
65 segments available
Dylan Patel runs Semianalysis, the leading publication and research firm on AI hardware: https://www.semianalysis.com/. Jon Y runs @Asianometry, the world’s best YouTube channel on semiconductors and business history. 𝐄𝐏𝐈𝐒𝐎𝐃𝐄 𝐋𝐈𝐍𝐊𝐒 * Transcript: https://www.dwarkeshpatel.com/p/dylan-jon * Apple Podcasts: https://podcasts.apple.com/us/podcast/dylan-patel-jon-asianometry-how-the-semiconductor/id1516093381?i=1000671564456 * Spotify: https://open.spotify.com/episode/6q1XODE2L5bqqBwe7434S7?si=seXQ6K_LQZeAV6776H6MhQ * Me on Twitter: https://twitter.com/dwarkesh_sp 𝐒𝐏𝐎𝐍𝐒𝐎𝐑𝐒 * Jane Street is looking to hire their next generation of leaders. Their deep learning team is looking for FPGA programmers, CUDA programmers, and ML researchers. To learn more about their full time roles, internship, tech podcast, and upcoming Kaggle competition, go here: https://jane-st.co/dwarkesh * Stripe builds financial infrastructure for the internet. Millions of companies from Anthropic to Amazon use Stripe to accept payments, automate financial processes and grow their revenue. Learn more here: https://stripe.com/ If you’re interested in advertising on the podcast: https://www.dwarkeshpatel.com/p/advertise 𝐓𝐈𝐌𝐄𝐒𝐓𝐀𝐌𝐏𝐒 00:00:00 – Xi’s path to AGI 00:05:05 – Liang Mong Song 00:09:10 – How semiconductors get better 00:12:01 – China can centralize compute 00:19:35 – Export controls & sanctions 00:33:36 – Huawei’s intense culture 00:39:36 – Why the semiconductor industry is so stratified 00:41:43 – N2 should not exist 00:46:38 – Taiwan invasion hypothetical 00:50:06 – Mind-boggling complexity of semiconductors 00:59:58 – Chip architecture design 01:05:21 – Architectures lead to different AI models? China vs. US 01:10:57 – Being head of compute at an AI lab 01:17:09 – Scaling costs and power demand 01:37:50 – Are we financing an AI bubble? 01:51:05 – Starting Asianometry and SemiAnalysis 02:06:55 – Opportunities in the semiconductor stack
In this opening segment, Dwarkesh introduces Dylan Patel and Jon Y, discussing their backgrounds and the significance of their work in the semiconductor industry. The light-hearted banter sets the stage for a deep dive into the complexities of AI hardware and semiconductor technology.
"Today, I'm chatting with Dylan Patel, who runs SemiAnalysis, and Jon, who runs the Asianometry YouTube channel. Does he have a last name? No, I do not. No, just kidding. Jon Y. Why is it only on..."
Dylan and Jon explore what actions Xi Jinping might take to advance China's AI capabilities. They discuss the importance of gathering intelligence and resources, drawing parallels to historical espionage efforts, and the implications for global AI competition.
"Alright here’s my first question. If you're Xi Jinping and you're scaling-pilled, what is it that you do? Don't answer that question, Jon, that's bad for AI safety. I would basically be contactin..."
The conversation shifts to the competitive landscape of semiconductor talent. Dylan highlights the poaching of skilled workers from TSMC to SMIC and Samsung, emphasizing the strategic importance of human capital in the semiconductor industry.
"Why doesn't that drive up their wages? It's because it's very compartmentalized. Back in the 2000s, before SMIC got big, it was actually much more open and more flat. After that, after Liang Mo..."
Dylan recounts the story of Liang Mong Song, a pivotal figure in the semiconductor industry known for his aggressive talent acquisition strategies. The discussion reveals the intense competition and the high stakes involved in semiconductor innovation.
"Very rapid. That guy's a genius. That guy's a genius. I don't even know what to say about him. He's like 78 and he's beyond brilliant. He does not care about people. What does research to make th..."
The segment delves into China's potential for centralizing compute resources to enhance AI capabilities. Dylan discusses the challenges and opportunities presented by China's infrastructure and the implications for global AI development.
"How do they get the 10 gigawatts data center up? What else do they need? There is a true question of how decentralized do you go versus centralized. In the US, as far as labs and such, you have..."
