（预览）Meta 的叙事难题、XBOX 重启、Token 成本问答、美国足球与自然中的 Starlink

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, it’s great to be back. How are you doing?

I think it’s great to be here. It looks like your location has changed.

That’s right.

Summer surroundings. You have a beach with an umbrella over your shoulder.

In the back, just living life here, working remotely for the next several weeks. So yes, life is good for me. How was the fishing boat?

It was great. No complaints. Happy to be back, happy to be there. I’m just a happy guy, Andrew. What can I say?

Did the Starlink hold up for you while you were out there?

Of course. I actually have an addition to my tech stack aboard.

Uh-huh.

But I can save it. I know someone wants to castigate me for taking a Starlink.

That’s right.

I think we have a bit of a mullet section today, but I do have a new product recommendation that goes with the Starlink, which I will get to at the end.

Okay. Well, I look forward to that. We will be returning to the Starlink discussion. We do have a robust mullet section—party in the back—today. For now, we will begin.

I thought you wrote a fabulous article on Tuesday.

Oh, thank you.

You were surveying the current state of Meta generally, and you used the words of a version of Mark Zuckerberg who’s a bit better at telling his company’s story. So before we get to the substance, what prompted you to write that piece and approach it the way you did on Tuesday?

1. Meta’s Missing Story

There’s been a Meta piece to write for ages. It’s the regular Meta check-in. I have a very good track record of writing, “Actually, this company’s fine,” when they’re at their bottom.

Calm down.

I have found them very frustrating—Mark Zuckerberg specifically—and that frustration is a long-running one. This isn’t just a frustration over the past few weeks. This is a frustration literally over the last—as long as I’ve been writing Stratechery.

10 years.

Right.

Yeah.

I feel like it’s been particularly bad in the last year or year and a half, but at the same time, I’m not a Meta bear. I wrote a couple of years ago about how AI is arguably the biggest opportunity for Meta of anyone.

Mm-hmm.

I still think that’s the case, but there’s a real—I don’t know that I can fully come out with the argument, “Everyone outside is stupid. Obviously, this is a great investment.” I think it is, but there is a consistent pattern by the leadership that, at some point, had to be addressed. That’s what I wanted to address, and it suddenly occurred to me how to address it: what I want Mark Zuckerberg to say.

Yeah.

What I want him to say is downstream of what I want him to believe.

Mm-hmm.

And I’m not sure he fully does. That was what I was trying to convey in the article.

He doesn’t necessarily grok the ultimate upside and the clarity of the opportunity here.

No.

Like—

I think he does. I think he probably does, but to me, all the frustrations that I’ve had from the outside looking in with Mark Zuckerberg are the exact same frustration repeated again and again.

Mm-hmm.

It’s this desire to be something that Facebook isn’t. That’s what it comes back to.

Yeah.

Well, there is an implicit criticism of Zuckerberg’s leadership in the format you adopted because you were basically doing his job for him, at least as far as messaging is concerned. But I also found the piece broadly positive on the state of Meta—the overarching strengths they have.

That’s good. I conveyed how I felt.

Yeah.

I’m positive about their opportunity, and it’s not even fully saying that this is a—

Mm-hmm.

What I’m frustrated with Mark Zuckerberg about is not sins of commission. There are lots of people who are mad about sins of commission, things that he’s done that they don’t like.

Yeah.

My problem is a sin of omission, in that I don’t believe he has fully embraced what Meta is, and because of that, has not delivered the messaging that would give him the freedom to do what has to be done. By the way, that has led to some sins of commission, spending tons of money on stuff that was in the wrong direction. But yeah.

There have been mistakes. You know, reality loves—

We’re talking around it. Do you want me to give the overall thesis, or do you want to lead into it?

I was actually going to go a different direction as far as the thesis is concerned because I’m curious: if you were trying to make an opposite argument that Meta has misplayed the last 2 years, what does that argument look like? There has been at least 12, maybe 24 months of agita surrounding Meta’s efforts in AI—whether they’re spending way too much, why they’re spending on X, Y, and Z. I’m curious whether there’s any legitimacy to all of those concerns.

