So as to escape the influence of GPUs, NVIDIA has begun to embrace corporations that produce their very own artificial intelligence (AI) chips. It is thought that Amazon, Microsoft (MS), Google, and Meta, in addition to Open AI, have contacted and discussed supporting the production of dedicated AI chips.
Citing nine well-informed sources, Reuters reported on the ninth (local time) that Nvidia is constructing a recent business unit focused on designing custom AI chips for cloud computing and AI corporations.
Based on this, NVIDIA is thought to have recently met with Amazon, Meta, MS, Google, and OpenAI to debate the production of customized chips.
This movement is concentrated on leading AI and cloud corporations to provide customized AI-specific chips to switch Nvidia's flagship 'H100' or 'A100' GPUs in response to the explosive growth of AI.
Nvidia occupies 80% of the worldwide AI chip market, and despite its high price, it’s so popular that it cannot meet the growing demand.
Nonetheless, experts indicate that the H100 and A100 GPUs are 'general purpose AI processors', so there isn’t any need to make use of expensive GPUs for specific purposes. Greg Raicho, general partner at Eclipse Ventures, identified, “There isn’t any have to rely only on the H100 or A100 to optimize power or application costs,” and added, “It’s efficient to combine and use the crucial chips in line with corporate computing purposes.”
In truth, big tech corporations are one after one other developing customized AI chips to suit their needs. The aim of that is to cut back dependence on GPU and increase power consumption efficiency.
Google already introduced the AI-specific chip 'TPU' in 2016 and continues to upgrade it. Last yr, Amazon unveiled an AI chip called 'Inferentia', and Microsoft unveiled its own chips called 'Maia' and 'Cobalt'. Meta recently announced that it’s going to introduce its own chip called ‘Artemis’ inside this yr. It was reported that OpenAI also reviewed the production of its own AI chip in October last yr and even searched for an acquisition goal.
Particularly, because the customized AI chip market is growing, corporations comparable to Broadcom and Marvell Technology are known to be actively entering this field.
Alan Eccles of the 650 Group, a research firm, predicted that the info center custom chip market will grow to as much as $10 billion (roughly 13.33 trillion won) this yr and $20 billion in 2025. Needham analyst Charles C. also said that the broader custom chip market might be value about $30 billion in 2023, corresponding to about 5% of worldwide annual chip sales.
Currently, custom chip design for data centers is led by Broadcom and Marvell, with annual sales of $10 billion and $2 billion (roughly 2.66 trillion won). Most chip manufacturing is left to TSMC.
The evaluation is that if Nvidia participates in this example, it’s going to pose an amazing threat to each corporations. In truth, after the Reuters report, Nvidia's stock price rose 2.75%, and Marvell's fell 2.78%.
Dinah McKinney, a former Marvell executive, is reported to be leading Nvidia's custom AI department. His LinkedIn profile says the team's goal is to “make technology available to customers within the cloud, 5G wireless, video games, and automotive sectors.”
Nonetheless, after an official request for comment was made, the content disappeared. NVIDIA, Amazon, Google, Microsoft, and OpenAI also declined to comment.
“Nvidia’s entry into custom chips is an actual threat to Broadcom and Marvell,” said Dylan Patel, founding father of semi-analytics, a silicon research group. “It’s clear that there might be more competition on this space as well.”
Reporter Lim Da-jun ydj@aitimes.com