101 Jefferson Dr, 1st Floor
Menlo Park CA 94025

(408) 623 9165

info at jpdata dot co
sales at jpdata dot co

After five years, how is the AI chip market doing

After five years, how is the AI chip market doing

I have been fortunate to follow AI chip market since 2016. When I wrote the first report, there were only a handful of start-ups trying to build AI chips. The need for AI acceleration was recognized, but it was unclear what applications would emerge and if ‘cat vs dog’ classification, the state of the art at the time, would be useful for anything from a market point of view.

Looking back, the journey feels almost like science fiction. AI makes Moore’s law look outdated. In the field of AI chips, there’s no 10% or 20% improvement, everything is of the order of  ‘X’ times. The overall market overall, the company revenue, and conferences – all have benefitted and experienced a similar ‘X’ times increase. In the past five years, AI chip companies have become Wall Street’s darling, and nerds have struck again!

7X increase in market forecast for 2025

The first forecast, that came out in 2016 predicted that the AI chipset market will reach $12 billion by 2025, a large number at the time. In 2016, chip companies had to struggle many years to reach $1 billion in revenue.  Nvidia for example, had needed 20 years to reach 5 billion in revenue. It was a bold prediction that AI chipset market will become twice Nvidia’s revenue in 8 years. I remember grilling by several investors, and top chip companies on how such scale is possible, and what use cases justify this. The forecast increased with every market update after that and by 2022, the same prediction for 2025 went up by over 7X (See Figure1  below). The data center market prediction increased 3X and the edge market by 16X.

Figure 1: Total AI chip market forecast increased by 7X from 2016 to 2025. It will certainly increase even more.

Today, it is routine to see AI chip forecasts ranging from $50 to $100 billion, and the number is likely to increase even more. AMD has already said that they see a $150 billion market opportunity in the data center for AI chips, which is vast compared to the current number.

Nvidia’s data center grows by 31X

Large semiconductor companies have been the primary beneficiary of this boom. Companies that have benefitted include Nvidia, Intel, and Qualcomm, among others. Nvidia’s data center business unit generated a staggering $3 billion in Q1 23 revenue, all driven by AI. Not only that, the company hit $1 trillion in valuation, a landmark for a chip company.

Figure 2: Nvidia and Qualcomm were the two biggest beneficiaries on the data center and edge market.

Nvidia was the top beneficiary in the data center market, with an estimated 31X revenue increase from 2016 to 2022 (Figure 2), whereas Qualcomm was the top beneficiary on the edge market, with an estimated 47X revenue increase driven by AI. 

Despite all the hype going to start-ups, they have been unable to keep up with larger chip companies. The highest yearly revenue for a start-up tops at $50 million in my estimate. Start-ups underestimated the need for software and, with passing every day, face an uncertain future.

5X or more increase in published papers and conference attendees

Academia and conferences have equally benefitted from the AI boom. A search for the ‘neural network’ phrase on Arxiv, the popular portal for publishing academic papers, shows 787 papers published in 2016, compared to 10613 in 2022 – a 6X increase (See Figure 3). The number of attendees to the AI Hardware Summit, where the first gathering of all AI chip companies took place in 2018, increased 5X from 2018.

Figure 3: Conference attendance and papers increased by 5X and 6X between 2016 and 2022.

Road forward

It seems that there’s no stopping for AI chip market as we move forward. AI has become so popular that even Snoop Doggy Dog thinks it is f—ing cool. Only AI chips will make this coolness come to reality, and AI chips will benefit. Generative AI is the current buzzword, and given the complexity of forward pass, and latency requirements, it is sure to become a significant contributor to the AI chip business. More silicon will be added for AI in every chip, including CPUs like Xeon and Snapdragon, increasing the market further. On the edge side, AI will become a feature in every AI chip, and a part of silicon will be dedicated to AI.

All this points to an optimistic forecast for the AI chip market. After five years, the AI chip market is going as strong as ever, and the forecast continues to increase.

I will be sure to write about AI chips after five more years!