（预览）OpenAI 驾驭世界：基础设施建设与无边野心，更多关于 Sora 与创作的内容

Hello, and welcome to a free preview of Sharp Tech. Hello, and welcome back to another episode of Sharp Tech. I'm Andrew Sharp and on the other line, Ben Thompson. Ben, we’re about 8 hours out from Game 4 of Brewers-Cubs. How are you doing? How are you feeling right now?

A little anxious.

Okay.

The Brewers probably should have put it away last night. Some real missed opportunities, particularly in the 8th inning. Caleb Durbin, you could have dived and scored.

Mm.

Fortunately, the Cubs still suck. So, as the saying goes, hopefully we’ll pull it out. But yes, it would’ve been nice to have this wrapped, dusted, and done with. It’s not great that our alleged number-two pitcher, who I wanted to start Game 2—and I guess I was wrong to want that—was knocked out in the 1st inning.

Oh, boy.

So yeah, we’ve got Fastball Freddy, man—Freddy Peralta. He’s going to carry us through. Confident.

Okay, I like it.

Somewhat confident.

They looked great. They looked great in the first 2 games. I was like, “Wow, maybe the Brewers could actually do this.”

Just a scrappy underdog, smallest market in the Major Leagues, doing the best we can.

For 3 or 4 months, you’ve been watching the Brewers every single day, so I hope, for your sake, they can at least make it to the NLCS. Then it really is Schlitz versus glitz with the Dodgers on the other end. I don’t have any faith in the Phillies’ ability to win that series, so Dodgers-Brewers would be pretty fantastic.

We have a lot to get to today, and it’s sort of been OpenAI week at Stratechery. Would you agree with that?

I was just looking up Major League payrolls. The Los Angeles Dodgers are number 1, with a total payroll of $350 million.

Yeah.

The Cubs, number 5, at $212 million. So we’re already up against it.

The mom-and-pop Brewers—

The mom-and-pop Brewers, with $122 million, half that of the Cubs. So yeah, we are the underdog. We’re lucky to be where we are. Everything is a gift.

Okay. Well, Brewers against the world. I’m right there with you, arm in arm. Let’s go do it. Let’s shock the world.

But as I said, it’s been OpenAI week at Stratechery, and we’re going to begin with a note from Roshan. He says:

“Ben and Andrew, I think it’s time to officially expand the Mag Seven to a Mag Eight and include OpenAI. They directly compete against Google, Meta, and in the future could compete with Netflix in content, Apple in devices, Microsoft in operating systems and APIs, Amazon in shopping and APIs, NVIDIA and Tesla in robotics. It’s also time to add Sam Altman to the pantheon of great business leaders and value compounders: Buffett, Gates, Jobs, Bezos, Musk, Zuck, Larry and Sergey, and Jensen, and now Sam.”

So, Ben, with that take as the premise, you have written several thousand words about OpenAI this week. You’ve interviewed Sam Altman. It’s a broad question, but how have you updated your thinking on what this company can be?

First of all, I think Roshan needs to relax. Let’s save the adulations for later. It’s one thing to speak to what you’re going to do, even to sign deals—

I was going to say—

To what you’re going to do.

Let’s let him stick the landing before we put him up there with Jobs and Gates.

This is a little bit like how we’re glorifying founders based on funding rounds. This is that on steroids, okay? Yes, there is a trillion dollars’ worth of infrastructure deals that OpenAI has signed in the last month. Let’s have that trillion dollars of deals actually be substantiated and actually paid for, and then there’ll be plenty of time to write the history books.

Yep.

I think the update is that it’s really a fascinating time. I was around in the dot-com era. I was a teenager and then in college, so I was sort of viewing it from afar. I wasn’t in the Valley. I wasn’t in the middle of it.

One of the interesting things about writing Stratechery is that you feel, particularly at this stage in the evolution—12 and a half years in—that lots of people pay attention and care what you think. There’s a certain sort of weight of expectations, or just a weight of everything you do feeling a little more weighty.

Yeah.

And then you remember you’re just writing a blog and no one cares, and you’re actually more invested in the Brewers winning. But it’s interesting to feel these dynamics again, not from the inside. I’m not in Silicon Valley. I’m not in OpenAI, but I do feel them in the ecosystem to a much different degree than I did 25 years ago.

I’ve been saying all along that you know it’s really a bubble when everyone sort of gives up on it being a bubble. That was an intellectual statement, but I wonder if it’s a—

When you say everyone gives up on it being a bubble, they say, “No, it’s not a bubble. This is all working and all rational, and we’re moving ahead.”

That’s when to be concerned.

That’s right. It is different this time. Exactly.

Yeah.

And I’m not sure that’s exactly right, based on how I feel right now, which is that I’m feeling concerned. This is probably bubbly, and there’s also a sense that you really have no choice but to go along with it and just see where it takes you.

