Loading summary
Transcript28 lines
- [00:00]
Sierra AI
How businesses connect with customers defines their brand. Sierra is the AI platform for building better, more human customer experiences. Fast answers. No canned responses. No hold music. No frustration. Visit Sierra AI to learn more.
- [00:18]
Bel Lin
Welcome to Tech News briefing. It's Friday, August 1st. I'm Bel Lin for the Wall Street Journal. People now watch YouTube on TV sets more than on their phones or any other device. An average of more than 1 billion hours each day. WSJ reporter Ben Fritz tells us how YouTube came to dominate our TV screens and what that means for the future of content on YouTube. Then there's little debate that AI consumes massive amounts of power and those demands are increasing exponentially. One potential way to address the problem? Redesigning computer chips that power how AI actually runs. But first, YouTube started as a website to watch videos on PCs, but now it's a bonafide media juggernaut and it's beating out Hollywood in the process. WSJ reporter Ben Fritz joins Patrick Coffey to discuss how YouTube overtook traditional media to become the most watched video provider on televisions in the US earlier this year.
- [01:20]
Ben Fritz
Ben YouTube became the most watched content distributor on TV this year. Would it be right to say that the gap between that platform and other big names like Disney seems likely to keep growing?
- [01:32]
Patrick Coffey
Yeah, the trend is just a bigger and bigger gap between YouTube and second place, which is currently Disney. And keep in mind, when we say Disney, we mean the ABC Network, the Disney Channel, ESPN, Disney, Hulu, everything that Disney offers for video. YouTube is beating them all combined. And the same is true for all other media companies as well as Netflix.
- [01:56]
Ben Fritz
Ten years ago, YouTube tried to enter the scripted drama arena with Cobra Kai, but at the time, at least, that approach didn't work. It was very telling to me that that show only became a hit when it moved to Netflix. What's different about YouTube's new ambitions for original scripted material?
- [02:15]
Patrick Coffey
I think the big reason Cobra Kai didn't work on YouTube is it was behind a paywall. It was just you had to subscribe to this premium version of YouTube to watch it. And the vast majority of the YouTube audience are not people who want to pay. The YouTube brand does not pay for premium content. It's pay nothing and get this huge array of content that's all the way from the most amateur to professional, but not Hollywood caliber professional. So when people want professional caliber Hollywood stuff, they go to Netflix, they go to Disney, they go to Paramount, and Netflix has way more paying subscribers than YouTube Premium ever had. So I think that's really why Cobra Kai ended up working so well on Netflix because YouTube just doesn't have much of an audience of people who want to pay for Hollywood caliber content.
- [02:57]
Ben Fritz
In terms of measuring YouTube's share of overall viewership, you know, we have subscriber totals for people like Rhett and Link of Mythical Morning, but can we really compare their popularity directly to that of similar programs on traditional or linear TV like Jimmy Kimmel or Good Morning America?
- [03:16]
Patrick Coffey
It's very hard to do because people watch video in so many different ways now. And you can look. Certainly the number of subscribers to a popular channel like Rhett and Link on Good Mythical Morning or Hot Ones or Chicken Chop Date is way higher than the number of people who are watching an episode of the Tonight Show. But not every subscriber watches every video, obviously. And people don't only watch Jimmy Fallon on tv. They're all, you know, he puts lots of clips on the Internet and they watch him that way too. On the other hand, these YouTube shows are not necessarily meant to be watched each day. You know, people watch them over time. They go back into the catalog and watch things from weeks ago, months ago, years ago. But bottom line, you can say that millions of people are watching the most popular Talk shows on YouTube and the numbers are growing.
- [04:04]
Ben Fritz
Now, live sports is, of course, the holy grail of entertainment and of advertising right now. How is YouTube advancing in the fierce fight with other platforms like Amazon Prime Net, Netflix, and then the NBC Universals of the world?
- [04:21]
Patrick Coffey
YouTube is really betting big on sports. And in fact, since they've given up on the premium scripted content business stuff like Cobra Kai, they don't do that anymore. The one kind of content they're spending the most of their own money on by far is sports. Everything else they let their creators make and YouTube is hosted for them. But YouTube is in the sports business in two ways. They're already the exclusive home of Sunday Ticket. All football fans surely know you have to pay for that. But what YouTube's doing, which I think is really scary for traditional entertainment companies, is it's starting to put sports on its free service. So YouTube is making a bet that they can cover the cost of sports rights solely through their massive advertising business.
- [05:01]
Bel Lin
That was WSJ reporter Ben Fritz. Coming up, designing new computer chips might be one way of addressing the immense power demands of AI. The these chips, designed to be more efficient than the likes of Nvidia's chips for running AI, are already showing promise. That's after the break.
- [05:24]
Sierra AI
No matter the industry, how businesses connect with customers defines their brand. And today, connection starts with a conversation. Sierra is the AI platform for businesses that want to provide better, more human customer experiences. With Sierra, your AI agent solves problems fast. No endless hold music, no canned responses, please press 1. No frustration, just better customer experiences built on Sierra. Visit Sierra AI to learn more.
