Meta Platforms (META.US) has initiated a significant five-year program to acquire AI infrastructure from Advanced Micro Devices (AMD), with plans to purchase up to 6 gigawatts of capacity starting in the second half of 2026. This deal, which is among the largest AI infrastructure agreements in the current capital expenditure cycle, marks a pivotal moment for AMD, solidifying its status as a “tier-1” supplier alongside Nvidia. Following the announcement, AMD shares surged over 14% in pre-market trading, reflecting investor optimism surrounding the infrastructure boom in AI.
Under the terms of this agreement, Meta has pledged to invest more than $100 billion in AI compute capabilities from AMD over five years. This includes the procurement of custom AI chips and equity warrants linked to AMD’s future stock performance, potentially granting Meta a 10% stake in the company. Such a commitment not only enhances AMD’s competitive edge against Nvidia in the burgeoning AI chip sector but also supports Mark Zuckerberg’s expansive AI strategy aimed at increasing data-center capacity and diversifying supplier options. AI infrastructure spending is projected to reach as much as $135 billion this year.
AMD emerges as a major AI beneficiary
According to AMD CEO Lisa Su, the total value of this transaction is anticipated to reach “double-digit billions” per gigawatt, indicating that Meta’s multi-year capital expenditure for this initiative could exceed $100 billion, depending on the specific configurations and hardware choices. This agreement transcends a typical supply contract; Meta is set to receive warrants to purchase 160 million shares of AMD stock, which will vest in stages based on project milestones and share-price conditions, potentially positioning Meta as a notable shareholder in AMD.
Market reactions to the announcement were immediate and favorable, with AMD’s stock seeing a notable uptick. Analysts suggest that this partnership may represent a substantial increase in AI-related revenues for AMD, enhancing its credibility within the accelerator segment. Conversely, Meta’s stock experienced a modest rise, indicative of investor perceptions that this deal aligns with its established compute-scale strategy rather than indicating any new risk factors.
Meta’s strategy centers on customizing components to fit its specific operational needs, allowing the company to gain a degree of influence over the development roadmap for future chip generations. This “co-design” approach aims to optimize both total cost of ownership (TCO) and inference efficiency—critical components in enhancing power consumption and cost per task during the deployment of AI models.
Insights from Santosh Janardhan, Meta’s Head of Global Infrastructure, suggest that AMD’s hardware will primarily focus on inference tasks, which are crucial for deploying trained models into production. Inference requires high energy efficiency and consistent capacity availability, as opposed to peak performance capabilities, which are more relevant during the training phase of AI models.
Meta has reiterated its commitment to a multi-sourcing strategy by continuing to develop in-house AI chips, maintaining partnerships with Nvidia, and now integrating AMD into its supply chain. This strategic diversification underscores how hyperscalers view AI accelerators not merely as commodities but as essential resources, thereby mitigating vendor-dependence risks.
For AMD, this partnership not only signifies an increase in order volume but also serves as a critical turning point for its reputation. Already the second-largest customer for AMD, Meta’s new agreement could significantly propel AMD’s growth in the coming years. Given AMD’s revenue of approximately $34.6 billion last year, even a modest increase in sales from this deal could substantially affect its trajectory within the accelerator market.
The scale and structure of this partnership indicate that both Meta and AMD are banking on a protracted investment cycle in AI monetization, even amid ongoing discussions regarding the potential risks of an AI capital expenditure “bubble.” With AMD shares currently testing the ~$191 area (EMA200) after a nearly 25% drop from recent highs, the Meta agreement may provide the necessary support and re-establish a fundamentally sound upward trend for the stock.
See also
Anthropic Launches Enterprise Agents Program with Finance and HR Plugins
Anthropic Accuses MiniMax, DeepSeek, and Moonshot AI of Massive Model Mining Scheme