2. Llama Four Goes Off Course

I don’t know. It’s hard to know what’s happening internally at a company. Llama 2 and Llama 3 both seem to be quite solid for what they were. Then something totally went off the rails with Llama 4. A lot of the Mistral team was former Llama people, so they lost a bunch of people at some point.

Okay.

I think Yann LeCun is a legend in his field, but he was totally the wrong person for Meta in the LLM area. Even if LeCun is a skeptic about how far that technology can go in terms of AGI, that doesn’t matter for Meta.

Mm-hmm.

LLMs as they exist—as they existed 2 years ago, when I first wrote “Meta’s AI Abundance” about this massive opportunity they had—were already good enough to realize massive opportunities on Meta’s side. They live in a probabilistic world. They have infinite ability to do things wrong.

Just as an example of this, Meta’s digital advertising: they serve trillions of ads a year.

Mm-hmm.

Most of them don’t work. There’s a phrase with baseball players: “If you only succeed a third of the time, you’re one of the greatest players ever.”

Yeah. Hall of Famer, sure.

If you’re an ad service and your ads only succeed 0.2% of the time, you’re the best ad platform on Earth. You have so much scope to experiment and do stuff here. That’s one of the things that makes it so intriguing and a very natural fit, I think, for this probabilistic world of LLMs for lots of different reasons. I got into it a little bit more in my Daily Depth on Wednesday as well.

3. Meta’s Ad Engine Advantage

And so, the overall structure of Meta—where they are, what they’re doing, and the fact that they are resolutely a consumer company and should not be a B2B company—gives them certain freedom of movement and certain incentives that are actually positive, particularly relative to, say, an OpenAI.

Mm-hmm.

All these structural things are there. To that end, I have no problem with the investments they’re doing.

Okay.

I don’t have a problem with the CapEx. I don’t have a problem with all the hiring that they did. It’s existential, for one. They’re a digital company. Anyone who’s purely digital has to be deep in AI.

Mm-hmm.

Number 2, they have so many ways to use this. That’s the thing. Of course, they should be building it. They can capitalize it. It’s still not clear—it’s getting pretty clear—but the frontier labs: are they going to earn a return? Meta has all the mechanisms to earn a return in place. They just need to build the engine here. So why do investors not believe this? I think—

Well—

I think that is ultimately because of Mark Zuckerberg, because he doesn’t fully believe it. He hasn’t fully embraced the thesis we keep talking around: Facebook is the greatest ad company of all time.

In history. Yeah.

That’s awesome. I think their ads are a societal good. You don’t have to be ashamed about it. There’s a bit where Mark Zuckerberg just wants to build a platform. He wants to be what Facebook isn’t.

Mm-hmm.

Platforms and advertising are totally opposite because, with an advertiser, you’re featuring the advertiser. With a platform, you’re featuring third parties. They don’t go together. So many mistakes they’ve made over the years are a refusal to accept the nature of what they are.

Mm-hmm.

4. The Hardware Contradiction

This idea that you’re going to build hardware, like with VR, when you’re a company that serves everyone—if you want to connect everyone, limiting your service to hardware you sell, by definition, limits your market.

Apple is not a company that serves everyone. They're not a social network company. By virtue of their business model, they are fundamentally constraining their market. That's the nature of hardware. Why is Facebook spending hundreds of billions of dollars building a hardware business? And they've been doing it since 2012. All of it—none of it makes sense. Yes, they're mad about Apple restricting their access, but at the end of the day, Apple delivered the entire world to them on a continuous basis on a platter.

You know what? I've said Apple needs to be grateful to Facebook for building the App Store. How about Mark Zuckerberg being grateful to Apple for building the smartphone—

Mm-hmm.

—and creating his entire business? This solipsism here is also kind of irritating. That's a mistake. There's a mistake. The whole TikTok thing—I wrote in 2015 or 2016 that Facebook is more than just connecting friends and family. It is entertainment. It is a thing people do when they don't have anything else to do. That's why it's so good at advertising: the advertising is just more content. You're in a lean-back mode. Oh, sorry—an ad interrupted you from your very serious work of watching yet another Instagram Reel. Honestly, the ads you look at are probably way more productive than half the crap you're watching anyway. It's a beautiful advertising environment.

