Nvidia commands an estimated 81 to 90 percent of the AI data center accelerator market, but this dominance is now being challenged by several major players. AMD has secured a total of 14 GW in deployment commitments from OpenAI, Meta, and Anthropic for its MI450 and Helios platforms. Cerebras launched its CS-4 rack-scale inference system delivering 750 PFLOPS of performance, roughly double its previous generation. Google signed a commercial deal with Marvell Technology on July 29, including a warrant for 58.97 million Marvell shares, and is targeting general availability of its TPU v8 by late 2026. Amazon, Meta, and Microsoft are all developing custom silicon, which could erode Nvidia’s long-term dominance despite the stickiness of its CUDA software ecosystem.
Source: Read the original article
