Nvidia is trading about 16% below its June 2024 high. According to the Boston Globe, possible reasons for this decline include slowing revenue growth and the possibility that generative AI, the driving force behind demand for Nvidia GPUs, is costing too much and delivering too little. There are two concerns:
Another concern for investors is competition for NVIDIA’s 90% share of the AI chip market. One such rival is AI chip maker Cerebras, which filed for an initial public offering on September 30, according to CNBC.
According to my September post in Forbes magazine, Cerebras’ competitive advantages over Nvidia (faster, lower-cost AI inference services) are less compelling than the company’s disadvantages. I think so.
To take market share from Nvidia, Cerebras must add the following features:
Application software that is more capable and convenient than Nvidia’s CUDA. Much less dependence on a single customer. A business model that generates positive cash flow.
As I mentioned in my book Brain Rush, there were 2,888 IPOs during the dot-com era. Cerebras’ IPO could be the beginning of the era of generative AI.
I hope the IPO is successful enough to encourage more AI startups to go public, but until the company has at least several quarters of better-than-expected growth, I wouldn’t rush to take Cerebras public. I have no intention of embarking on it.
Cerebras IPO File
Cerebras competes with Nvidia in designing AI chips and operates cloud services. According to CNBC, Cerebras claims its pizza box-sized WSE-3 chip has “more cores and memory than Nvidia’s popular H100.”
Cerebras’ revenue surged in the first half of 2024, but its net loss declined. Specifically, the chip designer’s revenue increased more than 14 times year-over-year to $136.4 million in the first half of 2024, but its net loss decreased from $77.8 million to $66.6 million, CNBC reported.
Like Nvidia, Cerebras designs its chips and partners with TSMC to manufacture them. According to the company’s IPO prospectus, Cerebras’ competitors include Nvidia, AMD, Intel, Microsoft, and Google, as well as “in-house developed application-specific integrated circuits and various private companies.”
Cerebras’ direct competitor Ampere Computing LLC (backed by Oracle) is not pursuing an IPO. With no clear path to listing, Ampere is “exploring the possibility of a sale,” Bloomberg reported.
Major investment banks Goldman Sachs and Morgan Stanley are not underwriting Cerebras’ IPO. That task falls to Citigroup and Barclays. CNBC noted that it is unclear why top investment banks are not involved.
Additionally, whether Cerebras’ IPO values the company higher than the “more than $4 billion” valuation it received in a $250 million funding round in 2021 should be of great interest to investors. .
Cerebras’ faster and cheaper inference engine
In August, Cerebras launched a cloud computing service for AI inference, or responding to natural language user queries, after training an AI chatbot. According to AI News, Cerebras claims its inference service runs “20 times faster at a fraction of the cost” than competing chips.
Specifically, Cerebras announced that its service can process 1,800 output tokens per second, which is 20 times more than typical hyperscale cloud products using Nvidia’s GPUs, AI News wrote. are. Additionally, Cerebras-powered services are “cost-effective,” sources told AI News.
Faster, cheaper inference should be attractive to enterprises, but Cerebras’ cloud service doesn’t address the costly process of training large language models. Additionally, Cerebras faces significant challenges competing with hyperscalers such as AWS, Microsoft Azure, and Google Cloud in building generative AI applications.
Although Nvidia’s chips are more expensive than its competitors, the company says customers are still reaping the quick benefits of the chips. “People who are investing in Nvidia infrastructure are seeing immediate returns,” Nvidia CEO Jensen Huang said on an Aug. 29 conference call with analysts. “This is the best ROI infrastructure, computing infrastructure investment you can make today.”
What is Nvidia’s source of investment return? Despite having higher prices than its competitors, its chips have better performance and lower operating costs, resulting in the lowest total cost of ownership. That more than makes up for the price, Brain Rush points out.
Cerberus’ over-reliance on a single customer
If Cerebras’ WSE-3 chips can deliver a good return on investment, the company could win over Nvidia customers. At the moment, Cerebras relies heavily on one customer, who is also a key investor.
More specifically, Cerebras derived 87% of its revenue from G42, a data center operator based in the United Arab Emirates, according to Bloomberg. The Commerce Department’s Bureau of Industry and Security has granted Cerberus a license to export its chips to the G42, which will be deployed in the United States.
In the future, BIS could hinder Cerebras’ efforts to sell more chips in the UAE because the approval process would take longer and the government “could impose burdens that businesses and customers find unacceptable.” Bloomberg pointed out that there is a possibility that
Investors should consider whether if G42 were to switch AI chip vendors, Cerebra could grow more quickly and diversify its customer base without putting the company’s survival at risk.
Nvidia’s superior application software
Cerebras expressed confidence in its ability to overtake Nvidia. When a reporter asked how much market share Cerebra could take away from industry leaders, CEO Andrew Feldman quickly replied, “Everything,” according to the New York Times. He later revised his response, saying it was “enough to offend them,” the Times reported.
Given Nvidia’s software dominance, Feldman may be too optimistic. Brain Rush pointed out that no other chip company has software comparable to Nvidia’s CUDA. “NVD is working with customers, and they have more customers than everyone else combined,” analyst Jon Peddie said in a September 2024 email. Ta.
“These customers are developing programs on NVD chips, and most are sharing their development with NVD as a way to influence next-generation designs,” Peddie added. So NVD has tens of thousands of SW and HW engineers (probably 5:1 SW), which is not only more than all competitors except Intel. There are also thousands more ghost troops at universities and customer sites. ”
It’s possible that CUDA will move from being a core competency to a core rigidity for Nvidia. That’s only if Nvidia customers start switching to Cerebras chips for faster speeds and lower prices.
Ultimately, the evidence will be in the relative growth trajectories of the two companies. Cerebras’ 14x growth in the first half is much faster than Nvidia’s 122% revenue increase in the quarter ended July.
Nvidia expects growth to be 80% this quarter, but it remains to be seen whether Cerebras’ growth will accelerate.
Cerberas’ cash-burning business model
Meanwhile, Nvidia (net profit margin of 55% in the most recent quarter) is making huge profits while Cerebras is wasting money.
“Since our inception, we have incurred operating losses and negative cash flow due to the development, marketing and expansion of our product portfolio and the continuation of our research and development activities,” Cerebras’ prospectus states.
IPO investors could be rewarded if all of Cerebras’ negative cash flow shifts market share away from Nvidia. However, I will wait a few quarters to see if the company can beat analyst expectations before considering an investment.