Why do you want to build something to help people do—well, we'll get to that later. But they didn't lean into it. TikTok came along and carved out—

They did get there eventually.

Yeah. I wrote that article before TikTok ever existed. I was spot on. The idea was that they needed to let go of “We're a social network” and lean into “We are an entertainment platform.” Had they done that when I wrote that in 2015 or 2016, they would have seen the TikTok threat immediately instead of—

Mm-hmm.

—waiting until it carved out a huge chunk of—TikTok built itself on Facebook, under Facebook’s noses.

I was going to say, a fun little twist with Facebook advertising—

So that's a mistake.

TikTok expanded its footprint.

5. AI Should Entertain

And then we get to the AI thing. AI right now is framed as productivity. I love AI as productivity. By the way, I'm up to 3 vibe-coded apps now, and it's fantastic. There are a gazillion use cases occurring to me, and I can spin stuff up super great.

Mm-hmm.

I'm not the normal person, okay? Part of the reason it took me a while to figure out what to build is because I so heavily optimized my computing environment 15 years ago that I watch other people—

What do I even need?

I watch other people use their computers and my skin just crawls. How do you live this way? How are you so—

Yep.

—unoptimized?

But most people are not me. They don't want to sit around and create a home inventory app or whatever the other apps are that I'm working on. Maybe we'll talk about them at some point. That's not what they want. They want to be entertained. They want to sit back. So why are you justifying your spending with, “We're going to build people assistants. We're going to be—”

Right.

—helping them be more productive? That's not what you do. What you do is you entertain people. You connect people. And by the way, in a world of AI, where you're increasingly isolated because you're just going back and forth with your chatbot—

You have the differentiation.

Do you know what's valuable?

Exactly.

You have the most differentiated spot ever. You're so well-positioned for the future. How can you not convince investors about this?

Well, and this is where it—

And the reason is—

6. The Coding Model Signal

I'm sympathetic to the investors. Let me be clear, because I see Zuckerberg tweeting today, his first tweet since 2023, by the way. Pretty remarkable.

“We're releasing Muse-Spark-1.1, a strong agentic encoding model at a very low price. It's available through our new Meta Model API and in Meta AI.” I don't fully understand—

Who cares?

—why they're doing that.

No one cares. No. It makes a ton of sense to have a coding model. All the coding stuff goes into these models. When you're doing your advertising things, their new image generation is super interesting. It basically is agentic at its core. It does reasoning; it has tool use.

What's a challenge when you're trying to generate an image? You say, “Make this image of Andrew Sharp”—one of my favorites, you know? Your favorite player now, Trae Young. I made an image of you—

Give me Trae Young's hair.

—looking like Trae Young. Right. Super complicated to pull—

His wispy, falling hair.

—to pull together.

Yep.

What if I could just type that, and it would go out, search the web, find the different images, and put them together? It could iterate on it, like reasoning. It's a really interesting model.

Mm-hmm.

You could see how this is amazing for advertising, for doing these generative ads. That makes a ton of sense. Part of that is coding. Coding, by the way, Facebook writes a lot of code. They are reportedly Anthropic's biggest customer. They shovel billions of dollars a year to Anthropic.

Wow.

That's something they should probably do themselves. It's a core part of their business going forward. So them building a coding model makes total sense. Framing yourself as a coding company is asinine. It's so dumb.

That's what's confusing to me.

It's just maddening.

Yeah.

And it's a great point. The first time he decides to tweet is about a coding model. That's neat for your buddies in Silicon Valley. I'm sorry, Mark Zuckerberg, you're not a Y Combinator startup guy. You are the CEO of Meta. What you say matters. You deciding this is the one thing you're going to talk publicly about on X, to your point, for the first time in years, just emphasizes my point. Your messaging—

Mm-hmm.

—and again, I think this is rooted in a deep-seated—this isn't a mistake. This is a—

He wants to be great.

