AMD CEO on New Data Center Chip, Demand

Helios, MI455X, and the Future of AI Compute – A Detailed Summary

Key Concepts:

• Helios: AMD’s first rack-scale system solution, housing the MI455X chip.
• MI455X: AMD’s first 2nm chip, designed for high-performance AI compute in cloud environments.
• MI440X: A chip utilizing similar building blocks to MI455X, targeted at enterprise AI applications and existing data centers.
• Yataflop: A unit of compute power defined as 10^24 FLOPS, representing the massive scale of future AI compute needs.
• AIPC (AI PC): Personal computers equipped with dedicated AI processing capabilities.
• Open Ecosystem: AMD’s strategy of fostering collaboration and interoperability with a broad range of partners.
• Compute Deficit: The growing gap between the demand for AI compute and the available supply.

I. The Growing Demand for AI Compute & Introduction of Helios/MI455X

The discussion centers around the escalating demand for AI compute, driven by the increasing adoption of AI technologies. Currently, there are over a billion active AI users, projected to exceed five billion within the next five years. To meet this demand, AMD has introduced Helios, a rack-scale system solution, and the MI455X chip. Lisa Su emphasizes that the MI455X is a significant leap forward, built on 2nm and 3nm technology, boasting 320 billion transistors. This chip is designed to address the growing need for powerful AI processing in cloud environments. The core message is that AI is at an inflection point, becoming pervasive across all aspects of computing.

II. MI455X Deployment & Partnerships

The MI455X is slated for deployment in the second half of 2026, with a ramp-up following its initial release. AMD has established strong partnerships to facilitate this rollout, notably with OpenAI (Greg Brockman appearing at the CES keynote) and Oracle. These partnerships will focus on exploring diverse use cases and accelerating the adoption of the MI455X. The emphasis is on a collaborative approach to address the compute deficit.

III. Addressing Diverse AI Needs: MI440X for the Enterprise

Recognizing the heterogeneity of the AI landscape, AMD has also launched the MI440X chip. While built on similar foundational technology as the MI455X, the MI440X is specifically designed for enterprise applications. It allows businesses to integrate advanced AI capabilities into existing data centers without requiring a complete infrastructure overhaul. This addresses a key need for enterprises wanting to control their data and deploy AI on-premise or in private clouds. Use cases include workflow optimization, financial services applications (where data privacy is paramount), and healthcare.

IV. Quantifying the Compute Deficit & the Concept of the “Yataflop”

A significant portion of the conversation focuses on the severity of the current compute deficit. AMD estimates that approximately 100 zettaflops (10^21 FLOPS) of compute power currently exists globally. However, to support the projected growth in AI users to five billion, this compute capacity needs to increase by a factor of 100. Lisa Su introduced the term “yataflop” (10^24 FLOPS) to illustrate the scale of compute required, highlighting that future needs will be 100 times greater than current capacity. This underscores the urgency of expanding compute infrastructure.

V. AMD’s Competitive Advantage & Open Ecosystem

When asked about competing with Nvidia, Lisa Su positions AMD as offering outstanding performance at a competitive total cost of ownership. Crucially, AMD champions an open ecosystem and deep partnerships, contrasting with a more vertically integrated approach. This open ecosystem allows for greater flexibility and innovation, enabling customers to leverage the best solutions from various vendors.

VI. The Role of AI in Various Industries & Economic Impact

The discussion highlights the broad applicability of AI across industries, including financial services, healthcare, entertainment, and even AMD’s own software development processes. AI is enhancing productivity and enabling new capabilities in diverse fields. Lisa Su believes that AI will have a significant positive impact on global GDP, driven by increased efficiency and innovation. She emphasizes that AI is becoming “everywhere and for everyone,” empowering individuals and businesses alike.

VII. Bottlenecks & Challenges in Deploying Compute

While acknowledging the rapid progress in AI, Lisa Su identifies several bottlenecks hindering the deployment of compute. These include challenges in ramping up production of advanced silicon (2nm and 3nm chips), securing sufficient high-bandwidth memory, and ensuring adequate power infrastructure. She stresses the importance of collaboration across the entire supply chain to address these constraints. Power availability is specifically mentioned as a growing concern globally.

VIII. AMD’s Future Roadmap: MI500 & Beyond

Looking ahead, AMD is focused on continued innovation. The MI500, slated for 2027, is projected to deliver 1,000 times the performance of the MI300 generation. This ambitious goal will be achieved through advancements in hardware, software, and system code design. AMD is also exploring opportunities in physical AI, exemplified by its partnership with Generative Bionics and the development of humanoid robots.

IX. China Market & Export Controls

AMD views China as an important market for its data center products. The company has received licenses from the US government for some previous-generation chips (MI308) and is currently applying for licenses for the MI325. Lisa Su expresses optimism about securing these licenses, emphasizing the importance of balancing national security concerns with market access. The demand for AI in China is high, and AMD is working with both the US government and Chinese customers to find viable solutions.

X. Market Share & Metrics for Success

To demonstrate its progress in the data center market, AMD will focus on metrics such as the adoption of the MI455X, the realization of tens of billions of dollars in AI revenue by 2027, and overall market share gains.

XI. PC Market & AIPC’s Impact

Despite challenging market conditions, AMD has been gaining market share in the PC market, driven by its strong product portfolio and early investment in AIPC (AI PC) technology. The refresh cycle with Windows 11 has also contributed to this success. While supply chain constraints, particularly related to memory, remain a concern, AMD sees opportunities for continued growth in the enterprise laptop segment.

XII. Government Support & AI Infrastructure

Lisa Su acknowledges the positive impact of the US government’s AI action plan, which aims to accelerate infrastructure development and streamline regulations. She highlights the Genesis mission as an example of successful public-private collaboration. AMD is working closely with the government to ensure a balanced approach to export controls and to foster domestic AI innovation.

Notable Quote:

“You ain’t seen nothing yet.” – Lisa Su, expressing her optimism about the future of AI and AMD’s role in it.

Conclusion:

AMD is aggressively positioning itself as a key player in the rapidly evolving AI landscape. Through innovative chip designs (MI455X, MI440X), strategic partnerships, and a commitment to an open ecosystem, the company aims to address the growing compute deficit and unlock the full potential of AI across diverse industries. The focus on both cloud-scale and enterprise solutions, coupled with a forward-looking roadmap (MI500), demonstrates AMD’s ambition to lead the next wave of AI innovation.