No Contest. Hi everyone. Everyone wants to beat Nvidia. Severalmedia outletsrecently espoused the narrative that the AI chip king would soon face tougher competition as large technology companies get help from Broadcom and Marvell to make better custom AI chips.
Last week, Broadcom shares surged 24% a day after CEO Hock Tan painted an extremely positive picture for the AI chip market, saying the opportunity would be “massive” over the next few years.
Tan said on the company’s earnings call with investors that the AI chip market opportunity for Broadcom’s top three major technology customers would range from $60 billion to $90 billion in fiscal 2027, up from $15 billion to $20 billion in fiscal 2024. He added that Broadcom would gain its “fair share.”
Nvidia stock has fallen since Broadcom’s results. Perhaps, some investors are extrapolating from Tan’s remarks that Broadcom could become a viable alternative to Nvidia.
How serious is the custom AI chip threat to Nvidia? I would say not very.
First, Tan’s positive outlook should be parsed and put in context. Broadcom’s actual financial results and current quarter revenue guidance were roughly in line with expectations. The forecast is also three years away and doesn’t mean the company will win a significant share of the market. The further out you go, the harder markets are to predict.
Second, custom AI chip competition from Big Tech isn’t a new story. Amazon Web Services launched its own AI chip in 2019, and Alphabet’s Google introduced its AI accelerator back in 2015. Both technology giants launched several successor versions, but Nvidia still wins nearly all of the AI chip business.
There are foundational reasons for this. Nvidia has the best, most mature technology stack. Its hardware, software, and networking solutions have been optimized with over a decade of fixes.
When I talk to AI startup founders or corporate technology officers, they all explain why they build their businesses on the Nvidia platform. Porting AI software to other chip infrastructure is never a seamless transition, they tell me. There are always problems and bugs. It’s not worth the risk to save a little money when Nvidia consistently releases vastly more powerful AI GPUs.
Using Big Tech has other issues, as well. If companies choose to build on top of Amazon’s or Google’s offerings, they’re locked into one cloud computing vendor and face the risk future chips may be shuttered. Nvidia has the best track record for backward compatibility, meaning large investments in software aren’t lost when Nvidia launches faster products in the future.
Of course, all this assumes Nvidia continues to innovate and offer the best performance. There are no guarantees, but thus far, the company hasn’t faltered. Instead, it accelerated the pace of innovation late last year, going from a two-year product release cadence to an annual one.
Earlier this month, a Bloomberg News anchor asked Amazon Web Services CEO Matt Garman if Nvidia had 95% AI market share. “I think it’s probably even higher than that,” he replied with a laugh. That’s roughly five years after AWS entered the market.
On Tuesday, Morgan Stanley’s technology team was slightly more generous, estimating the custom AI chip share of the market would rise from 11% this year to just 15% in 2030 with the balance going to AI GPUs, aka Nvidia’s wheel house. It suggests Nvidia will continue to dominate the surging AI chip market through the end of the decade.
History shows that going with the industry standard, Nvidia, is the safer bet for technology buyers. Investors should take note, too.