainly an interest in having diversity there and making sure that you have alternatives. But I think it’s going to be continued to be a very expensive investment. And so just by that nature, you’re not gonna have many, many players can afford to be able to be in that game.

Yin Fan: Any more questions?

Audience: Yes. Yeah, just thank you so much. Just to build on the last point on the manufacturing, but you spoke about how sort of countries are trying to incentivize some of these investments, even though they’re sort of very high upfront CapEx. What according to you, can governments do even more sort of to incentivize some of these investments? Let’s talk about India, for example, we wanna sort of get supply chain onshore manufacturing, but it’s a big risk. So is there anything more that governments can do to incentivize the manufacturing part of? the value chain.

Yin Fan: Thank you. The question is to? Anyone. Yes, Amandeep. This is a question for you, I think.

Amandeep Singh Gill: Right, the government incentives on the manufacturing side. I think because of costs and quality and scale considerations, there is going to be a degree of concentration in fabs and manufacturing. It’s a fact of life in some ways. But I think for sovereignty considerations, supply chain related, supply chain security related considerations, there are going to be national investments. Probably not of the scale that’s needed to really shift the landscape, but that’s also a fact of life. Those investments are going to come. And I think the generative AI turn has woken up governments to the coming era where compute would be a critical enabler of the digital economy. So you’re likely to see more investments. And I think that’s good in the longer term. But I think I would place my bets more on the innovation front where you end up doing manufacturing differently. I mean, that’s happening on design for some years already. So that landscape has become more diverse and more kind of dynamic in terms of responding to what’s needed. It may take a while, but we need to get there with manufacturing as well. Ed? I think it was.

Yin Fan: Yes. One moment. Yes, one second.

Audience: Just maybe like a question for everyone. You look at when GPT-3 came out versus now, I would say like the power consumption of intelligence has gone down remarkably, right? That same model that was like 170 billion parameters, probably there’s like an 8 billion parameter model. that is equivalently as good. NVIDIA, every year they’re coming out with, every two years they’re coming out with a chip that is more power efficient. And I don’t know, if you look at this trend, it’s getting better and better every year. Out of curiosity, why do you think something needs to change? Why doesn’t this trend keep going down? I guess maybe it’s similar to Moore’s Law with compute. If you look at the 1990s, the power efficiency of a chip

Yin Fan: in the 1990s for compute versus now,

Audience: it’s materially better. Why do we think that’s not gonna happen for GPUs to go continued?

Yin Fan: Christina?

Christina Kosmowski: Well, I think it’s just the pace, like the pace of improvement versus the pace of what’s required, right? I just read a stat that nuclear energy is gonna be, is outpacing 6X, like the energy consumption that is being used right now by the world. And so I just think this pace of innovation and the pace of implementing all these use cases

Amandeep Singh Gill: is just so, so high. The numbers don’t add up. Because you have efficiency improvements, but the demand is soaring. So there’s a 10X difference, perhaps some more. And then there’s this paradox. There’s Moore’s Law, but there’s Jevon’s paradox. As coal efficiency improved in Britain in the 19th century, coal use just skyrocketed. So use is going up. And I think we really need to wrap our minds around the energy implications, despite the welcome efficiency gains.

Rodrigo Liang: Yeah, I’d say, yeah. If we look at the high level trends, right? And on a per rack basis, each rack used to be 20 kilowatts. Now an NVIDIA rack is 140 kilowatts and going up. And then the total number of racks worldwide is increasing because the demand is increasing. So if you just take the power per rack multiplied by the growth of the number of racks, this is why we’re looking around for more power sources. This is why there are gigawatt data centers. is trying to be built, and that’s just a fact of the demand and other things that are going on. And so yes, the models are getting smarter, they’re getting more efficient, and our expectation of the models also increase. And so we expect the models to do more, and so that’s why you see the next generation model be bigger and faster. And so you’re seeing those trends, and I think that’s gonna continue. I think technology is gonna get better, it’s gonna continue to get better, but our appetite for more that it does will also increase and probably outpace it.

Yin Fan: Thank you, Rijigo. Thank you, Amandeep. Thank you, Christina. So thank you all for coming to this panel. Finally, I want to say is I’m happy, after this discussion, in terms of the chipmaking industry, we have some concerns, industrial policies, national rivalries, but the top concern is absolutely technology concern. And as I said at the beginning, the first chips may have been used in missiles, but it would be wise to avoid them becoming a flashpoint in the 21st century Cold War. So thank you very much, thank you all for coming, thank you very much. Thank you. Thank you. Thank you. Thank you.

Agreement Points

AI will drive future chip demand and applications

speakers

– Rodrigo Liang
– Amandeep Singh Gill
– Christina Kosmowski

arguments

AI will drive future chip demand and applications

Chips will enable hyper-connected environments across industries

Chips will power connected devices in everyday life

summary

All speakers agree that AI and connected devices will be major drivers for future chip demand and applications across various industries and everyday life.