Dylan and Jon analyze China's energy capabilities and how they could support massive AI data centers. They discuss the implications of China's power generation capacity and the potential for rapid scaling in AI infrastructure.
"centers ready. China could just build it in six months, I think, around the Three Gorges Dam or many other places. They have the ability to do the substations. They have the power generation cap..."
In this segment, the discussion turns to the future of semiconductor manufacturing in China versus the US. Dylan outlines the challenges and advantages each country faces in maintaining a competitive edge in chip production and technology.
"extremely energy intensive refining and rare earth refining and all these manufacturing industries here. It would be very easy to hide it. It would be very easy to just shut down like… I think ..."
Dylan and Jon critically assess the effectiveness of export controls imposed on China regarding semiconductor technology. They discuss the implications of these restrictions on the global semiconductor landscape and the potential for China's domestic industry to thrive despite them.
"Jon, Dylan seems to think the export controls are kind of a failure. Do you agree with him? That is a very interesting question because I think it's like… Why thank you. Dwarkesh, you're so ..."
Dylan Patel discusses Huawei's remarkable ability to compete in the global market despite facing significant restrictions. He attributes Huawei's success to its engineering capabilities and cultural mindset, suggesting that the company's drive for excellence stems from a nationalistic perspective and a belief in the importance of semiconductor technology.
"then all of the sudden you aggregate all the flops. There's no fucking way. China can be centralized enough to compete with each individual US lab. They could have just as many flops in 2025 a..."
The segment delves into the future of AI development in China and the potential for centralized computing resources. Dylan Patel speculates on China's ability to leverage foreign chips and domestic manufacturing to create powerful AI models, raising questions about the competitive landscape in the coming years.
"they could have a bigger model than any of the labs next year. I have no clue where all the Ascend 910Bs are going, but there are rumors about them being divvied up between the majors, Alibaba, ..."
Dylan Patel critiques the U.S. sanctions on SMIC, arguing that they are poorly targeted and may not effectively hinder China's semiconductor advancements. He discusses the ongoing capabilities of SMIC and the implications for the global semiconductor supply chain, emphasizing the need for a more comprehensive approach to sanctions.
"want to do with these semiconductors. There's two parts there. The way the US has sanctioned SMIC is really stupid. They've sanctioned a specific spot rather than the entire company. SMIC is stil..."
The discussion turns to the challenges of yield in semiconductor manufacturing, particularly for Chinese firms. Dylan Patel explains how the complexity of chip production leads to lower yields and the impact of U.S. export controls on China's ability to produce cutting-edge chips.
"Why do they have bad yield? Because it's hard. Even if everyone knows the number, like say there’s a 1000 steps. Even if you're 98-99% for each, in the end you'll still get a 40% yield. If it's ..."
Dylan Patel explores the relationship between AI demand and the economic viability of advanced semiconductor nodes. He discusses how AI's growth could justify investments in leading-edge technology, questioning whether the semiconductor industry can sustain its pace of innovation without significant AI-driven demand.
"the US, but when you flip to like… Sorry, I don't fucking know what I was going to say. Nailed it! We're keeping this in That's fine, that's fine. Hey everybody. I'm super excited to introduce our ..."
The segment examines Huawei's dominance in the tech industry and its implications for competition within China. Dylan Patel discusses how Huawei's success may hinder the growth of other Chinese tech firms, raising questions about the overall health of the Chinese semiconductor ecosystem.
"And now back to Dylan and Jon, 2026 if they're centralized, they can have as big training runs as any one US company… Oh, the reason why I was bringing up Shanghai. They're building 7nm capacity i..."
Dylan Patel reflects on the cultural factors contributing to China's success in technology and manufacturing. He discusses the influence of the Chinese Communist Party and the nationalistic drive for technological advancement, suggesting that this mindset fuels innovation and competitiveness.
"Do you think the dominance of Huawei is actually bad for the rest of the Chinese tech industry? Huawei is so cracked that it's hard to say that. Huawei out-competes Western firms regularly with tw..."
The conversation touches on the role of espionage in China's technological advancements. Dylan Patel acknowledges the impact of stolen intellectual property while emphasizing that China's engineering capabilities are also a significant factor in its success.
"The military, because it's the PLA. It's generally seen as an arm of the PLA. How do you square that with the fact that sometimes the PLA seems to mess stuff up? Oh, like filling water in rocke..."
