Here's where AMD stands as AI competition heats up

Key Concepts

• Total Addressable Market (TAM): The entire market demand for a product or service.
• Serviceable Addressable Market (SAM): The segment of the TAM targeted by your products and services which is within your geographical reach.
• Compound Annual Growth Rate (CAGR): The average annual growth rate of an investment over a specified period of time.
• Scale-Up Architecture: A system design that allows for the connection and utilization of a large number of GPUs.
• Training: The process of teaching an AI model by feeding it large amounts of data.
• Inference: The process of using a trained AI model to make predictions or decisions.
• Cost Per Token: A metric used to measure the efficiency of AI models, particularly in natural language processing.
• Enterprise SAS: Software as a Service applications used by businesses.
• Hyperscalers: Large cloud computing providers like Amazon Web Services, Microsoft Azure, and Google Cloud.
• RTX (Nvidia): Nvidia's line of graphics cards often used for AI inference in enterprise settings, typically air-cooled.
• Rockm (AMD): AMD's software ecosystem for its GPUs, particularly relevant for hyperscalers.
• ARM IP: Intellectual Property related to the ARM architecture, commonly used in mobile and increasingly in server processors.
• Stargate: A potential acquisition or investment target for OpenAI, as mentioned in the context of SoftBank's Nvidia stake sale.

AMD's AI Market Opportunity and Strategy

AMD CEO Lisa Su has projected the total AI chip market to reach $1 trillion by 2030. This figure represents the Serviceable Addressable Market (SAM), encompassing CPUs, GPUs, networking, and accelerators. Patrick Morehead believes AMD can achieve double-digit market share within this vast market.

Market Share Projections and Growth Drivers

• Server CPUs: AMD aims to secure over 50% market share in server CPUs, including ARM-based variants.
• Data Center Revenue: The company projects an 80% CAGR for data center AI revenue and a 60% CAGR for overall data center revenue.
• Conservatism in Projections: Morehead notes that Lisa Su's projections are conservative, and the exact percentage of double-digit market share (e.g., 20% or 30%) remains unknown. This uncertainty is attributed to the nascent stage of upcoming competitive products, with more advanced offerings against Nvidia expected in Q3 2026. The projections are also considered conservative over a three to five-year timeframe, which is significant in the rapidly evolving AI landscape.

AMD's Next-Generation AI Chips vs. Nvidia

AMD is developing next-generation AI chips that aim to compete directly with Nvidia's offerings.

Architectural Comparisons

• Scale-Up Architecture: AMD is focusing on a "scale-up" architecture, which allows for addressing a significantly larger number of GPUs. AMD is currently at eight GPUs and plans to scale to 72 to 144 GPUs, a capability comparable to Nvidia's Blackwell architecture.
• Performance and Cost Benefits: This scale-up capability is crucial for enhancing performance in both AI training and inference. It is expected to lead to much lower cost per token, a key metric for efficiency.
• Timeline: AMD's competitive AI chips are anticipated to launch in Q3 2026, a slightly more refined date than the previously heard "second half of 2026."

AMD's Enterprise Market Strategy

While AMD has a strong presence in the hyperscaler market, its position in the enterprise sector is currently weaker.

Challenges and Opportunities in Enterprise

• Market Share Weakness: Enterprise SAS and hyperscaler markets represent AMD's lowest market share segments.
• Sales and Marketing Focus: The primary challenge in the enterprise sector is not technology but rather sales and marketing efforts. AMD needs to increase its "feet on the street" to engage with enterprises and develop content that resonates with their specific needs.
• Enterprise AI Inference Solutions: A key roadmap item to watch is AMD's enterprise equivalent to Nvidia's RTX line. These are typically air-cooled, designed for traditional data centers, and specifically optimized for inference. While AMD is likely working on such solutions, a definitive roadmap is not yet available.
• CPU Performance: AMD is performing well in the CPU segment for enterprises and needs to focus on foundational aspects.
• GPU Software Development: Beyond the hardware roadmap, AMD must double down on its software ecosystem. While significant progress has been made with Rockm for hyperscalers and neoclouds, a more deliberate integration with enterprise software vendors is necessary.

AMD's China AI Business

The future of AMD's AI business in China is highly dependent on geopolitical factors and government regulations.

Current Status and Future Outlook

• Zero Data Center AI Business: Currently, both AMD and Nvidia have zeroed out their data center AI business in China due to restrictions.
• Potential for Wiggle Room: Any relaxation of these restrictions by the US administration or Chinese authorities could open up significant opportunities.
• Economic Impact: If restrictions ease, it could add tens of billions of dollars annually to AMD's projected $100 billion annual data center revenue over the next three to five years.
• Political Influence: Recent statements by political figures like Donald Trump regarding China could potentially influence these policies.

Nvidia's Market Dynamics and SoftBank's Investment Shift

Recent market movements, including Nvidia's stock decline and SoftBank's divestment, are significant.

SoftBank's Nvidia Stake Sale

• Sale of Entire Stake: SoftBank has sold its entire stake in Nvidia for nearly $6 billion.
• Investment in OpenAI: The proceeds are being redirected to invest in Sam Altman's OpenAI, indicating a strategic shift by SoftBank's CEO, Masayoshi Son, towards higher-growth potential areas.
• Rationale for Rotation: This move is interpreted as a classic "rotation" by Son to double down on OpenAI investments, potentially for initiatives like "Stargate" and other future acquisitions. He is also exploring investments in companies that could be combined with ARM IP.
• Timing of Sale: The sale occurred when Nvidia was at a high valuation, suggesting a strategic move to capitalize on current market conditions.

AMD Analyst Day Takeaways

AMD recently hosted its analyst day in New York City, where CEO Lisa Su outlined the company's AI strategy and demand outlook.

Key Announcements and Outlook

• TAM of $1 Trillion: Reiteration of the $1 trillion TAM for AI.
• Upcoming AI Chips: Plans for new AI chips, including the MI450 and MI500 lines, and F Helios line of RAT scale servers.
• Market Share Gains: AMD explicitly stated its intention to gain market share from Nvidia and Intel in both the data center and client (PC) markets.
• Revenue Leadership Goal: AMD aims to become the revenue leader in the data center market.
• Overall Upward Trend: The company sees a generally positive trajectory for its business.
• Market Correction: The current stock decline in the AI space is seen as a broader market trend affecting all players.