Meta Places Orders Worth Hundreds of Billions, AMD Competing with NVIDIA

On February 24th, before the U.S. stock market opened, Meta and AMD announced a five-year strategic partnership on their official websites. Meta will deploy 6GW (each gigawatt capacity transaction worth hundreds of millions of dollars) of AMD GPUs. In exchange, AMD will provide up to 160 million shares (about 10% equity) of performance-based warrants. Analysts believe that this strategic cooperation not only demonstrates Meta’s strong desire to expand in AI but also marks an important step for AMD to directly compete with industry leader NVIDIA.

10% Equity for 100 Billion Dollar Orders

According to the agreement, starting in the second half of 2026, Meta will purchase AMD-designed chips and server equipment tailored for AI models over five years. AMD CEO Su Zifeng revealed that each gigawatt of capacity will involve transactions worth “hundreds of millions of dollars.” This means the total value of the agreement could exceed $60 billion, potentially reaching over $100 billion.

As part of the deal, AMD is designing a semi-custom MI450 GPU specifically for Meta’s workloads. AMD expects initial shipments to begin in the second half of 2026. The products will include a customized Instinct GPU based on the MI450 architecture and the sixth-generation EPYC CPU codenamed “Venice,” running ROCm software, and built on the Helios rack-level platform.

Su Zifeng stated that Meta has long been deeply involved in AMD chip design and will also receive customized versions of the MI450 accelerator card and subsequent generations. Janardhan said that being able to more precisely define chip requirements is a key reason why Meta chose to partner closely with AMD.

“Our strategy is to accelerate at full speed and scale up,” Su Zifeng said. “Our cooperation with Meta has been very smooth from the start, and this agreement elevates our relationship to a new level. We are honored to expand our strategic partnership with Meta to help them break through AI limits on an unprecedented scale.”

Another part of the cooperation involves Meta obtaining warrants to purchase 160 million AMD shares in phases. When project milestones are reached and AMD’s stock price hits certain benchmarks, these warrants will gradually become exercisable, making Meta a significant AMD shareholder. The first warrants will be granted when Instinct GPU shipments reach 1 gigawatt, with subsequent warrants awarded as Meta’s procurement volume increases to 6 gigawatts.

Market insiders call this new financing approach “circular financing.” Under this arrangement, one company pays another, which then purchases products or services from the first. As AI heats up, the amount invested in such deals has surged.

Goldman Sachs, in its latest research report, pointed out that the core logic of this deal is that Meta exchanges potential equity for AMD’s deep customization services and long-term supply commitments, while AMD dilutes future shares in exchange for guaranteed orders and market share from a top large-scale cloud provider.

Renowned business consultant and corporate strategy expert Ho Hongyi said that this “circular financing” model, combining large orders with warrants, is actually a win-win design for both sides. From an industry perspective, this model breaks traditional buyer-seller relationships, creating a risk-sharing, profit-sharing community between tech companies and chip manufacturers. It could become a new trend for collaboration between compute-intensive enterprises and semiconductor firms, promoting a more collaborative industry ecosystem.

** Meta’s AI Obsession**

For Meta, this partnership not only provides customized chip components but also allows deep involvement in the design direction of next-generation semiconductors. This deal is part of Meta’s ongoing expansion. CEO Mark Zuckerberg has prioritized AI as a key strategic focus, committing to invest hundreds of billions of dollars in “aggressive” infrastructure development.

Last month, Zuckerberg officially launched the Meta Compute initiative, aiming to build “tens of gigawatts of computing power within this decade, gradually expanding to hundreds of gigawatts,” to establish a strategic advantage over competitors. For reference, 1 gigawatt of computing power is roughly equivalent to the output of a nuclear power plant, enough to power about 700,000 households.

Janardhan, who reports directly to Zuckerberg, also revealed that the company has not yet determined which data centers will host these new AMD chips. These processors will mainly be used for AI inference, i.e., the practical application phase after model training.

It’s worth noting that in October last year, AMD reached an agreement with AI star company OpenAI, with terms nearly identical to this Meta collaboration. OpenAI will deploy up to 6 gigawatts of AMD Instinct GPUs over the coming years, and AMD will issue up to 160 million warrants to OpenAI, with exercise conditions tied to deployment progress and stock price milestones.