Dylan Patel discusses the paranoia within Huawei and its impact on the company's performance. He draws parallels to the mindset of successful Western firms, suggesting that Huawei's sense of urgency and fear of competition drives its innovation and excellence.
"That’s also the flip side. How much false propaganda is there? There's a lot of, "SMIC could never, they don't have the best tools." Then it's like, “Motherfucker, they just shipped 60 million p..."
The segment explores the concept of 'struggle' within Chinese culture and its influence on Huawei's operations. Dylan Patel argues that this mindset fosters a relentless pursuit of excellence and innovation, positioning Huawei as a formidable competitor in the global market.
"Another thing I'm curious about is where that culture comes from, but also how it stays there. With American firms or any other firm, you can have a company that's very good, but over time it g..."
Dylan Patel analyzes the current state of the semiconductor industry, highlighting its stratification and the reasons behind the decline of vertical integration. He discusses how competition and specialization have shaped the landscape, leading to fewer but more specialized players in the market.
"“By the evil western pigs.” “Capitalist…” Or not capitalist, they don't say that anymore. It's more like, “Everyone is against China. China is being defiled. That is all on you, bro. If you can’t ..."
The discussion shifts to the future of wafer manufacturing and the challenges faced by foundries. Dylan Patel emphasizes the need for significant investment and market consolidation to support the next generation of semiconductor technology.
"that’s better.” That's the beginning of what we call the semiconductor equipment industry. In the seventies, everyone made their own equipment. Sixties and seventies. All these people spin off. Wha..."
Dylan Patel questions the economic viability of advanced semiconductor nodes, particularly 2nm technology. He discusses the challenges of funding and the implications for the industry's future, raising concerns about the sustainability of Moore's Law.
"A bunch in Japan. A bunch in Japan. They all did this thing. When going to leading-edge, it got harder, which means you had to aggregate more demand from all the customers to fund the next node. ..."
The segment concludes with a discussion on the power efficiency gains from moving to smaller process nodes. Dylan Patel explains the diminishing returns of node advancements and the importance of data locality in improving overall chip performance.
"What’s the point? Why is 2 nanometer not justified? I'm not saying it for N2 specifically, but N2 as a concept. The next node should technically... There will come a point where economically, the..."
This segment delves into how AI is reshaping the semiconductor landscape, particularly in terms of power efficiency and performance. The speakers discuss the importance of new process nodes for AI applications, emphasizing the need for higher density and lower power usage. They highlight the significance of data locality in improving power efficiency and the competitive pressure on data centers to adopt the latest technologies.
"2nm just for one player, TSMC? Ignore Intel, ignore Samsung. Samsung is paying for it with memory, not with their actual profit. Intel is paying for it from their former CPU monopoly… Private equi..."
The conversation shifts to the potential global ramifications of a crisis in Taiwan, a critical hub for semiconductor manufacturing. The speakers outline the immediate and long-term impacts on the tech industry, including market crashes and supply chain disruptions. They emphasize the reliance of major tech companies on Taiwanese chips and the cascading effects on everyday products, from cars to household appliances.
"I want to ask a normie question… I won’t phrase it that way. Not for you nerds. I think Jon and I could communicate to the point where you even wouldn't know what we're talking about. Suppose Ta..."
In this segment, the speakers discuss the current state of semiconductor manufacturing, particularly the dominance of Taiwanese companies like TSMC. They explore the challenges of producing chips across various nodes and the implications for global supply chains. The conversation highlights the complexity of modern chips and the extensive reliance on semiconductor technology in everyday devices.
"Cars are like 40% chips now. There are chips in the tires. There's like 2,000+ chips in every car. Every Tesla door handle has like four chips in it. It’s like, “What the fuck?” Why? It’s like shi..."
This segment addresses the high level of specialization required in the semiconductor industry, contrasting it with the more accessible field of AI. The speakers discuss the barriers to entry for new talent in semiconductors and the extensive educational requirements. They reflect on the challenges of knowledge transfer within the industry and the implications for innovation and workforce development.
"no automobiles, no weed whackers, because that stuff has chips. My toothbrush has Bluetooth in it. Why? I don’t know. There are so many things that would just go poof. We'd have a tech reset. We w..."
The discussion focuses on the siloed nature of knowledge within the semiconductor industry, where information is often not shared across different layers of the manufacturing process. The speakers highlight the challenges this poses for innovation and collaboration, emphasizing the need for better communication and understanding among different sectors of the industry.
