AMD's Compute Expansion Fuels Five Billion AI Users - Episode Hero Image

AMD's Compute Expansion Fuels Five Billion AI Users

Bloomberg Tech · · Listen to Original Episode →
Original Title:

TL;DR

  • AMD's MI 4555, featuring 2nm and 3nm chips with 320 billion transistors, represents a significant leap in AI compute capability, essential for scaling from one million to an anticipated five billion AI users within five years.
  • The MI 440 chip targets enterprise AI applications, enabling upgrades to existing data centers with new capabilities without requiring complete rebuilds, addressing the need for on-premise or private cloud deployments.
  • AMD projects a 100x increase in global compute capacity is needed over the next four to five years to meet the demand for AI, moving from zettaflops to yottaflops (10^24 flops).
  • AMD's strategy for humanoid robots positions them as the "brain" provider, supplying components for real-time local capabilities and the underlying software trained on AMD accelerators.
  • The PC market growth for AMD in 2025 was driven by its product portfolio and early bet on AI PCs, with an expectation of continued growth in enterprise laptops where they are currently underrepresented.
  • AMD is navigating supply chain constraints across silicon, memory, and power to expand compute environments, emphasizing the critical importance of ecosystem-wide partnership and planning.
  • Lucid's robotaxi ambitions, a partnership with Waymo and Uber, aim for rapid market rollout within 18 months, integrating luxury EV technology with autonomous driving systems.

Deep Dive

AMD's CEO Lisa Su articulated a vision of massive AI compute expansion driven by a family of new chips, signaling a significant inflection point for the company and the broader industry. The core argument is that the exponential growth in AI adoption, projected to scale from one million to over five billion users in five years, necessitates a corresponding increase in computing power, requiring a 100-fold increase in compute capacity over the next five years, reaching the yottaflop scale. This demand is being met by AMD's new MI 4555 for large data centers and the MI 440x for enterprise applications, aiming to provide high performance at an advantageous total cost of ownership within an open ecosystem.

The second-order implications of this compute expansion are profound and multifaceted. Firstly, the sheer scale of demand highlights a significant compute deficit, where current capabilities are already a bottleneck for advanced AI model development, as noted by OpenAI's Greg Brockman. This deficit is compounded by broader ecosystem constraints, including memory, energy, and electricity, which the industry must address collectively through coordinated supply chain efforts and partnerships. Secondly, AMD's strategy of offering a range of chips addresses the heterogeneous nature of AI use cases, from massive cloud training to on-premise enterprise applications, enabling businesses to upgrade existing data centers rather than build entirely new ones. This approach is crucial for enterprises, particularly in sectors like financial services and healthcare, that require data control and on-premise deployments.

Furthermore, AMD's emphasis on an open ecosystem and deep partnerships, contrasted with potential vendor lock-in, positions them to capitalize on this growth by providing turnkey solutions. The company's aggressive product roadmap, with the MI 500 series promising a thousand-fold performance increase over the MI 300 generation by 2027, underscores a commitment to pushing the bleeding edge of hardware capabilities through advanced technology and system co-design. The ambition to be the "brain inside" humanoid robots represents a strategic push into physical AI, leveraging existing FPGA and embedded real-time capabilities.

Finally, the discussion around AMD's ability to sell into China, contingent on US government licenses, reveals the geopolitical complexities influencing the global AI supply chain. While demand in China is acknowledged as high, navigating export controls requires careful collaboration between governments and companies. The ongoing dialogue signifies a recognition that addressing the global compute deficit and unlocking AI's full potential requires navigating both technological and regulatory landscapes, with the ultimate outcome pointing towards AI becoming increasingly integrated into daily life and business processes, driving productivity and economic benefits across the board.

Action Items

  • Audit AI compute demand: Quantify the projected 100x increase in compute needed over the next 4-5 years by analyzing user growth and current compute capacity (ref: Lisa Su, AMD CEO).
  • Implement AI development system: Build an AI development system to enhance software developer productivity and accelerate product time-to-market (ref: Lisa Su, AMD CEO).
  • Track AI adoption metrics: Measure the impact of AI on business processes and productivity by establishing key performance indicators for AI integration across 3-5 core business functions.
  • Evaluate AI hardware ecosystem: Assess the performance and total cost of ownership of AI accelerators from multiple vendors (e.g., AMD, Nvidia) for enterprise applications.

Related Episodes

Key Quotes

"Look Helios is a massive you know system you can you know see it in the background here and MI 4555 is uh just an incredibly powerful chip and probably the the context I would give Ed is you know one of the things that we're so clear about is that the demand for AI compute is just continuing to increase and you know we have seen that over the last five years when you think about just how much new capabilities have come on board we've now seen a real inflection in the number of people who are using AI."

AMD CEO Lisa Su explains the significance of their new AI chip, highlighting the rapidly growing demand for AI compute power. Su projects a massive scaling of AI users from over one million currently to over five billion within five years, underscoring the necessity for advanced computing solutions.

