-
Many of the AI chip players are based around an ARM core.
-
Customers mainly include enterprise AI projects and cloud companies.
-
Telcos are just beginning to experiment with AI.
Nvidia, AMD, Intel and Amazon’s Annapurna Labs unit are the companies making early headway in the artificial intelligence (AI) chip space, according to Appledore principal analyst Grant Lenahan.
“Unlike things like 5G…these chips are not actually part of the network, they’re definitely intelligent OSS or AI ops on the side of the network, crunching data and numbers to help you make decisions in not-real time,” Lenahan told Silverlinings on a call.
This is “exactly the flip” of 5G acceleration chips that can do crazy real-time calculations and “keep the latency down to microseconds," he said.
Unlike the x86 jamboree that constitutes much of the 5G virtualized Radio Access Network (vRAN) acceleration world, many of the AI chip players are based around an ARM core. Lenahan noted that the Amazon labs, AMD and Nvidia are all big on ARM cores.
Lenahan notes, however, that what makes one companies implementation different from another’s isn’t just reliant on the ARM cores. It’s also about how efficiently they do pipelining and parallel computations and how effectively they get data into and out of the cache and several other architecture concerns.
Nobody’s talking about that yet, he said.
Who is buying?
So which companies are buying these AI chips so far?
“To the best of my knowledge they’re either selling directly to enterprise AI projects or probably more often to public cloud companies and other cloud companies,” Lenahan said.
“Telcos are just beginning to experiment with real AI,” the analyst said. Part of the reason for this is because their data is “siloed” and telcos have been “very, very nervous” about customer data and personal information.
“They’ve always been afraid of the regulators,” Lenahan said.