CNBC AI News, August 17th – As tech behemoths aggressively invest in the burgeoning field of AI inference, a new report from Morgan Stanley reveals the remarkable profitability hidden within. The choice of AI GPU significantly impacts these profit margins.
Overall, a standard “AI inference factory” boasts impressive average profit margins exceeding 50%, regardless of the chosen chip vendor.
Morgan Stanley’s innovative “100MW AI Factory Model” provides a standardized framework for quantifying and comparing various AI solutions within a unified commercial context.
NVIDIA’s GB200 NVL72 emerges as the clear leader, commanding a near 78% profit margin. This dominance stems from its superior computational performance and the strong ecosystem built around its CUDA software, solidifying its position as the undisputed market leader.
Google’s in-house TPU v6e pod follows with a robust 74.9% profit margin, while AWS’s Trn2 UltraServer achieves 62.5%. Huawei’s Ascend CloudMatrix 384 platform also performs respectably, securing a 47.9% profit margin.
However, AMD’s MI300X and MI355X experienced unexpected setbacks, posting negative profit margins of -28.2% and -64% respectively.
The report suggests that despite significant upfront investment, AMD’s token generation efficiency doesn’t yield sufficient revenue to offset its substantial costs.
Original article, Author: Tobias. If you wish to reprint this article, please indicate the source:https://aicnbc.com/7373.html
