Nvidia Fuels Bets on AI Trade

**SPEAKER_1** (0:02)
Bloomberg Audio Studios, Podcasts, Radio, News.
This is Bloomberg Businessweek Daily, reporting from the magazine that helps global leaders stay ahead, with insight on the people, companies, and trends shaping today's complex economy. Plus, global business, finance, and tech news as it happens. The Bloomberg Businessweek Daily podcast with Carol Massar and Tim Stenovec on Bloomberg Radio.

**Tim Stenovec** (0:32)
Today's trade all about Nvidia. Shares were up by more than 8.7 percent. It added more than $440 billion in market cap just today.

**Carol Massar** (0:42)
That's incredible.

**Tim Stenovec** (0:42)
Yesterday, we talked to the CEO of Lam Research. Right. It's a big company. $400 billion in total market cap.

**Carol Massar** (0:50)
So what does it tell you?

**Tim Stenovec** (0:51)
Well, that's a question that Mandeep Singh is going to answer for us. Salesforce shares also having their best day in six years, more than 22.5%. Mandeep Singh is Bloomberg Intelligence Global Head of Technology Research. He joins us here in the Bloomberg Interactive Brokers Studio. Wow. So I'm going to pose the question to you that I posed to Ian in the 2 o'clock hour. And it's something that Elizabeth on our team pointed out earlier today.
This post-market reaction from the company's stock versus what the stock did after the call. A huge delta there. Why was that?

**Mandeep Singh** (1:25)
I mean, that 70% number that they gave for 2027 and their fiscal 2028 is what drove all that optimism.

**Tim Stenovec** (1:34)
So should we just ignore everything about Nvidia numbers on an earnings day until we hear from the CFO on the call?

**Mandeep Singh** (1:40)
Well, they don't normally give a future one year guide. So this was sort of an aberration in terms of how they guided in the second quarter. But look, that's where they have the visibility and also the supply commitments that they have made. It was up almost $160 billion to their suppliers. So this is your memory makers, your TSMC.
They have committed this much so that their suppliers can expand capacity and still they feel the demand is almost like 100 percent, but they are guiding for 70 percent and they can meet that demand.

**Carol Massar** (2:19)
So this is going beyond the hyperscalers?

**Mandeep Singh** (2:21)
This is absolutely going beyond the hyperscalers. And look, with hyperscalers, we've seen the CAPEX expectations have gone up because of what Microsoft said and what Google said on their earnings call. And what they said was, not only is the upside coming from neoclouds and these new players, which you alluded to, SpaceX could be huge in terms of, you know, the 100 billion dollars that they have added to that guide for 2027
I mean, this company could do 650 billion dollars in data center revenue next year in 2027 So a lot of that upside is coming from the likes of SpaceX, hyperscalers raising their CAPEX and neoclouds. And that's what they want. They want a much bigger ecosystem of buyers of their chips.

**Tim Stenovec** (3:11)
We're getting more just superlatives coming as the market closes and Ian King writing that Nvidia's sales forecast sends shares on its biggest rally going back to 2025 I'm wondering about the mode, Mandeep, and how deep Nvidia's mode is when it comes to its semiconductor technology.

**Mandeep Singh** (3:30)
I mean, look, there is no doubt they have competition. And Google TPUs is the best example that a custom ASIC is very effective both in training and inference workloads.

**Tim Stenovec** (3:41)
Okay, you're going to have to explain a little bit of the jargon.
Training versus inference, right? I think a lot of people understand what that is. But in terms of what Google's TPU is doing versus what Nvidia's core product is.

**Mandeep Singh** (3:54)
So Google TPU, which is in its ninth generation, they have been creating this custom chip to run first the search workloads and then the AI workloads. And Anthropic has used this chip for both training their LLM, and Anthropic is the best frontier LLM right now. So they didn't train on Nvidia. They've used Google TPUs for training. They've used Amazon Trainiam and Google TPUs for inferencing workloads. And now they have started to use Nvidia chips. But so far, your best LLM has not used Nvidia chips. And so even then, Nvidia continues to do so much better every year, partly because of what they have done with regards to the ecosystem they have created, the NeoCloud base that they have, where the likes of Corvive, Nibius. These companies are growing faster than your hyperscalers, which are growing fast as well. So because this pie is so big and these LLM companies have shown the revenue, had it not been for Entropic reaching an ARR of $65 billion and going public soon, this would not have been possible. But the fact that LLM companies have shown the revenue has allowed these chip companies like Nvidia and Broadcom to say, okay, we can guide for next year because the LLM companies want the gigawatt capacity. And so that's where the revenues have enabled these companies to really maintain this pace of growth. And so far, the LLM companies haven't disappointed. Now, once Entropic goes public and we learn about their margins and OpenAI's margins, and if they're burning a lot of cash, that story could change. Then people may not be willing to fund this at unlimited pace. But for now, it sounds like everyone is looking at the growth that they are able to generate from AI.