Yeah.

It makes me want to go back and talk to more people, particularly people in the markets or in senior leadership at that time of the bubble, to see if they had this sense that they never lost the knowledge and reality that this was going to end badly. It’s just an acceptance that that’s the way it’s going to be. You’re swept along with the tides of history.

This is a broad, roundabout way to answer your question. To take a specific example, I’ve been critical of OpenAI and its API. I’m like, you never have enough compute.

Mm-hmm.

Our issue with the GPT-5 launch is that you really sort of nerfed the Plus level of subscribers, or at least you made them feel like they were nerfed. You drastically reduced the number of thinking queries they had and said, “Go to regular GPT-5. It will think when it’s the right time to think,” using this router technology. Everyone feels kind of screwed, and there were a lot of complaints about GPT-5 that I don’t think were, that I think OpenAI should’ve ignored.

But this specific issue, where you have paying customers who feel like they paid for something and had it taken away from them, is a big problem. That’s one of the challenges of a paid business model. If someone’s getting it for free through ads, it’s like, okay, fine, we’ll give you a refund of $0. Deal with it.

Right.

Google could get away with that with a lot of stuff. OpenAI—this is one of the challenges of a subscription model—is that there is an implicit sort of promise that can really blow up and resonate on social media or wherever it might be if you don’t live up to it.

Why are you devoting so many of your resources to this API business when Microsoft is your partner? They have the exact same API in Azure, they’re good at serving enterprise, so partner with them and you can sort of do it.

Right. With the GPT-5 launch, the reason they had to limit and nerf some of the thinking queries for Pro users was because they had limited resources—limited compute resources.

That happens just constantly, right? They’re constantly short on resources, and who knows what technologies they have internally that are not released publicly. They’re starting to get a bit of the Apple problem.

Mm-hmm.

With Apple and the iPhone, there are things they can’t launch because they have to be able to make them at scale.

Yeah.

Actually, I wonder how much that’s the case with the—

They’re serving 800 million weekly users at this point.

Right.

Like, the scale is pretty astounding.

And so if you roll it out, it’s like, you just have it—it’s, everything now is a step change in terms of necessary capacity. You don’t just get to—

Mm-hmm.

—gradually sort of ramp it up. I mean, I wonder if this is the case with the Air too a little bit. I feel like we’ve gotten inundated with more iPhone Pro commercials than ever. Maybe it’s because the Brewers are in the playoffs, so I’m just really locked in on TV these days. I swear I see an iPhone Pro commercial every other commercial. I haven’t seen a single commercial for the iPhone Air at all, which is the greatest iPhone Apple has ever released.

Maybe you need to hold it in your hand to appreciate that sense. Maybe it’s also the case that, from Apple’s perspective, the iPhone Air being a slow burn—being something that sort of starts seeping its way into consciousness—is a good thing, because maybe they wouldn’t be able to make enough of them if they made as many as they can of the Pro.

Mm.

The inventories in the store don't support that, to be clear. But I can imagine that they probably had the Air idea for a while, but it took a while to feel like they could confidently ship it and serve it and all those things. They have volume. Now, that might be totally wrong. Maybe it's super trivial, and I think there's probably still inventory at the Apple Store right now.

Yeah.

It's just to illustrate the point about ChatGPT and their capacities.

But it is interesting. They're not actually pushing it all that hard with the Air.

Not at all. There's no pushing.

But that's a digression. But yeah.

Yeah.

I have not seen a single Air commercial. Every time I talk to you about the Air, though, it feels like I'm getting an iPhone Air commercial, so I'm all set.

But that was why the Air is amazing, because that's what Apple used to get all the time, right? Any Apple customer would be raving about their products at every opportunity.

You gotta try this. Yeah.

Oh, still love it. Still in love. I love it. Even though I've had a few baseball games since then, if you've noticed the lack of cameras, that's fine. I'll survive. But if you have small kids, like you, you should have a Pro. It's totally worth it. So you obviously should get a Pro, you should get one every year, because it—

There you go.

It just makes so much difference.

I'm resting easy.

Battery life is still off the charts with my Pro over here.

As another aside, as part of the home-screen reorganization, I've had Twitter in my dock for ages, and I stuck it in a folder. Man, I should have done this a long time ago. It's incredible.

Healthy choice.

Yeah. It's so much less of a just-one-tap thing that I'm sucking away time on Twitter. It's actually pretty amazing. So, I endorse that. Dude, it was replaced by ChatGPT in the dock.

Okay.