"And from that standpoint, you know, MI 4555 is a significant leap forward in terms of technology capability, you know, made up of two and three nanometer chips, 320 billion transistors, just a lot of performance and a lot of capability."

Lisa Su further elaborates on the technical prowess of the MI 4555 chip, emphasizing its advanced manufacturing process (2 and 3 nanometer chips) and substantial transistor count (320 billion). Su positions this chip as a major advancement in delivering the high performance and capability required for the escalating AI compute demands.

"Well, I think you see many enterprises now using AI all throughout their business processes, whether you're talking about things in their workflow. I can even AMD, we're using AI through every part of our development process. We're seeing a lot of applications in financial services in healthcare."

AMD CEO Lisa Su discusses the broad adoption of AI across various enterprise sectors. Su points out that AI is being integrated into core business processes and development cycles, with significant applications emerging in fields like financial services and healthcare.

"You know, we think we have to increase compute by another 100 times as you go over the next four or five years, and I introduced a term last night, the yottaflop. You know, people are like, what is a yottaflop? A yottaflop is actually 10 to the 24th in terms of flops."

Lisa Su introduces the concept of "yottaflop" to quantify the immense future compute needs for AI. Su projects a 100-fold increase in required compute power over the next four to five years, defining a yottaflop as 10 to the 24th flops to illustrate the scale of this demand.

"Well, our job as the industry is to push the bleeding edge. I mean, that is our job. And so, you know, when we think about like the MI 455 deploying, you know, 2 nanometer and 3 nanometer chips, having the latest generation memory, high end with memory that that is out there and really deploying these big systems, the important thing is that the entire ecosystem come together and we plan together for this next big inflection in compute."

AMD CEO Lisa Su emphasizes the industry's role in advancing technology to meet AI's escalating demands. Su highlights the importance of deploying cutting-edge chips and memory, stressing that a collaborative ecosystem approach is crucial for planning and executing the next major phase of compute advancement.

"Well, the PC market is a very good market for us. You know, we grew a ton in the PC market in 2025, and you know, that really came from the strength of our product portfolio. We bet early in AI PCs, so it was a clear area where we believed that the technology would generate demand."

Lisa Su discusses AMD's success in the PC market, attributing it to their product portfolio and early investment in AI PCs. Su explains that AMD anticipated AI technology would drive demand in this sector, positioning them favorably for growth.

Resources

External Resources

Books

  • "The President's AI Action Plan" - Mentioned as a forward-leaning strategy for US leadership in AI.

Articles & Papers

  • "AI Action Plan" (The White House) - Mentioned as a document outlining a strategy for US leadership in AI.

People

  • Lisa Su - CEO of AMD, discussed in relation to new chip announcements and AI compute demand.
  • Mark Winterhoff - Interim CEO of Lucid, discussed in relation to EV technology, robotaxi ambitions, and partnerships.
  • Min Liang Tan - CEO of Razer, discussed in relation to AI products for enhancing the gaming experience.
  • Greg Brockman - OpenAI President, mentioned as a partner discussing AI use cases and compute constraints.
  • Michael Kratsios - Mentioned as having joined Lisa Su on stage to discuss the Genesis Mission.
  • Jensen Huang - CEO of Nvidia, discussed in relation to new self-driving platforms, robotics, and demand for chips in China.
  • Roland Busch - CEO of Siemens, mentioned as a guest for an upcoming conversation.
  • Elon Musk - Mentioned in relation to concerns about Nvidia's self-driving platform.

Organizations & Institutions

  • AMD - Discussed in relation to new chip technology (MI 4555, MI 440), AI compute, and PC market strategy.
  • OpenAI - Mentioned as a partner discussing AI use cases and compute constraints.
  • Oracle - Mentioned as a partner of AMD.
  • Nvidia - Discussed in relation to self-driving platforms, robotics, and demand for chips.
  • Lucid - Discussed in relation to EV technology, robotaxi ambitions, and partnerships with Nero and Uber.
  • Nero - Mentioned as a partner in robotaxi ambitions with Lucid and Uber.
  • Uber - Mentioned as a partner in robotaxi ambitions with Lucid and Nero.
  • Santander - Mentioned as a partner of Nvidia.
  • Mercedes - Mentioned in relation to Nvidia's Drive platform for B2C customers.
  • Razer - Discussed in relation to AI products for enhancing the gaming experience and its ecosystem.
  • Siemens - Mentioned as a guest for an upcoming conversation.
  • LG - Mentioned in relation to its laundry folding robot.
  • Meta - Mentioned in relation to its smart glasses.
  • Aura Ring - Mentioned as an example of AI-enabled jewelry.
  • Samsung - Mentioned as having an AI ring.
  • BYD - Mentioned as the number one EV producer globally.
  • Tesla - Mentioned in relation to EV market share and concerns about Nvidia's self-driving platform.
  • Xiaomi - Mentioned as a growing EV company in China.
  • Generative Bionics (GBionics) - Mentioned as an investor and technology partner of AMD, unveiling a humanoid robot.
  • Nutonomy - Mentioned as a partner in robotaxi ambitions with Lucid and Uber.
  • J.P. Morgan - Mentioned in relation to its bullish outlook on compute demand for Nvidia and AMD.
  • Morgan Stanley - Mentioned in relation to its assessment of new information from AMD.
  • Barkley's - Mentioned in relation to market commentary on Tesla.
  • US Government - Mentioned in relation to licenses for selling products to China and export controls.
  • White House - Mentioned in relation to the President's AI Action Plan.