Power and energy constraints are significant challenges for chip technology

speakers

– Rodrigo Liang
– Amandeep Singh Gill
– Christina Kosmowski

arguments

Power availability is a major limiting factor for AI chip scaling

Efficiency gains offset by rapidly increasing demand

Need to balance performance demands with energy constraints

summary

All speakers emphasize the importance of addressing power and energy constraints in chip technology, highlighting the need for efficiency and balance between performance and energy consumption.

Similar Viewpoints

Both speakers highlight the challenges in chip manufacturing and the need for innovation in the industry, particularly in terms of diversifying the ecosystem and addressing high costs.

speakers

– Rodrigo Liang
– Amandeep Singh Gill

arguments

Fewer players able to manufacture advanced chips due to high costs

Need for diverse innovation ecosystem in chip design and manufacturing

Unexpected Consensus

Importance of industry collaboration and global cooperation

speakers

– Amandeep Singh Gill
– Christina Kosmowski

arguments

Need for collaborative leadership and digital cooperation globally

Importance of industry collaboration in chip innovation

explanation

Despite coming from different sectors (UN and private industry), both speakers strongly emphasize the need for collaboration and cooperation in the chip industry, which is unexpected given the competitive nature of the tech industry.

Overall Assessment

Summary

The main areas of agreement include the future role of AI in driving chip demand, the challenges of power and energy constraints, and the need for innovation and collaboration in the industry.

Consensus level

There is a high level of consensus among the speakers on the major challenges and future directions of the chip industry. This consensus implies a shared understanding of the industry’s needs and could potentially lead to more coordinated efforts in addressing these challenges across different sectors and stakeholders.

Different Viewpoints

Focus of future chip innovation

speakers

– Rodrigo Liang
– Amandeep Singh Gill

arguments

AI will drive future chip demand and applications

Chips will enable hyper-connected environments across industries

summary

While both speakers agree on the importance of future chip innovation, Rodrigo Liang emphasizes AI as the primary driver, whereas Amandeep Singh Gill focuses on the broader application of chips in hyper-connected environments across various industries.

Primary challenge in chip industry

speakers

– Rodrigo Liang
– Amandeep Singh Gill

arguments

Power availability is a major limiting factor for AI chip scaling

Institutional capacity limited compared to pace of chip technology advancement

summary

Rodrigo Liang identifies power availability as the main challenge for scaling AI chip technology, while Amandeep Singh Gill argues that the limited institutional capacity to handle rapid technological advancements is the primary concern.

Unexpected Differences

Role of government incentives in chip manufacturing

speakers

– Amandeep Singh Gill

arguments

Government incentives important but limited in shifting manufacturing landscape

explanation

Amandeep Singh Gill’s view on the limited impact of government incentives in shifting the manufacturing landscape is somewhat unexpected, given the general emphasis on national investments and policies in the chip industry. This perspective highlights the complexity of the issue and the potential limitations of policy interventions.

Overall Assessment

summary

The main areas of disagreement revolve around the primary drivers of future chip innovation, the most significant challenges facing the industry, and the best approaches to fostering innovation in chip manufacturing.

difference_level

The level of disagreement among the speakers is moderate. While there are differing perspectives on specific issues, there is a general consensus on the importance of chip technology and the need for innovation. These differences in viewpoints reflect the complexity of the chip industry and the various factors influencing its development, which could lead to a more comprehensive approach to addressing challenges and opportunities in the field.

Partial Agreements

All speakers agree on the need for innovation in chip manufacturing, but they differ in their approaches. Rodrigo Liang acknowledges the limitation of fewer players due to high costs, Amandeep Singh Gill advocates for a diverse innovation ecosystem, and Christina Kosmowski emphasizes the importance of industry collaboration.

speakers

– Rodrigo Liang
– Amandeep Singh Gill
– Christina Kosmowski

arguments

Fewer players able to manufacture advanced chips due to high costs

Need for diverse innovation ecosystem in chip design and manufacturing

Importance of industry collaboration in chip innovation

Similar Viewpoints

Both speakers highlight the challenges in chip manufacturing and the need for innovation in the industry, particularly in terms of diversifying the ecosystem and addressing high costs.

speakers

– Rodrigo Liang
– Amandeep Singh Gill

arguments

Fewer players able to manufacture advanced chips due to high costs

Need for diverse innovation ecosystem in chip design and manufacturing

Key Takeaways

AI will be a major driver of future chip demand and applications

Chips will enable hyper-connected environments across industries and everyday life

Power and energy constraints are a significant challenge for scaling AI chip technology

There is a need for more diverse and collaborative innovation ecosystems in chip design and manufacturing

Balancing performance, privacy, and protocols is crucial for future chip development

The pace of chip technology advancement is outpacing institutional capacity to manage its implications