There's a rational view of this API argument for normal businesses that is correct. And I went back to Microsoft, which I think is an interesting analogy. In my career of writing about Microsoft, it has always been in the context of Windows being eclipsed by the smartphone, then the necessity of Satya Nadella demoting the importance of Windows internally at Microsoft. He actually split up the Windows team. That team actually came back together a couple of weeks ago. I haven't mentioned it in the daily updates. I haven't had a chance to get to it. But it's fine now, because no one at Microsoft thinks Windows is the future of the company, right? There was a necessary interregnum where, yeah, you're so historically powerful internally, we have to split you up so you can't be running a coup on the rest of the company.

Mm-hmm.

And now they've accomplished that. It's clear what the future is, and actually, it is useful to have you all together, so let's get you all together there.

Terrific.

And I wrote about this early with Microsoft. I left Microsoft on June 30th. On July 2nd, I was out with a series of 3 articles on how Microsoft was just screwing up. Steve Ballmer didn't know what he was doing, in part because he needed to make a choice. He had the context that Windows was everywhere and could do everything. That was downstream of Windows' dominant platform position. They were an enterprise company. They got the consumer market for free. They were dominant horizontally, vertically, everywhere. They had the whole thing.

Mm-hmm.

And he was in the middle of this big push on devices, this vertical business model that didn't make any sense to me, and I'm like, “No, you need to focus on services. You have to serve everyone. You have to serve iOS. You have to serve Android. Your focus has to be the cloud,” et cetera, et cetera. Of course, I was completely right. Satya Nadella completely agreed with me. He was in the services business. He knew what was going on. That was the trend—the company had to make a choice.

Yeah.

And I think an error I've made is insisting OpenAI make choices, and they're not in that stage of a company. AI isn't in that stage of technology. It is in astronomical, massive growth. There was a time you could have made the argument in the ’80s or ’90s that you needed to make better choices in your software development because processors aren't slow enough to do what you want, and the answer was, “That's a waste of time. Processors are getting faster. Assume they'll get faster and develop accordingly.”

The analogy to OpenAI is: yes, you're low on capacity. You're going to be low on capacity no matter what you do. Go for it all and trust the capacity's going to come. That gives you not just the permission but the argument to go and make these infrastructure deals. You're painting this massive vision. And this is why OpenAI is not a Mag 7. This is the startup mentality writ large at an astronomical level. You're not selling your financial results. You're selling a dream and a vision of the future and asking people to fund it.

Mm-hmm.

That is where they are. It's just weird because they're so big and so impactful, which speaks to why they get the money. You can make the Mag 7, Mag 8 argument, but that's not how they're operating. They are operating like a startup, and it's pretty thrilling to watch. It also has that sense of, “This is very well gonna end up in a disaster.”

Crazy.

But whatever, we're all in. Here we go.

Well, and it's kind of dizzying to keep track of because, watching over the past couple of years, there was that initial phase when ChatGPT launched, took the world by storm, and the immediate reaction was, “Oh, wow, this company is the first threat to Google in 15 years, and maybe they're going to be competing with Google a couple of years down the line.” But then over the last 1.5 years—

That's what they do with every one.

It's like, yeah, exactly. We see the ambitions broaden every 3 to 4 months in various ways, and now I'm not even sure how to contextualize OpenAI in the context of big tech companies because they're trying to do everything. They're doing hardware. They're competing with Google. They're competing with Meta. They're doing entertainment. All of it. And now they've also integrated apps into ChatGPT, and it's like its own little OS layer. All of it is pretty insane to watch. And I wonder whether they’re trying to do too much and should be more focused. But it sounds like, for right now, just let it breathe—

No, this is what I mean.

—and see where the adventure takes us.

This is me leaning into the bubble: no focus, do it all. But the thing is, a lot of people are characterizing all these deals as them throwing spaghetti against the wall. I don't think so.

Mm-hmm.

I think it's remarkably cogent and clear if you expand your aperture sufficiently to see what they're going for, which is to be, yeah, the agentic AI layer for literally everything. That's why—

Yeah.

—I called it the Window. And maybe that was understating it: it was the Windows of everything. You put the WeChat analogy in this rundown, and that's probably actually a better analogy in some respects, where we talk about WeChat in China. WeChat diminishes the iPhone's power because it's such a dominant app, and it has all the mini apps in it and all the ways you interact with everyone, the way you pay, the way you interact with government services. It's all in this one super app, and that's—

Mm-hmm.

—that is also a good analogy, and it captures the sitting-on-top-of-everything-else aspect. I'm actually kind of annoyed I didn't include that in my article. I think Windows is—

Yeah.

—a better reference point for people in the US, but there are aspects of the WeChat analogy that are more applicable. And yeah, go for it. Why not? I've given up. Go all in. Even in the back of my head, I'm like, “This is going to end badly. You should probably focus. You should do X, Y, Z.” It's like, nope, this is the stage of evolution here.

Everybody's rowing in the same direction here.

Yep.

Even if they're going over a waterfall.