"possible. ArXiv is a free thing. The paper publishing industry is abhorrent everywhere else. You can't just download IEEE papers or SPIE papers or from other organizations. At least up until late..."
In this segment, the speakers explore the potential for breakthrough innovations in semiconductor technology, driven by advancements in AI and design methodologies. They discuss the importance of optimizing data movement and power consumption in chip design and the opportunities for significant efficiency gains. The conversation highlights the evolving landscape of semiconductor manufacturing and the role of AI in shaping its future.
"around something. Early on, we used to have SEMATECH where American companies came together and talked and hammered it out. But in reality it was dominated by a single company. Nowadays it's mor..."
This segment examines the architectural challenges and opportunities in chip design, particularly in relation to AI applications. The speakers discuss the importance of optimizing compute efficiency and the potential for significant gains through innovative design changes. They highlight the interplay between hardware capabilities and model architecture, emphasizing the need for a holistic approach to semiconductor development.
"Most of the equipment in semiconductor fabs runs on Windows XP. Each tool has a Windows XP server on it. All the chip design tools have CentOS version 6, which is old as hell. There are so many ..."
The conversation shifts to the complexity of semiconductor design, characterized by an enormous search space of possible configurations. The speakers discuss the implications of this complexity for innovation and the role of AI in navigating these challenges. They emphasize the need for collaboration between human designers and AI tools to optimize chip performance and efficiency.
"It's important to state that semiconductor manufacturing and design is the largest search space of any problem that humans do because it is the most complicated industry that humans do. When you..."
In this segment, the speakers analyze the trade-offs involved in developing AI models and semiconductor architectures. They discuss how hardware limitations influence model design and the strategic decisions companies must make to optimize performance. The conversation highlights the competitive landscape between American and Chinese companies in the semiconductor space.
"Over what time period? The question is how much can we advance the architecture. The other challenge is that the number of people designing chips has not necessarily grown in a long time. Company ..."
The final segment explores the differences in AI model development between the US and China, driven by varying hardware capabilities and investment strategies. The speakers discuss how these differences may lead to divergent architectural choices and performance characteristics in future AI applications. They emphasize the implications for global competition and technological advancement in the semiconductor industry.
"There are a few vectors to go here. One you mention is important to note. Hardware has a huge influence on the model architecture that's optimal. It's not a one-way street that better chip equal..."
Dylan Patel contrasts the capabilities of Chinese AI models with those from the West, particularly in video and image recognition. He notes that China's extensive surveillance infrastructure gives them a significant advantage in these areas. This segment explores how cultural and technological differences shape the development and application of AI technologies in different regions.
"investing hugely in. You go to conferences and there are like 20 papers from Chinese companies/universities about compute and memory. Because the FLOP limitation is here, maybe NVIDIA pumps up th..."
The discussion turns to the complexities of the global semiconductor supply chain and its impact on AI development. Dylan Patel highlights the intricate relationships between hardware design, architecture, and the evolving landscape of AI research. He emphasizes the challenges and opportunities presented by the current state of semiconductor technology and its implications for future advancements.
"You have this divergence in tech tree and people can start to design different architectures within the constraints they're given. Everyone has constraints, but the constraints different compani..."
Dylan Patel outlines the strategic considerations for leading a new AI lab, particularly in the context of US-Israeli firms. He discusses the importance of compute resources, data center locations, and the need for significant investment to compete with established players. This segment provides insights into the operational challenges faced by emerging AI labs in a competitive landscape.
"Jon and Dylan have talked a lot in this episode about how stupefyingly complex the global semiconductor supply chain is. The only thing in the world that approaches this level of complexity is th..."
In this segment, the conversation delves into the current state of the GPU market, highlighting the buyer's market conditions and pricing trends. Dylan Patel discusses the implications of these trends for AI research and development, including the cost of compute and the potential for increased accessibility to powerful hardware. This analysis sheds light on the evolving economics of AI infrastructure.
"Anyways, you can head to Stripe.com to learn more. If you were made head of compute of a new AI lab, if Ilya Sutskever’s new lab SSI came to you and they're like, "Dylan, we give you $1 billion. ..."
Dylan Patel discusses the future of AI compute costs, referencing Moore's Law and the decreasing costs of intelligence. He explains how advancements in compute efficiency and hardware design are driving down costs, making AI technologies more accessible. This segment highlights the potential for widespread adoption of AI as costs continue to decline.
