Twitch is opting every streamer into AI training by default, and NVIDIA just made a $500 billion bet that could make or break the AI data center boom. This week, Ashley and Daniel break down how to opt out of Amazon's new Twitch AI training policy, then dig into why NVIDIA is personally guaranteeing the resale value of its own GPUs to keep investors comfortable, and why that move echoes a warning sign from the dot-com bust.
Twitch Will Train Amazon's AI on Your Content — Unless You Opt Out (00:40)
Twitch now trains Amazon's generative AI models on streamers' broadcasts, clips, VODs, chat text, and images by default. Streamers who want out have to manually turn it off:
Opting out only blocks generative AI training. Twitch can still use AI for captions, recommendations, sponsorship tools, streamer growth features, monetization, and safety tools like AutoMod.
Ashley and Daniel discuss why Amazon didn't simply pay streamers for their data, and how this mirrors the original, uncompensated scraping that built today's AI models in the first place.
Source: Twitch will train Amazon’s generative AI on your streams, clips, and chat by default unless you manually opt out. (Apple Insider)
NVIDIA's $500 Billion Plan Is Risky but Brilliant — Especially for Aging GPUs (02:43)
NVIDIA has lined up commitments of up to $500 billion from major investment firms — including BlackRock, Blackstone, and Goldman Sachs — to fund new AI data centers. To win their confidence, NVIDIA is personally guaranteeing the resale value of the GPUs backing those deals: if the chips used as loan collateral lose value and a lender has to liquidate, NVIDIA will cover up to 25% of the shortfall.
Daniel draws a parallel to Lucent Technologies during the dot-com crash, which financed its own customers' equipment purchases and collapsed when the bubble burst. He also flags the "wrong-way risk" this creates for NVIDIA: its financial obligations grow biggest exactly when demand — and its own revenue — is weakest.
Ashley and Daniel also cover CEO Jensen Huang's public comments to investors and bond markets aimed at calming fears about NVIDIA's exposure, and debate whether this financing structure is a bold, necessary innovation or a warning sign of a bubble nearing its tipping point.
Source: Nvidia’s new $500B plan is risky but brilliant, especially for aging GPUs (TechCrunch)
Ashley Coffey (00:05)
Hi and welcome back to AI News That You Should Know About for August 18th. I'm your host, Ashley Coffey
Daniel Hill (00:11)
And I'm Daniel Hill. There's a lot of AI news out there. We give you the weekly highlights very quickly. Today on the show, we'll be talking about Twitch training Amazon's generative AI using your streams, clips, and chat by default, unless you choose to manually opt out. And NVIDIA's new $500 billion plan involving aging GPUs. We'll be covering all of that on our show today, but first, here's a quick word from our sponsors.
Ashley Coffey (00:40)
Welcome back. Twitch will train Amazon's generative AI on your streams, clips, and chat by default unless you manually opt out. This is coming from Apple Insider. Twitch now lets streamers block Amazon from using their broadcasts, clips, and other channel content to train generative AI. However, it's opt out, so here's how to do that.
Ashley Coffey (01:01)
Users can find it under security and privacy in Twitch account settings, where it's labeled, quote, training for generative AI.
Ashley Coffey (01:08)
Turning off the setting prevents Amazon from using a channel's streams, videos, on-demand, clips, highlights, chat, text, and images to train future models. Twitch says the opt-out covers Amazon models that generate or synthesize text, audio, images, or video. The bottom line here is that opting out of generative AI training won't stop Twitch from using AI for other parts of its service. The company can still process content for captions, recommendations, sponsorship tools, streamer growth, and monetization features.
Ashley Coffey (01:37)
In safety systems such as Automod. Very interesting move here. What are your thoughts, Daniel?
Daniel Hill (01:43)
I'm really curious why Amazon wouldn't just offer to pay its users or give them some kind of credit if they want this data to use it for their own generative AI purposes. It's not like Amazon doesn't have billions of available dollars that they could share and maybe they give a free month or something. I'm very curious why this is the approach that Amazon has taken. Obviously, we'll not easily understand that thought process, but
Daniel Hill (02:08)
kind of surprising that companies don't offer money to their users in exchange for that data.
Ashley Coffey (02:13)
It's because it would set a precedent. I see why, but also these companies trained all of the content on the internet in the first place to train the original models and they didn't get paid.
