Jul 23, 2025 ยท 2025 #27. Read the transcript grouped by speaker, inspect word-level timecodes, and optionally turn subtitles on for direct video playback
Edit labels for this show, save them in this browser, or download a JSON override for the production folder.
Human Transcript
Timed transcript
Blocks are grouped by speaker for readability. Expand a block to inspect word-level timing.
Speaker 2
Hello, everybody. It is Saturday, August the 9th, 2025. We are back with tech. We never left it, of course. Last week in my weekly roundup with my old friend Keith Teer, I led with the headline, Tech Insider Claims Open AI Will Be Worth $10 Trillion. Has Silicon Valley Finally Gone Bonkers? And in my own editorial, I began by saying I've always considered my friend Keith Teer a bit weird, but maybe this week he lost his mind speculating that OpenAI will be worth $10 trillion. Well, in the last week, of course, OpenAI has not only launched its new GPT-5 software, which many people, including critics like Gary Marcus, seem impressed with, But there's also news that they are organizing a $500 billion valuation for a new round of financing. So, Keith, that should shut me up, shouldn't it? I'm the weird one, not you.
Well, it's half a trillion. Let's think of this as year two in the life of OpenAI, and it's a half a trillion. And I think that's a 10x on where it was a year and a half ago. If it does 10x again, it's going to be at 5 trillion. And who would bet against OpenAI doing 10x from here on? Because what they did this week is they set up their platform for the next phase of impact. And much more commercial impact as well, actually, in terms of what they ship this week. So one would expect their revenue, which I think currently is around 13 billion, probably gets well over 100 billion in the next year and a half.
So our difference, and it's less of a difference, I actually am beginning to think that you, and I don't like admitting it publicly, but you're correct here and I'm probably wrong, is I've always assumed that these AI companies, and particularly OpenAI and also Anthropic, they're pouring ridiculous amounts of money into their platforms, which are still not profitable, to generate the kind of revenue to get these kind of valuations. But it's... standard internet game isn't it amazon did the same thing and now they're incredibly amazon's a very profitable company in the future presumably open ai even if it's not profitable today will become one of the most profitable companies in
the history of the world yeah the way you think of it as an investor is um you you you ignore expenses By expenses, I don't mean the cost of sales of doing what you do. I mean expenses, like people, offices. You ignore expenses. And in the case of OpenAI, most of what it spends are expenses because it's building infrastructure.
Yeah, you ignore all of that. And what you focus in on is what's called the gross margin. That is to say, what are you selling? How much do people pay for it? And in terms of unit economics, what percent of that represents future profit? And in the case of OpenAI, it's roughly 90%. In the case of Amazon, it's as low as 20%. So gross profit is the future capability of being profitable. Then what you want to do is build the engine to be as big as possible. Think of it as a printing machine for money. You should spend as much money as possible on the printing machine. And at some point, you turn the expenses off because you're done. And now it's pure profit. So you're basically timing profitability at the point of peak profit. growth. And that's the right thing to do. So early profit actually would be mismanagement.
Early profit is mismanagement. That may be our headline today. And of course, this $500 billion valuation, as you suggested, is just the beginning. Altman knows what he's doing, though. I mean, I'm always a bit suspicious of him because he's such a player, but he's a very good player, isn't he?
Yeah. And the way they played it this week, by the way, it wasn't a great week for them in some ways. Their presentation when they launched it was pretty amateurish. They were trying to be Apple-like and they failed. Their demos took a long time and didn't really nail the points they were trying to make as well as they could have done. And
They've changed their algorithms so that this ChatGPT5 is very, very good at some things that the previous versions were not good at, but it's bad at some things that the previous versions were good at. So it's a bit of a mixed bag, but the core of what they delivered does set them up for massive revenue growth.
So where's the risk? If you're an investor, Keith, and you are an investor, And someone said, well, we want $5 or $10 billion at the $500 billion valuation. Seems to me the biggest risk is that the product will become commodified and be given out for free by some Chinese company, which would undermine their future revenue. Is that a risk?