"Yes, you're going to make compute efficiency wins, but with a billion dollars you probably just want the biggest cluster in one individual spot. Small amounts of GPUs are probably not possible to ..."
The discussion focuses on the challenges of scaling AI infrastructure, particularly in the context of large cluster sizes and power requirements. Dylan Patel outlines the complexities of building and operating data centers, including the need for efficient power management and the implications of regulatory constraints. This segment provides a comprehensive overview of the logistical hurdles faced by AI companies.
"Right. Anyway, if you were head of compute at SSI… Okay, I’m head of compute at SSI. There's obviously no free data center lunch, in terms of what we see in the data. There’s no free lunch if you..."
Dylan Patel emphasizes the critical role of power in the operation of AI data centers. He discusses the importance of efficient power delivery and the challenges associated with energy consumption in large-scale AI operations. This segment highlights the interplay between power infrastructure and AI performance, underscoring the need for innovative solutions in energy management.
"Now it makes the most sense to build your own cluster instead of renting, or get a very close relationship like OpenAI/Microsoft with CoreWeave or Oracle/Crusoe The next step is Bitcoin. OpenAI ..."
In this segment, the conversation shifts to global trends in AI data center development, with a focus on emerging markets. Dylan Patel discusses the significant investments being made in countries like Malaysia and the Middle East, as well as the strategic partnerships forming between tech companies and local governments. This analysis provides insights into the future landscape of AI infrastructure worldwide.
"but it's you have 14 mobile generators and you're just burning natural gas on site on these mobile generators that sit on trucks. Then you have power directly two miles down the road. There's no..."
Dylan Patel speculates on the future of AI training regimes and the potential for new methodologies that could enhance scalability. He discusses the implications of synthetic data and distributed training techniques for the evolution of AI models. This segment explores the innovative approaches that could shape the next generation of AI research and development.
"It's the ability to generate new power for these activities. That’s why it's really difficult, the economic regulation around that. But the real thing is if you look at the cost of ownership of a..."
The discussion continues with a focus on power generation and its economic implications for data centers. Dylan Patel highlights the potential of untapped energy resources in regions like Ethiopia and the challenges of establishing data centers in politically sensitive areas. This segment provides a nuanced view of the intersection between energy policy and AI infrastructure development.
"the ability to get the power. I don't want to turn it off eight hours a day. Let’s zoom out a bit. Let's discuss what would maybe happen if the training regime changes and if it doesn't change. Yo..."
Dylan Patel provides an overview of the global AI compute landscape, discussing the distribution of data center capacity across different regions. He emphasizes the competitive dynamics between the US, China, and emerging markets, highlighting the strategic investments being made to enhance AI capabilities. This segment offers a comprehensive look at the future of AI compute infrastructure.
"Are people bidding for that power? I think people just don't think they can build a data center in Ethiopia. Why not? I don't think the dam is filled yet, is it? No, the dam could generate that power..."
In this segment, Dylan Patel discusses projections for scaling AI clusters over the next few years. He outlines the anticipated growth in GPU clusters and the technological advancements that will enable this expansion. This analysis provides insights into the future capabilities of AI systems and the infrastructure required to support them.
"Middle East, and the rest of the world. Let’s go back to your point. You have synthetic data. You have the search stuff. You have all these post-training techniques. You have all these ways to s..."
Dylan Patel concludes the discussion with a forward-looking perspective on the future of AI infrastructure. He speculates on the potential for massive GPU clusters and the implications for AI research and development. This segment encapsulates the key themes of the conversation, emphasizing the transformative potential of AI technologies in the coming years.
"Microsoft/OpenAItheir partners for them. It’s potentially even more. 500K GB200s is like a gigawatt and that's online next year. The year after that, if you aggregate all the data center sites, a..."
In this segment, Patel elaborates on the future of GPU clusters, predicting that by 2026, we could see clusters with 300,000 to 500,000 GPUs. He discusses the challenges of connecting these clusters across multiple sites and the efficiency losses that may occur. The conversation touches on the rapid scaling of AI capabilities and the financial implications for companies investing in these technologies.
"Let's just give the number to like, “Okay, it’s 2025 and Elon's cluster is going to be the biggest…” It doesn’t matter who it is. There's the definition game. Elon claims he has the largest clust..."
