As the AI arms race enters deeper waters, tech companies' thirst for computing power has shifted from "just having it" to "precision and efficiency." At a recent Morgan Stanley technology, media, and telecommunications conference, AMD CEO Lisa Su clearly defined the direction for AI chip development, a topic of great market interest: future AI infrastructure will not have a "one-size-fits-all" chip; heterogeneous computing is the ultimate solution.
At the same time, she also gave her first in-depth response to the recent conflicts between AMD and Meta and OpenAI."Warrants" Deeply Integrated Cooperation ModelFurthermore, it observes the market anxiety surrounding the current shortage of AI server CPUs and rising memory prices.
AI computing power is no longer a single standardized product: the era of heterogeneous computing has arrived.
In the early stages of AI development, powerful GPUs with a single architecture solved the basic computing power needs. However, Lisa Su points out that AI infrastructure has reached a critical turning point.
• Workload fragmentation:The requirements for computing resources vary greatly depending on the training of ultra-large language models, medium-sized models, and various specific inference tasks.
• Balancing performance and power consumption:When AI models enter large-scale daily inference operations, customers are most concerned about the "price per watt of computing power", and a single chip can no longer be optimized in all aspects.
The Rise of ASICs and Customization: Lisa Su believes that Application-Specific Integrated Circuits (ASICs) will definitely have a place in future computing needs. AMD's strategy is to "have your cake and eat it too," providing both flexible standard products and deeply customized chip designs for specific workloads to achieve the fastest computing optimization.
Deeply Aligned with Tech Giants: Analyzing Meta and OpenAI's "Warrant" Strategy
AMD recently signed a computing power cooperation agreement with Meta worth hundreds of billions of dollars, amounting to 6 gigawatts (GW), and last October, it signed a massive computing power supply agreement with OpenAI. Both deals have one thing in common: AMD released warrants or equity.
Regarding this unique business model, Lisa Su provided a clear strategic logic:
• Accelerating Procurement and Ecosystem Building:Such infrastructure investments, amounting to hundreds of billions of dollars, require an extremely high level of trust. Issuing warrants based on "performance" can substantially accelerate the purchasing decisions of these giants, while simultaneously binding their interests together.
• The driving force for achieving common goals:When customers hold AMD stock options, it's no longer just a simple buyer-seller relationship; both companies have a strong incentive to help each other achieve their technological and commercial goals. For example, Lisa Su revealed that AMD and OpenAI are actively planning their first gigawatt-level computing power installation, and the next-generation MI450 chip is essentially "jointly validated" by both parties.
Supply Chain Status Quo: The Dual Challenges of CPU Shortages and Rising Memory Prices
In addition to her forward-looking AI strategy, Lisa Su also explained the two major pain points currently facing the supply chain.
First, there's the tight supply of CPUs needed for AI. She admitted that the market growth far exceeded predictions from 3 to 6 months ago, and even hyperscalers acknowledged that they "significantly underestimated" the computing demands of AI infrastructure on CPUs. AMD is currently working actively with its supply chain and expects to further expand production capacity in 2026 and 2027 to meet the gap.
Secondly, there's the price surge in memory chips (such as DDR4 and DDR5). Lisa Su warned that this wave of price increases is substantially pushing up systemic pricing and will put greater cost pressure on the personal computer (PC) market. While market volatility is expected to moderate in the second half of the year, the dynamics of the memory market still need close monitoring.
Analysis of viewpoints
It is clear from Lisa Su's speech at the Morgan Stanley conference that AMD is waging an "asymmetric war" that is completely different from its competitors.
While market leaders (such as NVIDIA) attempt to dominate all AI computing with a standardized high-end GPU architecture, AMD has chosen to embrace heterogeneous computing and semi-customization. Through stock options, AMD has effectively turned some of the world's largest AI buyers (such as Meta and OpenAI) into partners. This deep collaboration model, binding companies from the underlying silicon chip architecture all the way to company equity, not only secures massive orders for the next few years but also builds an ecosystem moat for AMD that is extremely difficult to replace.