These Chinese startups already exist. They already exist. What they don't have and what OpenAI does have is they've captured the consumer mindshare associating their brand with artificial intelligence. So very few people outside of the Silicon Valley echo chamber, and by that I include Chinese entrepreneurs, very few people will choose an AI different to OpenAI's ChatGPT. Well, I've chosen, or Anthropic, because you would include them, right? Even Anthropic, it's kind of people like you and me who are somewhat insiders who know. It's not going to be... You said I was normal, Keith, not an insider. You're a normal insider, but... That's my brand.
Exactly, exactly. Or a normal insider. But, you know, it's pretty clear that OpenAI has the mind share of, you know, more than a billion people. Anthropic probably has the mind share of 100 million. Google probably... It doesn't have Mindshare at all of consumers.
I have to admit... that I'm becoming increasingly, I've always been vaguely bullish on AI, but last weekend I was in New York with my son who 20-something graduated from NYU. He says all his friends are entirely addicted to AI. We're getting a lot of pieces on AI and you always cover them in your excellent newsletter, but I wonder whether journalists are really beginning to understand the impact that AI is having on AI. everybody that it's changing our lives well in a way that's the subject of the
editorial this week which is we massively underestimate how the world's going to change due to where AI has now gotten to. The word software is going to mean something completely different to what it means today. The word hardware is going to mean something completely different.
Right, so the title of your newsletter this week is The Week Everything Changed, although that seems to be every week in Silicon Valley. AI native software and hardware is here. But it sort of turns the concept of software and hardware upside down, doesn't it?
Yeah. So, you know, I was meeting with an entrepreneur yesterday who's building something using AI in the education space, focused on tutors and his mental image of his software was a website where you could choose a tutor, a bit like Khan Academy, but where the tutor was an AI. And I said to him, why will a webpage be the place to deliver an AI tutor, surely you'll deliver it through a chat interface that has access not just to math tutors and physics tutors, but also has access to therapists, consultants, you name it. So software is going to be delivered in an interactive spoken or typed interface where you don't even see the software. It doesn't have a visual form.
It's like water, I mean, or like electricity. And of course, the electricity metaphor is one we've used a lot. I'm quoting you here from your very interesting editorial this week. What if the defining assumption of modern tech that software is something humans write and others humans use and hardware is something we buy and use just became wrong?
This stuff is... How would you describe it? You talk about... It's beginning to happen. Software has become... I'm just quoting you here because you put it nicely. Software is becoming an actor, not an app. And hardware is going to have to have that embedded. Imagine a child talking to a toy and getting answers back. For some people, that would be chilling. For others, it's very exciting.
Yeah, I mean... replace the ipad as child uh friendly pacifier which a lot of parents use these days with you know a a doll or a fire truck or whatever toy you want to think of where the child is fascinated by it because it talks back and and the physical world will be turned into an interactive place using ai embedded in objects that is a very different world to the one we live in today, but it's coming up very fast now.
Just one point. Tesla just did the first delivery of a new Tesla to a buyer without a driver being in the car. The car brought itself to the buyer and introduced itself using Grok.
So you've got this interesting piece about the Stone Age, Welcome to the Stone Age, and you work off that in your editorial. Are we in a new, so to speak, Stone Age, Keith, or is it post-Stone Age?
Well, that particular article is quite clever because it says, when did humans start to progress? And it uses the Stone Age as a metaphor, but also historically correctly as the tipping point between not using tools and using tools. So if you think of AI up until this week as being a little bit like a human that has no tools, what happened this week is AI got access to tools and without you explicitly having to tell it to, it just knows it has tools and it uses them when it decides it needs to. And of course, when humans started using tools, farming could happen. The whole of progress since then is predicated on tool use and better and better tools requiring less and less humans to produce more and more things. Well, AI just got to that tipping point between the stone age and the tool age. And the article kind of is very good at capturing that and crystallizing it. And I think that is the key. Now, ChatGPT has always had the ability to write Python code and run it. Now it can use spreadsheets, it can write Word documents, it can do queries against databases. It's got an unlimited number of tools, starting with a small set that were released this week. That means it can do things, hence the actor versus an act. It can do things, and so you can direct it to produce outcomes using whatever are the appropriate tools for those outcomes. Once it's in the physical world, those tools can include real-world tools.