Patel predicts a significant leap in compute power by 2028, potentially reaching 1e30 flops, which would represent a massive increase in capabilities compared to current models like GPT-4. He discusses the complexities of measuring compute power and the factors that contribute to this exponential growth, including advancements in AI training methodologies and the integration of synthetic data.
"No. Sam has a superpower. It's recruiting and raising money. That's what he's like a god at. Will chips themselves be a bottleneck to the scaling? Not in the near term. It's more about concentratio..."
This segment focuses on the financial aspects of scaling AI infrastructure, with Patel discussing the potential costs associated with achieving 1e30 flops. He emphasizes the need for substantial investment in semiconductor manufacturing and the importance of TSMC's aggressive plans for new fabrication technologies. The conversation highlights the relationship between investment, innovation, and the future of AI capabilities.
"Wow. Okay you’re saying 1e30 you said by 2028-29. That is literally six orders of magnitude. That's like 100,000x more compute than GPT-4. Yes. The other thing to say is the way you count flops o..."
Patel analyzes TSMC's critical role in the semiconductor supply chain and its ability to meet the growing demands of the AI industry. He discusses the necessity for TSMC to adapt to the increasing requirements for silicon and the importance of demonstrating consistent growth in AI-related revenues to secure future investments. The segment underscores the strategic decisions that TSMC must make to remain competitive.
"Given the fabs that are currently planned and being built, is that enough for the 1e30, or will we need more? I think so, yeah. Okay, so then the chip goal doesn't make any sense.The chip goal st..."
In this segment, Patel addresses the implications of rising power consumption from data centers, which currently account for a small percentage of total US power usage. He discusses the challenges of scaling power generation and distribution to meet the demands of rapidly growing data centers. The conversation highlights the potential for innovation in the power sector as companies respond to increased demand.
"Who will need to see? Investors or companies? More so, TSMC needs to see NVIDIA volumes continue to grow straight up and Google's volumes continue to grow straight up, and so on down the list. Chi..."
Patel discusses the financial dynamics of AI development, including the significant investments required for training advanced models like GPT-5. He emphasizes the lag between investment and revenue generation, explaining how companies like OpenAI and Microsoft are planning for future growth despite current financial risks. The segment explores the broader implications of these investments for the AI industry.
"that people can innovate when given the need to. It's one thing if it's a shitty industry where my margins are low and we're not growing really. All of a sudden it’s like, "Oh, this is the sexiest..."
In this segment, Patel and his co-host discuss the revenue potential of the AI industry and the need for it to reach smartphone-level opportunities to sustain growth. They analyze the return on investment for major tech firms and the psychological factors driving investment decisions in AI. The conversation highlights the importance of perceived value and market confidence in shaping the future of AI.
"This is intense. It's very intense. Jon, actually, if he's right, or not him, but in general. If the capabilities are there, the revenue is there… Revenue doesn't matter. Revenue matters. Is ther..."
Patel concludes the discussion by speculating on the future of AI models, particularly GPT-5, and the expectations surrounding its capabilities. He emphasizes the importance of delivering groundbreaking advancements to maintain investor confidence and drive revenue growth. The segment encapsulates the urgency and excitement surrounding the AI landscape as companies prepare for the next wave of innovation.
"On AI in particular? No, just generally as a company. Now, obviously there's other factors here. Like what is Meta's ad efficiency? How much of that is AI, right? That’s super messy. But here's th..."
Dylan Patel discusses the staggering investments required for developing advanced AI models like GPT-4 and GPT-5. He emphasizes that the return on investment (ROI) must increase exponentially to justify these costs, and the success of future models hinges on their ability to impress users. This segment highlights the financial dynamics of AI development and the expectations surrounding new releases.
"need the revenue in 2025-2026 to support the $10 billion that OpenAI spent in '23, or Microsoft spent in 2023 and early 2024 to build the cluster, which model they trained in mid 2024, which the..."
The conversation shifts to the broader implications of AI investments, with a focus on the potential for significant capital inflow into the sector. The speakers compare the current AI landscape to past tech bubbles, suggesting that the ongoing investment frenzy could lead to substantial advancements, even if some companies fail. This segment explores the cyclical nature of tech investments and the optimism surrounding AI's future.
"What kind of socks are you wearing, bro? Show them. AWS. GPT-5 is not here. GPT-5 is late. We don't know. I don't think it's late. I think it's late. Okay. I want to zoom out and go back to the end ..."
