The pricing in the open market is very flexible, which means that devices that are difficult to find can receive high premiums. Todd Burke, Global Business Development President of independent distributor Smith, stated that as of November, there is very little inventory of AI chips on the public market. "From the perspective of the open market or distributors, it is interesting that not many materials have entered our field. Nvidia has strict control over inventory and customers in demand. Therefore, there is indeed a large demand."

The US ban on selling artificial intelligence chips to China has not relaxed supply. Smith's market intelligence manager Todd Banker said that many original equipment manufacturers have stockpiled chips, and companies that acquire these devices are also retaining chips. "If you can get them, even if you don't plan to put them into immediate use or production, you have to keep them," he said.

Will AI chip supply face a boom and bust cycle?

According to Deloitte Consulting, in the spring of 2023, the price of more advanced artificial intelligence chips was approximately $40000 per chip. The shortage/allocation of chips (mainly due to a lack of advanced packaging capabilities) has delayed the launch of generative artificial intelligence products by thousands of companies. The production speed of the companies manufacturing these chips is not fast enough, and it is expected that this imbalance will continue until 2024.

According to Deloitte, the market opportunity value driven by generative artificial intelligence in 2024 is approximately $50 billion. It is estimated that this opportunity will reach $400 billion by 2027, but Deloitte warns that this number may be too optimistic. Some practitioners worry that there may be an artificial intelligence foam from 2023 to 2024.

For example, the generative AI GPU market in 2023 is dominated by a single designer, relying on wafer fab partners with limited production capacity. Deloitte stated that once production capacity increases, the prices of artificial intelligence chips may decrease.

Then there is the bullwhip effect, which is currently being experienced in the electronic product supply chain. When customers allocate, they often overbook. At present, the supply chain is reducing inventory based on orders placed during the most severe period of global chip shortages. Deloitte warns that once the supply and demand of artificial intelligence chips become more balanced, buyers may receive chips that exceed demand and then withdraw them when new production capacity is launched.

Finally, Deloitte pointed out that training and almost all generative artificial intelligence reasoning are completed using the same data center generative artificial intelligence chip. But over time, a significant portion of generative artificial intelligence reasoning may be completed on edge processors. Doing more at the edge may expand the market or lead to a decrease in the price of artificial intelligence chips generated in data centers.

According to TrendForce, AMD has recently increased production of AI chips, and China is moving towards self-sufficiency in AI chip manufacturing, although the development of high-end AI chips is still limited.

Signal from the supply chain

The possibility of artificial intelligence chip foam has attracted attention in the electronic product supply chain. Enterprises seem to embrace artificial intelligence without considering strategic goals.

Smith's Todd Burke shared, "A few days ago, I had a phone call with a client to discuss their artificial intelligence plan and how they will implement and use it. The key point was, 'We use artificial intelligence, we have a solution, we are looking for a problem to solve.'" "They don't know exactly what they want to do with it, but they know that it's clearly a powerful tool, and they just want to figure out how to apply it to their niche markets."

Similar views were also expressed at the ECIA executive meeting in October.

Response methods

Artificial intelligence chips are one of the few semiconductors that are still in short supply. Overall, electronic components are in a surplus state in the supply chain. Smith and other distributors help manage inventory imbalances - they resell excess inventory while searching for difficult to find components. Original equipment manufacturers, component manufacturers, and EMS suppliers typically collaborate with independent distributors to provide these and other services.

"Our sales channel is through original equipment manufacturers, and we know that similar original equipment manufacturers in the same vertical field may utilize excess inventory. Alternatively, we have (original equipment manufacturers) and contract manufacturers available for procurement - we have door-to-door relationships, we buy and sell both," Burke said.

The current surplus is due to non cancellable, non returnable (NCNR) orders during the shortage period. At that time, the price was very high, but as inventory increased, the value of the equipment also decreased. "In some cases, we are helping them transfer it to the open market, and buyers are looking for opportunities to price. In other cases, manufacturers will write down inventory that has lost pricing power," Todd Banker said.

Dealers have stated that due to weak demand in the fourth quarter, it may take several more quarters to consume excess inventory. Banker concluded, "It is difficult to predict when demand will rebound in order to alleviate overcapacity, and it is also difficult to predict when wafer fabs will start production, thereby exacerbating overcapacity." "It is hoped that by the second quarter of 2024, excess capacity will be consumed before a large number of wafer fabs start production."

The article is translated from EPSNews, a sisters magazine of International Electronic Commerce, with a link to the original text：Will AI Chip Supplies Face a Boom-Bust Cycle?