Yeah, and of course, the science fiction writers will start talking about, and they've already written many books about imagining these tools having minds of their own. I was just in Africa a couple of weeks ago, went to the Kenya Museum in Nairobi, which showed those early, the skulls at least, of those early Stone Age people. There must have been some... of our prehistoric species, Keith, who reacted against the tools. What happened to them?
So they probably all thought to themselves, we don't need tools. And we have those types today. Yasha Munk writes an interesting piece in Persuasion about what he calls the peculiar persistence of AI denialists. And then there's a piece in the New York Times this week as well about some college professors who will not, and I'm quoting the Times here, bow down to AI, whatever that means. I mean, these are the people who are just going to die out. AI denialists and college professors who won't even use AI themselves or allow their students to use it. I mean, these are extreme reactionaries, aren't they?
Yeah. So it's fine to be a writer, let's say, and enjoy doing it and resolutely not use AI tools. Society will give you the space to do that and may even reward you for doing that. So it isn't like Stalinism where everyone must get in line. But the ones who do get in line and use the tools are going to prosper, certainly economically compared to everyone else, because they'll be more impactful.
And we've had every day there's a new article about how AI is coming from one group or another. The journal wrote a piece about existential shift, which is really an existential crisis, and McKinsey. I mean, do you think these traditional consultancies for which AI essentially undermines, can they even exist in the age of chat GPT?
They have to change their function for sure. I once worked with Boston Consulting Group on a project at VeriSign back when... just before I met you, Andrew, at Santa Cruz Networks. And if you really were to strip away what did Boston Consulting Group do, they ran models in spreadsheets, they did industry analysis to look at the competition and the opportunities, and they presented a narrative from those things. to a management team who had the option to give a thumbs up or a thumbs down.
Yeah, in my sense, I mean, my experience with these consulting groups is they charge you an enormous amount of money to tell you what you already want to know.
Well, I can tell you clearly, I was getting paid about half a million dollars a year by VeriSign for running the project. Boston Consulting, who did that work, were getting paid probably 20 million on a contract. And everything they did can now be done by AI already. So if they're going to continue to be useful, they're going to have to change what they offer, what the value is, because the management team can now use AI to replicate them.
Yeah, I don't think anyone's going to be shedding any tears over McKenzie or Boston. Although you have another interesting piece in the FT about the risk of letting AI do your thinking. On the one hand, of course, it undermines the work of people like McKinsey and Boston Consulting Group. On the other hand, it also adds value to human thinking, because once we find the kind of thinking that these machines can't do, they become particularly valuable, don't they, Keith?
Yes. At a personal level, I use an app that I built using AI for producing my newsletter. It's called Creator Automation. If anyone wants to play with it, you can. It's free, and it's at creatorautomation.ai. And I built this week an editorial writing component of it to see what would happen. And I did it because of ChatGPT5, which is, I think, a lot better than the previous ones. So this tool, I told it, go and read my last five editorials and look at the content that that editorial sat on top of. Then go and look at this week's content and try to write a Keith Teer editorial in line with this week's content, And it did it. It was about 100 times better than any previous editorial I've ever got an AI to write, but it still definitely works.
Yeah, hence why I agree with that title, Don't Let AI Do Your Thinking. AI is very good at being your servant, or intern or co-pilot, it is not good at replacing thought. And this goes back to a theme we've talked a lot about on this show, Andrew, which is the people at risk with AI are specialists who have specialist skills like mathematicians, scientists, not the inventors, but the kind of the ones who understand the rules of a specialism. The ones who are not at risk are creative generalists.
And if you add more audacity into your normalness, AI can't be an audacious path breaker yet. That may change, but at least for now, it's constrained by what it already knows, and it doesn't do a good job of thinking outside the box. But if you think outside the box, it can do a great job of helping you.
