Nvidia Goes Big on Buybacks as SpaceX Reaches Orbit

SPEAKER_1 (0:02)

Bloomberg Audio Studios, Podcasts, Radio, News. Bloomberg Tech is live from the heart of Silicon Valley with Ed Ludlow in San Francisco.

Ed Ludlow (0:22)

This is Bloomberg Tech coming up. Nvidia boosts the size of its share buyback plan by a record $150 billion, taking the total amount to $235 billion. Plus a milestone for SpaceX's Starship rocket, reaching true orbit for the first time after a brief scare due to an engine failure during ascent. And Meta taps MongoDB's president and CEO to lead a new AI platform for enterprise customers it deems as the next major pillar of its business. Let's get right to today's big number, $150 billion. That's how much Nvidia has added to its share buyback plan, taking the total remaining authorized amount to $235 billion, a massive number reflecting CEO Jensen Wong's confidence in its continued growth. The company expects to complete the program through fiscal 2028 According to a statement, they're currently in the third fiscal quarter of their fiscal year 27 At the same time, Nvidia says it has a new way to rein in rogue AI agents, introducing new security tools that can shut them down in milliseconds. The company even going so far as to say it could have stopped that recent hugging face breach, which involved OpenAI's models. Bloomberg's Maggie Eastland has the details and it's kind of in two parts. Explain them.

Maggie Eastland (1:39)

Absolutely. So this new product release has two layers. So first, there's OpenShell that is an open-source software program that can run on Nvidia's central processing units. And that's designed so that a user can impose certain guardrails and enforce those guardrails. Now, this was first announced in March back at their GTC conference being rolled out now in combination with the completely new products called Nvidia Sentry, which is another separate system that runs on Nvidia's data processing units. And essentially, it's an added layer that can, like you said, isolate an agent that's acting suspiciously, the company says in milliseconds.

Ed Ludlow (2:16)

The analogy that I've been going with this morning is like, it's putting an AI agent in a locked room where part one controls what that room looks like and the doors in and out of it. Part two is a security guard that's always outside. But it's kind of in line with Jensen Wong's school of thought that this is an engineering issue at the heart of the AI safety debate. Well, he's given us an AI engineering solution.

Maggie Eastland (2:41)

Absolutely, Ed. That's the line I've been using. Jensen Wong, one of the biggest beneficiaries of the AI trade, has really continually voiced that AI progress needs to continue and that the executives who want to slow down just need to find a solution to their problem. And now he's offering them the solution. Importantly, we don't know whether OpenAI and Anthropic are going to adopt this for their training runs specifically, which is where they've had the most problems with these rogue agents.

But if this really can solve the issue, it could really quiet some of these calls that we've been hearing for a slow down.

Ed Ludlow (3:14)

There is an element where each of those solutions runs on other parts of Nvidia's ecosystem hardware-wise. Just very quickly, what's the Nvidia School of Thought about what this means for the training runs that are happening in industry?

Maggie Eastland (3:30)

Yeah, I think the idea here is that Nvidia has said this potentially could have stopped the hugging face incident, which was an incident where agents got out during a training run.

If the technology works in the way that they're saying it works, it could enable OpenAI and Anthropic to continue with some of these training runs that they've been more hesitant about. We saw just last Friday, OpenAI said they were pausing training after new disclosures about some government-related incidents and a breach of an Australian government website. Again, this could be really a breakthrough if technology does work the way Nvidia is claiming.

Ed Ludlow (4:06)

Nvidia shares up 2% on multiple pieces of news. Bloomberg's Maggie Eastland on the AI safety component of it. For more, let's get to Erica Klauer, founder and CIO of Science and Technology Partners. And this is so interesting because this is Nvidia acting on their warning that this is an engineering problem, here's a solution. What do you make of that? Nvidia just offering that to industry.

Erica Klauer (4:33)

So good morning Ed, nice to see you.

It's such an interesting time once again for Nvidia showing and flexing its muscle of innovation. I think one of the things that's sometimes hard to remember is that over half of Nvidia's engineer are software engineers. These are engineers who are experts, not only in terms of software expertise, but also in cybersecurity. So it makes total sense that the company would be stepping in to alleviate some of the concerns around the cyber area.