Oracle Corp's stock closed up 4% today after reports surfaced that the company is negotiating a $20 billion agreement to provide cloud infrastructure for Meta Platforms Inc.
According to sources cited by Reuters, the parent company of Facebook will utilize this hardware for artificial intelligence training and inference operations. Bloomberg previously reported that the deal's value could exceed $20 billion. It was also noted that other terms might change before final contract completion.
These reports emerge two months after Oracle signed an agreement to build 4.5 gigawatts of data center capacity for OpenAI. A gigawatt equates to the power consumption of hundreds of thousands of households. Shortly after announcing the partnership, sources told The Wall Street Journal that this project could be worth $300 billion over five years.
Artificial intelligence infrastructure demand from clients including OpenAI and Meta has driven Oracle's stock upward by more than 80% year-to-date. Most of this gain occurred following the company's recent earnings release, where Oracle revealed a 359% surge in total remaining performance obligations (a measure of future sales) to $455 billion.
The database technology company is making substantial investments in new infrastructure to meet client requirements. Oracle anticipates increasing capital expenditures by 65% to $35 billion in the current fiscal year.
Oracle Cloud Infrastructure (OCI) provides access to AI clusters equipped with more than 100,000 graphics processing units from Nvidia Corporation. These clusters incorporate networking switches and SHARP technology from the chip manufacturer. SHARP reduces data exchange requirements between GPUs working collaboratively, thereby preserving bandwidth for additional workloads.
Should Meta finalize the $20 billion cloud contract with Oracle, the company might commission AI clusters featuring alternative architectures. Notably, the Facebook parent organization could seek to replace some Nvidia products with internally developed silicon solutions.
In March, sources revealed to Reuters that Meta had begun testing its first custom AI training chip. Previously, the company detailed an accelerator called MTIA optimized for inference workloads. These chips will initially support recommendation algorithms.
OpenAI may adopt a similar approach in its collaboration with Oracle. According to Financial Times reporting, the ChatGPT developer plans to begin mass production of custom AI accelerators developed in partnership with Broadcom Inc. beginning next year. OpenAI has indicated that its commissioned 4.5 gigawatt infrastructure from Oracle will incorporate more than 2 million chips.
Meanwhile, Meta's cloud agreement with Oracle could represent part of its extensive data center initiative detailed in July. This project involves the Facebook parent company investing hundreds of billions of dollars in new AI infrastructure. The first two data center clusters under development - Prometheus and Hyperion - will each require multiple gigawatts of power capacity.