Resolutions and Action Items

Focus on developing more power-efficient chip architectures to address energy constraints

Increase collaboration and digital cooperation globally in chip innovation

Update institutional and policy frameworks to better handle the rapid pace of AI and chip technology advancement

Unresolved Issues

How to effectively balance the concentration of advanced chip manufacturing with needs for supply chain security and national sovereignty

Specific ways governments can further incentivize domestic chip manufacturing investments

How to close the gap between efficiency improvements in chip technology and rapidly increasing demand

Suggested Compromises

Balancing national investments in chip manufacturing for sovereignty with acceptance that some concentration of manufacturing is inevitable due to costs and scale

Focusing on both improving chip efficiency and managing demand growth to address energy constraints

I’m very excited personally about neuromorphic chips, so new ways of training AI models using architectures that have not been successfully used earlier. And then there is the whole field of quantum, where more powerful quantum chips are on their way.

speaker

Amandeep Singh Gill

reason

This comment introduces cutting-edge technologies that could revolutionize AI chip development, going beyond current architectures.

impact

It broadened the scope of the discussion beyond current AI chips to include potential future innovations, setting the stage for a forward-looking conversation.

The supply issues that we’ve experienced over the last two, three years, that was all for training models. And for the most of the world, we believe that inferencing is gonna be a significant larger use case than training.

speaker

Rodrigo Liang

reason

This insight highlights a potential shift in the AI chip market from training to inference, which could have major implications for chip design and production.

impact

It shifted the focus to consider not just current demand but future trends in AI chip usage, prompting discussion on how to prepare for this shift.

We’re in a deeply connected world. And so if you think about every experience you have in your life, from your healthcare to your entertainment, and so this connection of bringing the physical and digital world together just is becoming more and more important.

speaker

Christina Kosmowski

reason

This comment emphasizes the pervasive nature of chip technology in everyday life, highlighting the broader implications of chip development.

impact

It expanded the conversation beyond technical aspects to consider the societal impact of chip technology, leading to discussion on various applications and industries.

I think that’s probably the most under-highlighted aspect of this latest AI drive is we’re running out of power. That’s a worldwide issue. It’s, if you’re going to scale AI, if you’re going to make AI a proliferated technology that everybody has access to, we got to solve the power issue.

speaker

Rodrigo Liang

reason

This comment brings attention to a critical challenge in AI chip development that had not been prominently discussed – the power consumption issue.

impact

It shifted the conversation to focus on sustainability and energy efficiency in chip design, leading to a discussion on balancing performance with power constraints.

Top concern for me is that our institutional capacity is very limited compared with the challenge we are facing. The technology is moving very fast. Its implications are enormous. But our institutions, economic, social, political, international cooperation later in institutions do not have the capacity to handle this tech challenge.

speaker

Amandeep Singh Gill

reason

This comment highlights the gap between technological advancement and institutional readiness, introducing a crucial policy perspective.

impact

It broadened the discussion from technical aspects to include policy and governance considerations, emphasizing the need for international cooperation and updated regulatory frameworks.

Overall Assessment

These key comments shaped the discussion by expanding its scope from purely technical aspects of chip development to include broader considerations such as future trends in AI usage, societal impacts, sustainability challenges, and policy implications. The conversation evolved from focusing on current chip technologies to exploring potential future innovations, market shifts, and the challenges of scaling AI technology globally. The comments also highlighted the interconnected nature of chip technology with various aspects of society and economy, emphasizing the need for a multifaceted approach to chip development that considers performance, power efficiency, and institutional readiness.

How can the chip industry balance performance demands with energy constraints?

speaker

Yin Fan

explanation

This is crucial as power consumption is becoming a major limiting factor in AI and chip development.

How can governments further incentivize investments in chip manufacturing, particularly in countries like India?

speaker

Audience member

explanation

This is important for developing more diverse and resilient global supply chains in chip manufacturing.

How can the industry develop more diverse talent globally to bring new use cases into the AI economy?

speaker

Amandeep Singh Gill

explanation

This is essential for democratizing AI development and ensuring it addresses a wide range of global needs.

How can the industry develop common protocols and standards to ensure interoperability in the growing AI and chip ecosystem?

speaker

Amandeep Singh Gill

explanation

This is critical for maximizing the value and horizontal diffusion of AI technology across various use cases.

How can the industry improve institutional capacity to handle the rapid pace of technological change in AI and chip development?

speaker

Amandeep Singh Gill

explanation

This is necessary to ensure appropriate governance and policy frameworks keep pace with technological advancements.

How can the industry develop more power-efficient chip architectures to support the growing demand for AI computation?

speaker

Rodrigo Liang

explanation

This is crucial for scaling AI technology globally, especially in areas with limited power infrastructure.

How can the industry better leverage existing data to create more value and interconnected services?

speaker

Rodrigo Liang

explanation

This represents a significant opportunity for innovation and value creation in the AI-driven economy.