Daniel Hill (02:20)
Right. Yes.
Ashley Coffey (02:23)
So
Ashley Coffey (02:24)
We have a problem to fix and I don't know where the solution is. We're gonna take a quick break. When we come back, we'll be talking about NVIDIA's new $500 billion plan is risky but brilliant, especially for aging GPUs.
Daniel Hill (02:43)
Welcome back. Continuing with our next story, it has to do with NVIDIA's new $500 billion plan involving aging GPUs. By way of background and going back to when the dot-com tech bubble burst back in the early 2000s, there was a telecommunications equipment provider called Lucent Technologies. It went up sharply in value, but then crashed. And the reason was it was lending its customers money to buy its equipment.
Daniel Hill (03:10)
I actually didn't realize this at the time. I didn't know that at all. In researching this story, I realized I learned that. So, in a very similar vein, NVIDIA announced this week that large investment firms, including BlackRock, Blackstone, Goldman Sachs, and others, are willing to commit up to $500 billion to build AI data centers. That's one part of the story. Here's the second part that's
Daniel Hill (03:35)
Similar to what I talked about with Lucent, to convince those investment firms, NVIDIA has agreed to guarantee with its own money that its chips used as collateral in these deals will retain their value. Now, I think we can all agree that hardware decreases in value over time. As new hardware is created, the older hardware is less expensive and less valuable.
Daniel Hill (04:00)
We understand this is a core concept of how technology works. This is financial maneuvering on the part of NVIDIA. They're promising that if the GPUs used as collateral don't retain their value as expected, the company will cover up to 25% of the difference. This article explains so if a data center owner defaults on a loan and the lender has to liquidate, but the chips can't command.
Daniel Hill (04:26)
The price that NVIDIA said they should on the books, NVIDIA will chip in to cover part of the difference. The danger here is for NVIDIA creating something that financiers call wrong way risk. So NVIDIA's obligations grow as demand weakens. So if this happens, the revenues are not going to be there either. I've
Daniel Hill (04:49)
Probably done a terrible job of explaining this, but hopefully I conveyed the concept. Ashley, what is your take on this situation?
Ashley Coffey (04:55)
I like that you picked this and I wanna hone in on the specific part about Jensen Huang, it's interesting that he had to go speak to investors essentially to help quell some of their risk.
Daniel Hill (05:07)
Jensen Wong is the CEO of NVIDIA. I don't think I said that previously, but yes, please continue.
Ashley Coffey (05:10)
Yes, Jensen
Ashley Coffey (05:11)
Huang is the CEO of NVIDIA and you mentioned that
Ashley Coffey (05:14)
The bond markets got so spooked that NVIDIA CEO Jensen Huang took to X and Business TV to better explain how NVIDIA's risk would be limited. I think this is a great idea. It's innovative, someone's gonna have to do something different these days. We can't continue on this cycle that we've been on. So there's no way for us to figure out if it'll work except for trying it. And what does NVIDIA have to lose? Yes, they have billions and millions of dollars, but they have the ability to take on risk, and that's the big thing here.
Daniel Hill (05:40)
That is true, but they are also taking on risk with money that they are borrowing. And I think that's concerning since I feel like when we look back on this time period, the way we look at the book Michael Lewis wrote, The Big Short, right? And that movie, I think this will be a critical moment in that story when we look back at things. That's my take on it, but who knows how it will end.
Ashley Coffey (06:01)
We're just really on that tipping point and threshold of that collapse. And when that happens, it's going to be a deck of cards falling.
Daniel Hill (06:09)
I agree. That is it for our AI news you should know about this week. The show notes have links to everything we've discussed. If you want more of this kind of content, that is what you can find us on social media for. You can find me on Instagram, threads, and YouTube at Daniel Hill Media.
Ashley Coffey (06:24)
And I am Ashley Rcoffey89 on Instagram. That's A S-H-L-E-Y-R-C-O-F-F-E-Y-89. And if you want the tech highlights without the twice-a-week commitment, my AI newsletter is monthly. You can find the link in my bio.
Daniel Hill (06:39)
Please follow the show on YouTube, send us a topic you want covered, and we will be back later this week for more AI news you should know about. Thanks for listening!