OpenAI Builds Its Own AI Chip, Wall Street Braces for Nvidia Earnings

Zaid Admani (0:00)

Public.com presents The Rundown. Your daily market update in 10 minutes. My name is Zaid Admani, and today is Wednesday, August 26th. In today's episode, we'll break down the fresh inflation numbers that just dropped this morning, and what it means for the Fed. We'll also tell you about OpenAI's new Jalapeño AI chip, and preview Nvidia's earnings that are coming up. Then stick around to the end of the show to find out why your life insurance policy might have lent LeBron James $300 million. We got a great show for you today.

Let's go.

Stocks were back in the green on Tuesday, with the S&P 500 gaining 0.3%, while the NASDAQ jumped 0.7%, led by a rally in chip stocks. And I think stocks also got a boost yesterday from treasury yields pulling back and oil prices dropping. The 10-year treasury yield had its biggest one-day decline yesterday since June, following to around 4.64%. And Brent Crude fell nearly 4% yesterday after reports of negotiations between Oman and Iran about reopening the Strait of Hormuz. So there was some good news to hit the tape yesterday, and that brings me to inflation. We just got the PCE inflation report for July this morning. This is the Fed's preferred inflation gauge, and the numbers weren't so bad. The July PCE report showed headline inflation running at 3.7% year over year, slightly hotter than the 3.6% economists expected, but core PCE, which trips out food and energy, and is probably the number the Fed cares about the most, that came in right in line with expectations at 3.3%. Now, keep in mind, this still means that inflation is running above the Fed's 2% target. You know, inflation has been stubbornly sticky in the 3% range for a while now, but I guess at least it's not getting worse. I think this report basically gives the Fed zero reason to do anything right now when it comes to interest rates. Inflation is still too high to start cutting rates, and I don't see the Fed hiking rates either, given the recent soft job market numbers. So we'll see what Fed Chair Kevin Warsh has to say about all this. Remember, the Fed is gathering in Jackson Hole this weekend, where Kevin Warsh will give a keynote speech on Friday. So really looking forward to seeing what he has to say. We'll continue to stay on top of all of this, so definitely get subscribed to the podcast if you haven't already, and tune in every day to stay in the loop.

Let's run through some headlines, starting with OpenAI. OpenAI announced that their in-house built AI chip can outperform Nvidia in certain tasks. Their new chip is called Jalapeño, which honestly, I kind of like that name. And according to OpenAI, it definitely brings the heat. This chip beats Nvidia's Grace Blackwell Ultra AI chip in two areas, how much AI work it can handle per unit of power and how quickly it responds. So that's a pretty impressive feat by OpenAI, but there are two important caveats here. Jalapeño is mainly designed for inference work, which is the part of AI where a model is already trained and it actually answers your questions. This chip is not optimized to train massive AI models though. That's still an area where Nvidia dominates. The other caveat here is that OpenAI didn't test Jalapeño against Nvidia's latest system called Vera Rubin. That ship just started the ship, so Nvidia didn't test it against that. So I wouldn't call Jalapeño an Nvidia killer just yet. In fact, OpenAI's chip chief, Richard Ho, said that Nvidia remains a key partner and that OpenAI still needs a lot of Nvidia chips. So I wouldn't panic if you're an Nvidia investor just yet. But I do think big picture, this has to be kind of concerning for Nvidia because some of their biggest customers, including Google and OpenAI, are all developing their own chips to reduce their reliance on Nvidia. Now both Google and OpenAI have partnered with Broadcom to help develop their own chips. So Broadcom has become one of the big winners here. And you know, it's easy to see why these companies are developing their own chips. You know, running AI products like ChatGPT is insanely expensive and requires a ton of Nvidia chips. And right now, Nvidia is able to charge a premium for their chips and make a large margin because they have the best product. But if OpenAI can move more of their inference workloads onto their own chips, that could eventually save the company a ton of money and reduce their reliance on Nvidia. OpenAI said they plan to start using their Jalapeño chips later this year to run some of their AI models. So Nvidia has to be a little bit nervous here. And for OpenAI, I think what they're trying to do is vertically integrate, kind of like how Apple is. OpenAI already has one of the best frontier AI model, but they also want their own chips and their own data centers. And who knows, maybe their own hardware devices too. So we'll see if that strategy pays off. By the way, somewhat related here, the executive who has been overseeing OpenAI's data center without Chris Malone just left the company last week. So I gotta say, there's a lot of turnover at OpenAI these days. He is now the latest senior executive to leave the company.