Q: In response to why NVDA might be experiencing weakness, CNBC has been talking about:
MON, DEC 16, 2024
The Exchange with Kelly Evans
FROM THE DESK OF KELLY EVANS
 
AS OF MON, DEC 16, 2024 • 11:24 ET
What Just Happened.
 
"The market's most important stock is faltering," the CNBC headline aptly reads this morning. And as our Fred Imbert catalogs, Nvidia shares are down 3% in December while the rest of the market is up nearly 4%, and are down nearly 10% from their November 7th all-time highs.
Now, a tiny correction in a stock that's up 164% this year, and 9x in two years, is hardly reason for concern. But there could be a meaningful reason why the shares are stalling out here--and why shares of another chip company, Broadcom, are suddenly soaring.
It goes back to this fascinating discussion between Chetan Puttagunta, a general partner at venture capital firm Benchmark, and the anonymous fund manager known as "Modest Proposal" on Patrick O'Shaughnessy's podcast a couple of weeks ago. I'm no technical expert in AI, but here's my best effort to summarize their discussion.
Namely, has the arms race to develop the biggest, best, and fastest large language model--the kinds of model that uses hundreds of thousands of Nvidia chips in massive data centers--begun to stall out? A few marquee players, like Meta's "Llama" and Musk/X's "Grok," are still plowing ahead, but the broader market may be starting to shift.
The shift is happening because (1), Meta's Llama model is open-source, and therefore start-up teams are now able to use it to produce smaller, more targeted AI models for specific use cases, and (2), the "training" of large language models using both real and synthetic data has stalled out, giving way to a new era of grading them based on their inferencing ability, also known as "test-time compute."
And if this shift is happening--and the podcast only barely got into the chip implications of this--then it would make sense if demand were also starting to shift from a land grab of Nvidia's workhorse chips, to a market where Broadcom's "custom" chips could suddenly become a very important player. Indeed, Stifel CEO Ron Kruszewski told us that's exactly where his firm is looking as they begin to deploy AI.
And boom--Broadcom's earnings last week confirmed its sudden rise as one of the foremost players in AI. Its overall revenues soared 50% from the year earlier, and its AI revenues were up a whopping 220%. The shares surged more than 20% the next day, putting it above the trillion-dollar market-cap mark for the first time. And they've kept rising, adding another 6% today.
Now, if this shift is real, there could be further implications, ranging from expected data center power usage to perhaps reigniting a start-up boom in AI that many beleaguered Silicon Valley investors thought might never come. And Nvidia could still come out just fine, as top analyst Vivek Arya told us last week, even as he raised his Broadcom price target.
But the shift would certainly explain why Nvidia's performance has been more muted lately.
A final player to watch, by the way, is Marvell, another custom chipmaker. Its stock also surged 20% earlier this month--and is now up 102% this year--after stronger-than-expected earnings. For now, though, it's still a much smaller $106 billion market cap.
So perhaps what we're learning this month, in other words, is that Nvidia may be ceding its crown (to whom exactly, we don't know yet) as the most important stock in the market.
See you at 1 p.m!
Kelly
MON, DEC 16, 2024
The Exchange with Kelly Evans
FROM THE DESK OF KELLY EVANS
 
AS OF MON, DEC 16, 2024 • 11:24 ET
What Just Happened.
 
"The market's most important stock is faltering," the CNBC headline aptly reads this morning. And as our Fred Imbert catalogs, Nvidia shares are down 3% in December while the rest of the market is up nearly 4%, and are down nearly 10% from their November 7th all-time highs.
Now, a tiny correction in a stock that's up 164% this year, and 9x in two years, is hardly reason for concern. But there could be a meaningful reason why the shares are stalling out here--and why shares of another chip company, Broadcom, are suddenly soaring.
It goes back to this fascinating discussion between Chetan Puttagunta, a general partner at venture capital firm Benchmark, and the anonymous fund manager known as "Modest Proposal" on Patrick O'Shaughnessy's podcast a couple of weeks ago. I'm no technical expert in AI, but here's my best effort to summarize their discussion.
Namely, has the arms race to develop the biggest, best, and fastest large language model--the kinds of model that uses hundreds of thousands of Nvidia chips in massive data centers--begun to stall out? A few marquee players, like Meta's "Llama" and Musk/X's "Grok," are still plowing ahead, but the broader market may be starting to shift.
The shift is happening because (1), Meta's Llama model is open-source, and therefore start-up teams are now able to use it to produce smaller, more targeted AI models for specific use cases, and (2), the "training" of large language models using both real and synthetic data has stalled out, giving way to a new era of grading them based on their inferencing ability, also known as "test-time compute."
And if this shift is happening--and the podcast only barely got into the chip implications of this--then it would make sense if demand were also starting to shift from a land grab of Nvidia's workhorse chips, to a market where Broadcom's "custom" chips could suddenly become a very important player. Indeed, Stifel CEO Ron Kruszewski told us that's exactly where his firm is looking as they begin to deploy AI.
And boom--Broadcom's earnings last week confirmed its sudden rise as one of the foremost players in AI. Its overall revenues soared 50% from the year earlier, and its AI revenues were up a whopping 220%. The shares surged more than 20% the next day, putting it above the trillion-dollar market-cap mark for the first time. And they've kept rising, adding another 6% today.
Now, if this shift is real, there could be further implications, ranging from expected data center power usage to perhaps reigniting a start-up boom in AI that many beleaguered Silicon Valley investors thought might never come. And Nvidia could still come out just fine, as top analyst Vivek Arya told us last week, even as he raised his Broadcom price target.
But the shift would certainly explain why Nvidia's performance has been more muted lately.
A final player to watch, by the way, is Marvell, another custom chipmaker. Its stock also surged 20% earlier this month--and is now up 102% this year--after stronger-than-expected earnings. For now, though, it's still a much smaller $106 billion market cap.
So perhaps what we're learning this month, in other words, is that Nvidia may be ceding its crown (to whom exactly, we don't know yet) as the most important stock in the market.
See you at 1 p.m!
Kelly
