Following Groq’s accusation that Nvidia is delaying AI GPU orders for companies working with other suppliers of data center processors for AI applications, Scott Herkelman, a former vice president at AMD, said that Nvidia pursues similar tactics. The high-ranking executive even went as far as to call Nvidia a GPU cartel.
“This happens more than you expect, NVIDIA does this with DC [data center] customers, OEMs, AIBs, press, and resellers,” Herkelman wrote in an X post in response to a tweet from Tom’s Hardware. “They learned from GPP to not put it into writing. They just do not ship after a customer has ordered. They are the GPU cartel, and they control all supply.” You can expand the embedded tweet below to see his reply.
This happens more than you expect, NVIDIA does this with DC customers, OEMs, AIBs, press, and resellers. They learned from GPP to not put it into writing. They just don’t ship after a customer has ordered. They are the GPU cartel and they control all supply.February 27, 2024
Herkelman’s accusation is a big deal. He ran AMD’s graphics business unit from 2016 until he left in 2023, and that group had to compete against Nvidia both on consumer and data center fronts. Perhaps more importantly, he worked as general manager of Nvidia’s GeForce business from September 2012 to May 2015.
A recent Wall Street Journal article brought to light allegations against Nvidia, claiming that the company tends to delay shipments of data center GPUs to customers who are considering AI processors from competitors, such as Groq. Jonathan Ross, Groq’s CEO, suggested to the Journal that the situation has created an atmosphere of discretion among clients, with some even concealing and denying their interactions with rivals.
Nvidia’s chief executive, Jensen Huang, has countered concerns about preferential shipping in the past, stating that the company aims to allocate supplies fairly and provide alternative solutions, such as renting GPU performance from cloud service providers, while customers await their orders.
Tech giants like Amazon Web Services, Google, and Microsoft are developing their own AI accelerators while they continue to ingest massive amounts of Nvidia silicon, though they say they do not intend to directly compete with Nvidia. These companies are also some of Nvidia’s largest customers, so it is likely they’ll continue to place orders and see shipments flow in.
Herkleman compared Nvidia’s moves to the company’s GeForce Partner Program, which earned the ire of the industry and plenty of negative press back in 2018, eventually forcing Nvidia to axe the program.
The allegations against Nvidia, if true, also draw parallels to past tactics employed by Intel to discourage partners from using AMD products in the 1990s and 2000s. However, the report does not provide any concrete evidence of Nvidia’s alleged behavior. The situation raises questions about the company’s practices, but the lack of clear proof leaves the matter unresolved.