6 Gigawatts of GPU Capacity Meta and AMD Agree on Delivery of AI Chips Worth 60 Billion US Dollars

Related Vendor

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

Meta intends to purchase AI chips from AMD worth USD 60 billion over the next five years. The two companies agreed to supply GPUs with up to 6 GW TDP as well as a participation option for Meta in AMD.

The US semiconductor manufacturer Advanced Micro Devices (AMD) and the Facebook group Meta Platforms have signed a multi-year agreement to expand AI computing capacities. Meta intends to purchase up to 6 GW of graphics processing units (GPUs) over the next five years. According to both companies, the order volume is around 60 billion US dollars.

The first expansion phase comprises 1 GW and is scheduled to begin in the second half of 2026. A customized version of AMD's upcoming MI450 architecture will be used. The chips will be optimized for Meta's AI inference applications, i.e. for running trained models on the fly.

Participation Option for 160 Million AMD Shares

Part of the agreement is an option commitment for up to 160 million AMD shares. The tranches will fall due in stages, depending on delivery volumes, technical milestones and share price performance. If exercised in full, this would correspond to a share of around ten percent that Meta could gain in AMD in this way.

AMD had already agreed a similar model with OpenAI. Such participation structures are increasingly being used in the industry to secure delivery commitments and long-term purchase guarantees.

AMD CEO Lisa Su described the agreement as a multi-year partnership spanning several chip generations. Meta will also be one of the first customers of the sixth generation of EPYC processors, which are being developed under the code name "Venice".

Expansion of the Data Centers

In addition to GPUs, Meta is also planning to purchase customized CPUs from AMD. These are to be specially tailored to the requirements of Meta's own data centers and achieve a favorable ratio of performance to energy consumption.

Meta has been significantly increasing its investments in data centers for years. The company is planning investments of between USD 115 and 135 billion for 2026. In the previous year, expenditure amounted to around 72 billion US dollars. A new data center in the US state of Indiana is estimated to cost around USD 10 billion, another in Louisiana around USD 27 billion.

The expansion is linked to the growing demand for infrastructure for training and operating large AI models. Industry-wide investments of more than USD 600 billion in data centers and AI chips are expected for the current year.

Several Suppliers Planned

Despite the agreement with AMD, Meta continues to rely on several chip suppliers. The company has also concluded an extensive supply agreement with Nvidia and purchases billions of GPUs from them. Meta is also working with ARM on its own chips and is in talks with other suppliers.

Meta's Head of Infrastructure, Santosh Janardhan, emphasized that the scale of the planned data centers would require different technologies. No single chip could efficiently cover all workloads.

For AMD, it is the second major order of this kind following the agreement with OpenAI. Here, too, the delivery of up to 6 GW of GPU capacity over several years and processor generations was agreed. The company is thus expanding its position in the market for AI accelerators, which has so far been heavily dominated by Nvidia.

Observers point out that large technology companies are broadening their supply chains in order to avoid bottlenecks and reduce their dependence on individual manufacturers. At the same time, investors are focusing on the high level of investment in AI infrastructure in view of potential overcapacity.

The partnership between AMD and Meta is designed for several chip generations. Deliveries are to increase gradually until the agreed 6 GW is reached. How quickly this expansion takes place depends on the demand for AI services and the progress of data center construction.(sg)

Digitalisation in production

Digital infrastructure for plastics processing machines