Yeah, I think it's very well put. So we want more audacity. But you've also got to be daring. You had an interesting piece, actually, speaking of daring, from Daring Fireball, changing the subject slightly, making fun of the valuation of Substack at 100 million. Of course, that's that pales in comparison to OpenAI's $500 billion valuation. But John Gruber at Daring Fireball, who's the ultimate Apple fetishist fanboy, he's not a big fan of Substack. And there are others as well. What's wrong with the latest Substack $100 million round? At least according to Gruber, I know you're more bullish on Substack.
They probably didn't pay him to come on the platform.
Words and timings
Theyprobablydidn'tpayhimtocomeontheplatform.
Speaker 1
Yeah. and I don't know if Hamish McKenzie at Substack will listen to this, but for what it's worth, I think I side with Substack. But I do understand what Gruber's getting at. He's making the point that Substack is an umbrella brand that is in the sense a walled garden, and it forces the author and the brand to be contained inside of the Substack brand. And Substack... pulls most of the strings when it comes to who gets to read what, its algorithms. And it made a mistake last week in which its algorithms promoted what, I haven't read it, so I don't know if this is true, but it was claimed it was a Nazi publication. Oh, what?
And they apologized for it and said they've changed the algorithm. And historically, Substack always said it doesn't have automated algorithms creating feeds. And it kind of showed that it does. And so there was a bit of a backlash, which Gruber led. And so the context here is, should publishers publish on platforms like Ghost, which released a new version this week, which is truly independent, and you pay them a fee for using the platform, or should an author be subsumed inside another brand like Substack, where you give them a share of your subscription revenues? And that's an ongoing debate. What nobody contests is that authors need to have their own independent presence and directly be speaking to their audience.
Everyone agrees with that. Yeah, my sense of Substack is they're trying to be half pregnant. They should either go one way fully or the other. Given how bad I think the mainstream publications are in New York Times, or Washington Post, all the others. I think the Substack has an opportunity to actually become the New York Times of the future if it was curated properly. They've got so much incredible content on. If you could find, if I would love to go to a Substack front page every day where I have trusted editors who will pick out the very best stuff published that day or that hour on Substack, and I would follow it much more loyally, I think, or much more aggressively than I would follow the New York Times, which is so predictable and annoying.
Yeah, the other thing with Substack, which is weird, is that I use it. I don't really care. I mean, people can pay me for subscriptions. I'm happy to take their money. But it's much less interesting to me than the attention you and I and my other stuff gets. So I'm basically using their platform, their distribution, their video for free. They're not getting anything back from me.
Yeah, me too. But it's just, it's not the point. So, yeah. Anyway, finally, you don't have a post of the week, but there is an interesting post with one word headline. Enough, the math in the headlines. Tell me about this piece and why we might use it as our post of the week, even if you don't have an official one this week.
Right, which Andrew kindly explained to me who they were, because I am not a novel reader. So I'm the worst person in the world to understand writers. But that said, there's a point in this, which is happiness and money. You know, money can't buy you love, as the Beatles said, is kind of the core point in this article. And it's contrasting the absolutely enormous amounts of money sloshing around in the AI venture ecosystem, which I hesitate to even call venture, and the small amounts of money sloshing around in the creative ecosystem. And it's correct and staggering, but also completely rational, because you can't make a lot of money back in the writing ecosystem compared to how much money you can make back in the AI ecosystem. So by definition, money goes towards money.
Well, certainly one of the things about AI is going to make, or is making, just as it's making McKinsey consultants redundant, it's making editors redundant. God knows how editors will survive in the age of AI. Anyway, Keith, you've done a good job. Last week I accused you of being weird. This week you accused me of being normal, so we are... equal. Maybe next week we can find some other ways to describe each other and insult and compliment each other. Have a good week. That was the week that was. It was a remarkable week, the week that broke software and hardware. Maybe next week something else will happen. We'll talk again in a week. Have a good week.