Did a flawed Goldman Sachs report roil the market on Friday?

On Friday, shares of Nvidia fell 4%, dragging the Nasdaq to a three-week low.

One of the triggers for the sell-off was a Goldman Sachs memo from Peter Oppenheimer claiming that traffic to ChatGPT was in steep decline. Goldman posted this chart, which later went viral (including in the Financial Times). Show the number of visits to ChatGPT:

The chart was further picked up by the usual suspects who claimed that ChatGPT was just a hoax, that Meta/Grok/Anthropic was eating its lunch, that it was woke or whatever other agenda they were pushing.

The truth is simple and embarrassing.

ChatGPT’s URL has been changed to chatgpt.com from chat.openai.com.

When both URLs overlap, the traffic volume is:

Thanks @edsuh

In any case, traffic speeded up.

For the “smartest people in the room,” this reflects an insulting lack of critical thinking. There was absolutely no reason why ChatGPT usage dropped by +80% in just two months.

Did this mistake wipe $110 billion off Nvidia’s market cap on Friday (that’s 72% of Goldman Sachs’ market cap, for the record)?

I doubt this was the main trigger, but I don’t doubt it was painful. Research and critical thinking are rare in this meme-driven world.

NVDA daily

What worries me about Nvidia is the life cycle of the investment boom. The H100 chip is one of the best products ever made and demand for it is huge. By all accounts that demand will be at least equal for the Blackwell generation that will be released later this year.

But analysts are factoring in this level of demand—which is increasing every year—and that means the expected price-to-earnings ratio in 2025 will be 37 times.

The first problem is that they need to keep iterating to expand their moat, which is hard to do with profit margins approaching 80%. Now I wouldn’t bet against them on that, but the amount of money being spent on chipmaking right now is extraordinary and is essentially a bet against capitalism.

Second, there has to be a return on investment from the buyers. Right now, all the big tech companies are investing in chips, but at some point, those investments have to pay off. We’re pricing that level of investment year after year, and I find it hard to believe that all these companies will continue to spend that much money in a world where technology is going to dominate everything.

Third, comments by Broadcom CEO Hock Tan on Thursday after earnings suggest there is a significant threat to Nvidia’s demand from the same hugely capped tech companies:

“I used to think that general-purpose commercial silicon would eventually win. Well, based on the history of semiconductors so far, general-purpose, small commercial silicon tends to win. But like you, I’ve changed my mind. And by the way, I did, last quarter, maybe even 6 months ago. But still, catching up is a good thing. And I actually think that because I think there are two markets here for AI accelerators. There’s one market for companies in the world, and none of these companies are capable of, and don’t have the financial resources or the interest to build silicon, custom silicon, and large language models and software, maybe, to be able to run these AI workloads on custom silicon. It’s too much and there’s no return for them to do it because it’s too expensive. But there are these few cloud guys, developers with the sheer scale of the platform and the financial wherewithal to make it absolutely rational, economically rational, to build their own custom accelerators because right now, I’m not trying to overstate it, it’s all about the compute engines. It’s all about specifically training these large language models and enabling them to Your platform. It’s all about the limitations, to a large extent, when it comes to GPUs. Seriously, it’s gotten to the point where GPUs are more important than the engineers, who are very expansive in terms of how they think. These GPUs are much more important – or XPUs are much more important. And if that’s the case, what better thing to do than to bring control, control their destiny by building your own custom silicon accelerators. And that’s what I see them all doing. They’re doing it at different rates and starting at different times. But they all started“.”

A week ago, everyone was regretting not buying NVDA stock when it dropped to $90 (and the 69% rally to $130 certainly vindicated those buyers for a while). But after reading these comments, I’m not so sure I’ll buy the stock when it drops back to $90.

flawedFridayGoldmanmarketReportroilSachs