- [06:00]
Bel Lin
Nvidia is the undisputed leader in AI chips, the hardware that AI runs on and is trained by. But a new class of chips that are designed expressly for inference or running AI models is aiming to challenge Nvidia's dominance in the space. WSJ tech columnist and co host of the Bold Names podcast, Christopher Mims joins us to discuss how these new inference chips could address some of AI's demands for power as well as present an alternative to Nvidia's chips. Christopher let's start with the big picture. Why do we need better solutions for all of AI's huge projected power demands?
- [06:40]
Christopher Mims
The main reason we need new ways to do AI and rethink how we're building it from the ground up is that the projected growth in the amount of power that AI requires is pretty unsustainable. So one estimate has it that 50% per year increase in the amount of electricity required for AI is a reasonable scenario through 2030. And we're already at the point where companies like Meta and Google and others are building multi gigawatt data centers and 1 gigawatt of power that's like a pretty decent sized coal fired power plant. So Meta has said, hey, we're going to build this data center, it's going to be as big as Manhattan and someday it's going to take five of those power plants. And Google has said we're going to need nuclear power, we're investing in fusion, we're going to, we're investing in exotic ways to store energy so we can make this work in Europe where energy is even more expensive.
- [07:35]
Bel Lin
Let's focus on one solution, which your column focuses on the idea that these new AI chips could use less energy than the ones from Nvidia. Tell us about these chips.
- [07:45]
Christopher Mims
So there are a dozen startups at least coming up with new AI chips that deliver AI. And going forward it seems like that's where all the demand is going to be, or most of it. And so in addition to these dozen startups, Microsoft, Amazon, Google are all working on custom chips that will go into their own data centers that will power their AI tools and also allow them to sell AI compute to outsiders. So Google, for example, just announced a deal with OpenAI, where OpenAI is going to power some of their AI on Google's new data centers, and presumably on the chips that Google makes that are more efficient at delivering AI.
- [08:29]
Bel Lin
And why are so many companies, you said almost a dozen or so, or possibly more, are building these inference chips. Why is the race so heated?
- [08:37]
Christopher Mims
There is huge demand. There simply aren't enough chips being produced by Nvidia, for example, to deliver AI at scale. Also, there's just this huge problem with Nvidia's new class of chips. They're so power hungry that you have to basically do a gut renovation on old data centers. So some of these new companies, the value proposition is, hey, you can deliver inference. You can deliver AI using our chips, using existing old style data centers where individual racks use less power than what Nvidia says you should have if you're going to use their latest chips.
- [09:17]
Bel Lin
So it sounds like you can retrofit an existing data center and that might save a lot of resources, time and energy when building out data center infrastructure.
- [09:27]
Christopher Mims
Yeah, absolutely. As one data center hardware guy put it, he's head of hardware for Cloudflare and they deliver about 20% of the Internet. He said we're going to need more efficient chips, more efficient ways of doing AI in general, so that we can get this growth in power consumption under control.
- [09:45]
Bel Lin
Right. And let's talk about Positron, this startup that is building inference chips and is working with Cloudflare and the source that you mentioned there. What's unique about their approach to building these inference chips?
- [09:57]
Christopher Mims
Positron is a startup that has taken the approach of what if we design chips that were only for delivering the kind of AI that people are most likely to use nowadays? So the kind of AI that makes chatbots possible and other things that are based on what's called a transformer architecture. And so they are taking chips from existing providers and they program them in a way that allows them to be super efficient, and they're just for AI. And so by stripping down what they can do and really focusing it on AI, they can deliver AI much more efficiently.
- [10:32]
Bel Lin
But one sort of big caveat which you call out in your column, is that even if chip makers and designers can make these more efficient inference chips, why is energy production the biggest bottlenec?
- [10:44]
Christopher Mims
The problem is you can build these huge data centers, you can fill them with chips. Where do you get the power from? When these companies hook up to the grid, they can't instantly demand all the extra power they need to run their data center. So they're having to produce power on site from fossil fuels in order to get enough power long term, they're gonna have to work with local utility companies that's already increasing electricity bills for everybody who pays into that utility. So for everyday folks as well. So there are a lot of challenges and no good solutions right now in terms of getting enough power for AI.
- [11:16]
Bel Lin
That was WSJ tech columnist and co host of the Bold Names podcast, Christopher Mims. And that's it for Tech News Briefing. Today's show was produced by Charlotte Gartenberg. Logging off for the weekend. I'm your host, Bell Lynn. Additional support this week from Zoe Culkin and Julie Chang. Jessica Fenton and Michael Lavelle wrote our theme music. Al our supervising producer is Melanie Roy. Our development producer is Aisha El Moussleem. Scott Salloway and Chris Sinsley are the deputy editors. And Falana Patterson is the Wall Street Journal's head of news audio. We will be back this afternoon with TNB Tech Minute. Thanks for listening.
- [12:00]
Sierra AI
No matter the industry, how businesses connect with customers defines their brand. And today, connection starts with a conversation. Sierra is the AI platform for businesses that want to provide better, more human customer experiences. With Sierra, your AI agent solves problems fast. No endless hold music, no canned responses.
- [12:21]
Patrick Coffey
Please press 1.
- [12:22]
Sierra AI
No frustration, just better customer experiences built on Sierra. Visit Sierra AI to learn more.