On February 17th, Meta also announced a multi-year, cross-generational strategic partnership with NVIDIA, covering local deployment, cloud, and AI infrastructure. Meta will build large-scale data centers for AI training and inference, supporting its long-term AI infrastructure roadmap.

Meta’s Global Infrastructure Head, Santosh Janardhan, said: “Our goals are very ambitious.” Meta will continue developing its own AI chips while also purchasing NVIDIA products. The three types of chips will support different AI computing tasks. “With our current scale, all three solutions (self-developed, AMD, NVIDIA) have their roles,” Janardhan said in an interview.

Meta’s demand for computing power stems from its aggressive AI expansion plans. The company’s Q4 2025 financial report projects that AI-related capital expenditure will reach $115 billion to $135 billion this year—almost double last year’s amount. Zuckerberg stated during the earnings call that the company will continue heavy investments in infrastructure to train leading models and provide personal superintelligence for billions of people and businesses worldwide. Meta CFO Susan Li further revealed that the company still faces “capacity constraints.”

Yuan Shuai, co-founder of New Wisdom New Quality Productivity Salon, said that deploying AI computing power can support Meta’s metaverse vision, helping realize functions like virtual avatar generation and scene rendering, moving the metaverse from concept to interactive product form, and gaining an edge in the AI race within the tech industry.

** AMD’s Race to Catch NVIDIA?**

Less than a month ago, AMD released its Q4 2025 and full-year financial results. Data shows that in Q4 2025, AMD’s revenue reached a record $10.3 billion, up 34% year-over-year. Full-year revenue hit $34.6 billion, also a record high. Notably, data center revenue in Q4 reached $5.4 billion, up 39% year-over-year, setting a new record.

Meta’s investment clearly bets that AMD’s stock price is still in the early stages of a long-term upward trend. Some warrants become exercisable if the stock price hits $600, but AMD closed Monday at only $196.6. After the announcement, AMD’s stock rose over 8%.

Analysts believe this major deal marks a key step forward for AMD in catching up with NVIDIA. It also shows that despite concerns about potential investment bubbles, overall spending on AI equipment in the tech industry continues to accelerate.

Currently, Meta is AMD’s second-largest customer and will play a more critical role in AMD’s growth strategy. Wall Street expects AMD’s revenue to grow 34% this year. Even an additional $10 billion in revenue could significantly accelerate AMD’s pace in catching NVIDIA.

AMD CFO said, “We expect this cooperation to significantly boost the company’s revenue over the coming years and improve our non-GAAP EPS, marking another important step toward achieving a sustainable long-term financial model.”

As this order begins to be fulfilled in the second half of 2026, AMD’s AI business will shift from “promising prospects” to a “performance ramp-up” phase. Su Zifeng emphasized during the analyst call after earnings that the second half of 2026 will be a critical turning point, with the Instinct MI450 series AI accelerators scheduled for mass production and shipment, starting from Q3. The next-generation MI500 series, based on the CDNA6 architecture and 2nm process, is expected to launch in 2027.

Goldman Sachs, in its latest research report, stated that this announcement is a major positive for AMD, providing greater certainty about its market share among Tier-1 large-scale cloud service providers. Due to equity ownership and product customization, Meta has a strong motivation to use AMD’s products.

However, Goldman Sachs also noted that they will adopt a more positive stance on AMD only after seeing actual deployment timelines from Meta and OpenAI, maintaining a “neutral” rating with a target price of $240.

Yuan Shuai believes that this five-year partnership with Meta is a key step for AMD to break NVIDIA’s AI chip monopoly. By securing long-term, guaranteed orders from Meta, AMD can not only generate hundreds of billions to trillions of dollars in stable revenue but also enhance its brand recognition in the AI chip market through Meta’s industry influence, demonstrating its capacity to supply large-scale compute needs for top cloud providers.

Senior industry economist and Ding Technology founder Ding Shaiji also shared this view: “Binding Meta to AMD can accelerate AI chip iteration, expand market share, and deepen cooperation through equity ties, providing critical leverage in the race to catch NVIDIA.”