Tools & Software

  • Sierra AI - Mentioned as an AI-powered customer experience platform.
  • ChatGPT - Mentioned as an example of a large AI model.
  • Gemini - Mentioned as an example of a large AI model.
  • Grok - Mentioned as an example of a large AI model and an AI wearable assistant.
  • Nvidia Drive - Mentioned as a platform used by Lucid for B2C customers and robotaxis.
  • Windows 11 - Mentioned as a factor in the PC refresh cycle.

Websites & Online Resources

  • sierra.ai - Mentioned as a website to learn more about Sierra AI.

Other Resources

  • MI 4555 - AMD's first rack scale system solution and a powerful chip.
  • MI 440 - AMD chip focused on enterprise applications.
  • MI 308 - Previous generation AMD chip for which licenses were obtained for China.
  • MI 325 - Recent generation AMD chip for which license applications are in process for China.
  • MI 355 - Previous generation AMD chip from which MI 4555 is a 10x improvement.
  • MI 500 - Future AMD generation planned for 2027 with 1000x performance of MI 300.
  • Helios - AMD's first rack scale system solution.
  • Yottaflop - A unit of computation equal to 10^24 flops.
  • Genesis Mission - A public-private partnership approach to advance science in the United States.
  • Project Motoko - Razer's AI wearable device unveiled at CES.
  • AI PC - Personal computers with AI capabilities, discussed as a driver for the PC market.
  • FPGAs - Field-Programmable Gate Arrays, mentioned in relation to AMD's work in physical AI.
  • H200s - Nvidia chips with strong demand from China.
  • Blackwell - Nvidia's next generation architecture.
  • Verra Architecture - Nvidia's next generation architecture, with prototypes already received.
  • Ryzen - AMD's PC processor line, discussed in relation to new innovations and future shipping.
  • Gravity - Lucid's SUV, mentioned in relation to its luxury experience and integration with autonomous driving technology.
  • Robotaxi - Autonomous vehicles for ride-hailing services, discussed by Lucid, Nero, and Uber.
  • Humanoid Robot - Robots with human-like form and capabilities, discussed in relation to AMD's strategy and Generative Bionics.
  • Physical AI - AI applied to physical systems and robots, discussed as a future market for AMD.
  • AI Gaming - The integration of AI into the gaming industry, discussed by Razer.
  • QA Companion - A tool for game developers to shorten quality assurance cycles.
  • DRAM - Dynamic Random-Access Memory, mentioned as a potential constraint in the PC market.
  • EVs (Electric Vehicles) - Discussed in relation to Lucid's technology and market position.
  • Global EV Market - Discussed in relation to Lucid's position and competition.
  • AI Wearables - Wearable devices with AI capabilities, discussed in relation to Razer's Project Motoko and other form factors.
  • Smart Glasses - Wearable devices with visual display capabilities, discussed as a category and compared to other form factors.
  • AI Agents - Software agents that perform tasks using artificial intelligence, mentioned in the context of customer experience platforms.
  • Compute Deficit - The gap between the demand for computing power and its availability, discussed in relation to AI development.
  • Export Controls - Regulations governing the export of technology, discussed in relation to US government policy on China.
  • Tariffs - Taxes on imported goods, discussed in relation to Lucid's supply chain strategy.
  • Open Source Models - AI models whose source code is publicly available, mentioned in the context of China.
  • Closed Models - AI models whose source code is not publicly available.
  • Total Cost of Ownership - The overall cost of owning and operating a product or system, mentioned by AMD as a competitive advantage.
  • Ecosystem - A network of interconnected components and participants, discussed by AMD and Razer.
  • Supply Chain - The network of organizations and activities involved in producing and delivering a product, discussed by AMD, Lucid, and Razer.
  • Memory Storage - The capacity to store digital information, discussed in relation to Sandisk and Micron.
  • Water Cooling - A method of cooling electronic components using water, mentioned in relation to cooling Nvidia chips.
  • AI Avatar - A digital representation of a person or entity powered by AI, mentioned by Razer.
  • AI Dog - A robotic dog with AI capabilities, mentioned as a cute implementation of AI.

---
Handpicked links, AI-assisted summaries. Human judgment, machine efficiency.
This content is a personally curated review and synopsis derived from the original podcast episode.