"Everyone wants to challenge NVIDIA"
Recently, multiple media outlets have promoted the viewpoint that the king of AI chips, NVIDIA, will soon face fiercer competition as large tech companies such as Broadcom (AVGO.O) and Marvell Technology (MRVL.O) are helping other companies develop better custom AI chips.
Last week, Broadcom's stock surged 24% in a single day. CEO Hock Tan painted an extremely optimistic picture of the AI chip market during the company's earnings call, stating that opportunities in this market will be 'very huge' in the coming years. He mentioned that by fiscal year 2027, the opportunities for Broadcom's three major tech clients in the AI chip market will grow from $15 billion to $20 billion in fiscal year 2024 to $60 billion to $90 billion, adding that Broadcom will capture a 'reasonable share.'
After Broadcom's earnings report was released, NVIDIA's (NVDA.O) stock price immediately dropped. Some investors may have inferred from Chen's comments that Broadcom could become a strong competitor to NVIDIA.
But how serious is the threat of custom AI chips to NVIDIA? The answer is: not very serious.
Broadcom's actual financial performance and this quarter's revenue guidance are generally in line with expectations. However, this outlook is for a forecast three years down the line and does not represent the company's ability to win a large share of the market. The further out the prediction, the greater the market uncertainty.
Amazon Web Services (AWS) launched its own AI chips back in 2019, while Google introduced AI accelerators even earlier, in 2015, and has released multiple generations of products. However, despite this, NVIDIA has still won nearly all of the AI chip business.
NVIDIA has the most advanced and mature technology stack, with its hardware, software, and network solutions having been optimized and refined over more than a decade.
After speaking with founders of AI startups or corporate tech executives, they unanimously expressed that choosing the NVIDIA platform to build their business is the safer choice. Porting AI software to other chip architectures is not an easy task, and problems and vulnerabilities always arise in the process. It is clearly wiser to use NVIDIA's more powerful AI GPUs than to take risks just to save a bit of cost.
If companies choose chips from Amazon (AMZN.O) or Alphabet (GOOGL.O), they will be locked into a single cloud computing provider, and the risk of future chip discontinuation cannot be ignored. Meanwhile, NVIDIA excels in backward compatibility, meaning that customers using NVIDIA will not waste their substantial software investments due to the launch of faster products.
Although no company can guarantee to stay ahead forever, NVIDIA has never faltered so far. Since the end of last year, NVIDIA has accelerated its pace of innovation, shifting from a two-year product release cycle to an annual one.
Earlier this month, in an interview with Bloomberg News, Amazon Web Services CEO Matt Garman was asked if NVIDIA holds 95% of the AI market share, and he responded with a smile, 'I think it might be even higher.' This comes five years after AWS entered the market.
On Tuesday, Morgan Stanley's tech team estimated that the share of custom AI chips in the market will rise from 11% this year to 15% by 2030, while the remaining market will still be dominated by AI GPUs (NVIDIA's stronghold). This indicates that by the end of this century, NVIDIA will still dominate the booming AI chip market.
Investors should also take note of this. Although Broadcom's potential in the AI chip market is exciting, NVIDIA's leading position is hard to shake in the short term.
Article reposted from: Jinshi Data