They're making themselves the table where all the chips go in. And I also think there's an aspect of the aggressiveness here that might have the potential to starve out competition to a certain extent, where, for every marginal dollar of investment, would you rather put it in OpenAI or would you rather put it in Anthropic? Would you rather put it into OpenAI, or would you rather invest in Meta or whatever it might be?

Mm-hmm.

Would you rather put it in OpenAI, or would you rather put it in NVIDIA? That’s where the real value-chain battle is, I think. There’s a lot happening there. Sam Altman is very good at game theory. He’s very good at sketching out how this stuff all links and ties together, and he’s willing to risk it all.

Yeah.

And that’s exactly what’s happening.

I mean, you talked to Sam Altman this week, as I said, for your interview. You only talked to him for about 40 minutes, and you covered a ton of ground in that interview. I was very impressed.

Well, this is the third time I’ve talked to him, the second time I’ve talked to him one-on-one. I’ve learned my lesson. Usually, for an hour-long interview, I need to prepare 20 questions or so, which is probably good, and I won’t get to all of them.

Yeah.

With Sam, I had 30 minutes, so I knew to prepare about 40 questions. We actually got an extra 10 minutes, and I went through all the questions, I’m pretty certain.

Well, it was very solid, and there were a lot of different threads that you discussed that I wanted to follow up on here. I’m going to read some quotes from Sam, and you can just riff and take it in any direction you want.

On the infrastructure build-out, he said, “When we are done with this new set of deals and get to explain, ‘Here’s where we’re going and why,’ I think it will all make sense. People may say it’s too ambitious, it’s too risky, whatever, but at least why we’re doing the stack that we’re doing, I think, will make sense.”

As far as the stack of investments and announcements—

I mean, I’m confused about what else he’s going to announce. I thought that his interview, what I wrote this week, was pretty clear. Again, I think all these investments are very cogent. They are looking to be the AI for everyone.

Okay.

Whether in a business context or a home context, you’re going to have your ChatGPT. That ChatGPT is going to continually gain contextual knowledge about you and what you’re interested in. You can use these connectors, where you get your email and your calendar in there.

But this is the worst it’s going to be, as we say again and again. You’re going to have this companion. It’s going to be total lock-in because it knows so much about you that no one else does. You’re going to use it everywhere. Right now, if you have 8 billion people in the world, OpenAI’s goal is to have 8 billion personal assistants that are on their platform and that you interact with and use basically 24/7.

Mm-hmm.

Why 24? The AI is going to monitor your sleep and give you feedback about it. That’s what they’re building. It’s omnipresent, everywhere, with you on every device and every platform. Wherever it might be, they’re going to sit on top, and you’re going to pull them with you everywhere.

That is what gives them leverage in the market. To achieve that is going to require astronomical amounts of—

Unbelievable amounts of infrastructure.

Yeah.

And so what’s interesting—I mean, the most entertaining sequence was the investment from NVIDIA into OpenAI. $100 billion was committed by Jensen Huang, and then two weeks later they turned around and, on Monday, announced the deal with AMD, NVIDIA’s direct competitor. That led to a 40% jump in AMD’s stock, all part of the frothy atmosphere that we all inhabit right now.

But in terms of what they’re trying to do with—

By the way, did you see what Jensen Huang had to say about this on CNBC the other day?

No. What did he say?

He was like, “You know, I’m really surprised that AMD was so excited about their next chip. It seems surprising they’d give away 10% of their company for someone to buy it, but hey, it’s very clever.”

Just a great—

I did get the impression—

Great passive aggressiveness there.

I did get the impression from Altman that NVIDIA did not know about this deal before it happened.

Mm-hmm.

Well, strategically, they’re trying to lock in their pricing power, or OpenAI’s leverage, in terms of not being too reliant on NVIDIA.

Okay, so you have a value chain—

With that—

You have a value chain in anything.

In terms of not—

Right. Well, you have a value chain in anything, and the most profits, there’s... I wrote an article ages ago, “The Law of Conservation of Attractive Profits,” in the context of Netflix. That phrase, “law of conservation of attractive profits,” is a Clayton Christensen phrase.

It’s basically the idea that in a value chain, you have different suppliers. Say you have 10 companies in a value chain. Usually, one company comes to dominate that value chain by virtue of becoming the bottleneck in the chain. They have the key piece, and everyone around them has to supply them or has to be their customer.

Those people around them lose pricing power because they’re dependent on this one thing that is highly differentiated and can’t be competed with.

All right, and that is the end of the free preview. If you'd like to hear more from Ben and I, there are links to subscribe in the show notes, or you can also go to sharptech.fm. Either option will get you access to a personalized feed that has all the shows we do every week, plus lots more great content from Stratechery and the Stratechery Plus bundle. Check it out, and if you've got feedback, please email us at email@sharptech.fm.