Summary
Highlights
Oracle's recent deal with OpenAI, worth hundreds of billions for cloud services, is likely to benefit AMD due to their close partnership. The partnership is grounded in engineering and performance, with AMD's EPYC line of chips being central to Oracle's non-GPU compute. Oracle claims AMD chipsets offer 40-50% cheaper operational costs for like-for-like workloads, making AMD an economic disruptor.
AMD's MI300 series GPUs, with their higher memory throughput, are well-suited for globally distributed inference, particularly at the edge. This design allows for fewer GPUs per rack, leading to better power efficiency and a smaller footprint, which is crucial for edge computing. Oracle views this as a perfect technical partnership, enabling them to leverage AMD's technology for inference at the edge efficiently.
Oracle's $300 billion deal with OpenAI for the 'Stargate' project signifies a massive investment in AI infrastructure, roughly half of OpenAI's total infrastructure plan. This deal underscores the immense and ongoing investment in AI hardware, with no 'bubble' in this sector. Oracle is strategically positioning itself as a key provider of data centers for companies with overflow needs.
Oracle is planning to build 4.5 gigawatts of compute infrastructure for OpenAI over five years, which is 4.5 times the size of Elon Musk's Colossus 2 cluster (approximately 1 million GPUs and 1 gigawatt of compute). This highlights the massive scale of the partnership. OpenAI is not Oracle's only customer; XAI also relies heavily on Oracle for inference, showing Oracle's growing role in AI infrastructure provisioning.
Of the $300 billion OpenAI is investing in data centers, roughly half (around $150 billion) will be allocated to GPU providers like Nvidia and AMD. While Nvidia may take a larger share, AMD is poised to capture a significant portion, especially by differentiating itself in high-cost-performance inference at the edge. AMD's ability to offer cheaper workloads with higher memory capacity makes them a compelling choice.
AMD has a history of being an exceptional 'second mover' competitor, challenging market leaders like Intel and now Nvidia. Oracle views AMD as a critical hedge against Nvidia's dominance, especially since Oracle lacks its own in-house silicon projects like other hyperscalers. Oracle's heavy investment and cooperation with AMD aim to prevent Nvidia from having unlimited pricing power in the AI infrastructure market.
AMD's MI300 and MI355 GPUs are specifically designed for efficient inference at the edge, offering smaller physical and energy footprints at a lower cost. Inference, the process of running AI models in real-world applications, is expected to be a significantly larger workload than training in the long term. AMD is aggressively positioning itself to capture a majority of this inference market, relying on its enhanced memory capacity and a strong ecosystem to compete effectively.