Nvidia Could Be About to Counter a Big Artificial Intelligence (AI) Threat With This Move

Nvidia (NVDA -0.06%) is the dominant force in the market for artificial intelligence (AI) graphics cards, with an estimated 92% market share. This is the reason the company has been growing at a terrific pace, up 230% over the past year.

However, the company faces a potential threat from some tech giants. Let’s examine that threat before checking how Nvidia is reportedly going to take care of it.

Custom AI chips pose a threat to Nvidia

While companies including Microsoft, Amazon, Meta Platforms, and Alphabet are Nvidia’s customers and have spent billions of dollars on its graphics processing units (GPUs), it’s no secret they have been developing custom AI chips to reduce their reliance on the graphics-card specialist.

Alphabet, for instance, has deployed custom AI accelerators known as tensor processing units (TPUs) in Google Cloud to “scale cost-efficiently for a wide range of AI workloads, spanning training, fine-tuning, and inference.” Similarly, Meta Platforms is expected to deploy a new custom AI chip this year in a bid to reduce its dependence on Nvidia.

Microsoft has built custom AI chips for deployment in its Azure data centers, and they’re expected to hit the market this year. Meanwhile, Amazon revealed its own chips for training AI models in November, making them available for use by Amazon Web Services (AWS) cloud customers.

There are two reasons these tech giants have been developing chips in-house. First, Nvidia has been unable to keep up with the massive demand for its AI GPUs. The waiting period for the company’s flagship H100 AI graphics card can reportedly stretch up to a year.

Nvidia is trying its best to increase the supply of its graphics cards with the help of its foundry partners, but customers may not be comfortable waiting for so long to get their hands on these chips.

Second, Nvidia’s AI GPUs are very expensive. The H100 processor reportedly carries a price tag between $30,000 and $40,000. However, investment banking firm Raymond James estimates that it costs Nvidia just over $3,300 to manufacture one H100 GPU, pointing toward the immense pricing power the company enjoys in this market.

So it’s not surprising to see tech giants looking to cut down on such massive spending by developing custom chips internally to tackle specific AI workloads for which an H100 may not be required.

Formally known as application-specific integrated circuits (ASICs), these custom chips are dedicated entirely to performing specific operations rapidly while being energy efficient. Semiconductor research group SemiAnalysis reportedly estimates that a successfully developed custom AI chip could help Nvidia’s customers save hundreds of millions of dollars.

All this tells us why Nvidia may be looking to enter the custom AI chip market.

Nvidia wouldn’t want to let go of this potentially $55 billion revenue opportunity

Broadcom (AVGO -1.55%) and Marvell Technology (MRVL -1.70%) are two leading manufacturers of ASICs, and both companies have witnessed a sharp jump in AI-related orders. Marvell, for instance, could generate $1 billion in revenue from selling custom AI chips this fiscal year. Broadcom, on the other hand, is expected to sell custom AI chips worth $8 billion to $9 billion in 2024, according to one estimate. These two companies together control a 47% share of the ASIC market.