The discussion delves into historical tech bubbles, drawing parallels between the dot-com era and the current AI boom. The speakers argue that while some companies may collapse, the foundational technologies developed during these periods will pave the way for future innovations. This segment provides insights into how past failures can lead to significant advancements in technology.
"to radically get reshaped for many people. Every time you increment up the intelligence, the amount of usage of it grows hugely. Every time you increment the cost down of that amount of intellig..."
Dylan Patel emphasizes the importance of capital in the AI sector, noting that the current investments are not heavily debt-financed, unlike previous tech bubbles. This segment discusses how the financial backing of major companies can sustain growth and innovation in AI, and the potential for a larger bubble compared to past tech eras.
"How many billions of dollars a year is this bubble right now? For private capital? It's like $55-60 billion so far for this year. It can go much higher. I think it will next year. Let me think abo..."
The speakers reflect on the unpredictable nature of the AI market, discussing how companies must adapt to changing conditions. They highlight the importance of being prepared for potential downturns while also recognizing the opportunities that arise from technological advancements. This segment underscores the need for strategic thinking in the rapidly evolving AI landscape.
"At the turn of the 1990s, there was an immense amount of money invested in things like MEMS and optical technologies because everyone expected the fiber bubble to continue. That all ended in 20..."
The conversation shifts to the competitive pressures faced by CEOs in the tech industry, emphasizing the need for continuous innovation. The speakers discuss the risks of inaction and the importance of seizing opportunities to maintain relevance in a fast-paced market. This segment highlights the drive for innovation as a critical factor for success in the tech sector.
"where it is happening. What is that Warren Buffett quote? I don't even know if it's Warren Buffett. You don’t know who's swimming naked until the tide goes out? No, no, no. The one about how the ma..."
Jon Y shares the story behind the creation of Asianometry, detailing his journey from a tourist channel to a leading source of information on semiconductors and Asian business history. He reflects on his early experiences in Taiwan and how they shaped his content. This segment provides a personal insight into the motivations and evolution of the Asianometry channel.
"I want to hear the story of how both of you started your businesses, the thing you're doing now. Jon, how did it begin? What were you doing when you started the YouTube channel? It’s all about you..."
Jon Y discusses the challenges and strategies behind producing content for Asianometry, including his research process and the types of topics he covers. He shares insights into how he transitioned from a tourist channel to focusing on semiconductor technology and business history. This segment highlights the creative process and dedication required to build a successful YouTube channel.
"What year did people start watching your videos? Let's say like a thousand views per video or something? Oh my gosh, I started the channel in 2017 and it wasn't until 2018 or 2019 that it actual..."
Dylan Patel recounts his journey from a curious child to a semiconductor expert, detailing his early experiences with technology and how they influenced his career path. He shares anecdotes about his learning process and the evolution of his interests in the semiconductor industry. This segment provides a comprehensive look at the personal and professional development that led to his current role.
"should know this but in the end, I just try my best to bring interesting stories out. How do you make a video every single week? These are like… Two a week? You know how long he had a full-time job?..."
Dylan Patel discusses the growth of his semiconductor research and consulting firm, highlighting the importance of networking and continuous learning in the industry. He shares insights into the team's composition and the diverse expertise they bring to the table. This segment emphasizes the collaborative nature of the semiconductor field and the value of specialized knowledge.
"It’s all over the world and across many ranges of… You have ex-hedge funds as well. You kind of have this amalgamation of tech and finance expertise. We just do the best work there, I think. Are y..."
Jon shares his perspective on the memory sector within the semiconductor industry, arguing that breakthroughs in memory technology could have a transformative impact. He reflects on the stagnation of memory advancements since 2012 and discusses the potential for innovation in memory integration with accelerators, emphasizing the importance of passion and hard work in pursuing opportunities in this field.
"do. How it got there to where it is, is just like, “Just try and do the best and try to be the best.” If you were an entrepreneur who's like, “I want to get involved in the hardware chain somewher..."
Dylan and Jon conclude by discussing the vast opportunities for innovation within the semiconductor supply chain. They encourage aspiring entrepreneurs to explore various layers of the industry, leverage AI for efficiency, and pursue their passions to drive success. The conversation highlights the potential for significant advancements and the importance of curiosity and engagement in the field.
"If you have a passion for copper wires… I promise to God, if you make the best copper wires, you'll make a shitload of money. If you have a passion for B2B SaaS, I promise to God, you'll make fu..."