No—

—and wants to be revolutionary, and doesn't treat the ad business as revolutionary and great.

He's vaguely embarrassed about it, right?

Yeah.

Boz goes in and cleans everyone up and whips them into shape. Sheryl Sandberg gets on the earnings calls and talks about the beautiful anecdotes of how advertisers are changing people's lives. Mark Zuckerberg wants to talk about anything else.

And by the way—

Mm-hmm.

—this is why Apple cleaned their clock in the debate about ATT. ATT was, hilariously, one of the most egregious antitrust actions a company has ever taken.

Abuses of power. Yeah.

And they completely got away with it, because who was speaking up for Facebook other than me? And yet, at the end of the day—

Mm-hmm.

—despite the fact he can't talk about it, can't seem to bring himself to express it, they are still spending billions of dollars in the face of the market telling them not to.

Yeah.

That's his redeeming quality. He just—

Well, and—

He vaguely goes in the right direction, and he doesn't care what anyone says, and he gets it done. And so Facebook is in, I think, a fine position. I think it looks like their models are really good. It looks like the last year of upheaval and pain has been worth it. They are moving in the right direction, and people will figure this out. Their ad product is going to be amazing. All the stuff I want to happen is going to happen.

It just doesn't need to be this difficult. You don't need to go through this massive stock drawdown. You could be out there painting this image. You could avoid all the internal morale issues of people feeling like, “My stock's down, I'm doing all this—

Yeah.

—grunt work. It's so unfair.”

They're measuring me every day—

There's an aspect, Mark—

—every second.

No, this is the Mark Zuckerberg experience. His redeeming quality is that he doggedly goes in the right direction, no matter what barriers are in front of him.

Mm-hmm.

His weakness is that most of the barriers, he put there himself. This could be much easier.

Yeah. Well, that's not always true. Honestly, when you look back at Reality Labs, that was the wrong direction. They literally renamed the company after the metaverse, and that's now a giant flop that hangs over everything. So it's tricky.

This is probably a stupid normie question, but what else should they be doing with that money in order to satisfy investors? If you're an investor with a pitchfork demanding that they do something different—

No, investors are—

What would that be?

Investors want them to just sit there and harvest it all, make all this money, and, I don't know, pay it back through a share buyback. At least a lot of them.

In a competition that they're not going to win, with Anthropic and OpenAI on the frontier of AI.

So here's an analogy. I've talked to a lot of people who want to do what I do, who want to write a blog or a newsletter or whatever it might be.

Mm-hmm.

The mistake everyone makes is that they want to do exactly what I do. They want to write about the big tech companies in their basement. The problem is, I got here first, and I'd like to think I do a decent job of it. So if you want to replicate Stratechery, you're going to compete directly with Stratechery, which is fine. I think that's a tougher road than it could be.

But good luck, and it's hard to—

Right.

It's harder today than it was 15 years ago, certainly.

Yeah. You're diving into my pond, and this is a pretty big pond. We're talking about the big tech companies, so you're diving into my very large lake. I just drove past Lake Superior on this fishing trip. You're diving into Lake Superior. Good luck.

Great.

You could be the biggest fish in Lake Superior. Go for it. Or what you can do is, instead of trying to be the biggest fish in a very large lake or the ocean, find your own pond.

Mm-hmm.

That's just a much better route for you to take. Be the sole person writing about this specific thing, and then charge way more money than I do because there's a very narrow number of people who want it. But if they want it, they really want it, and you can charge them a lot of money.

And they will pay for it, yeah.

I think that—yeah, there are lots of people who make very good money charging very high prices for very niche content, and that is absolutely a viable business model. The analogy here is that Mark Zuckerberg is insistent on jumping into Anthropic and OpenAI's lake.

Yes.

That's not a very fun thing to be in. You have an entire ocean that's all yours. By the way, the value of that ocean is being enhanced by AI. It's becoming more attractive. The idea of, as we just talked about it, connecting people, entertaining people, and getting out of the productivity obsession that's taking over everything—

It should be durable and even more dominant in the years to come.

They are already the biggest fish in the biggest lake, which is becoming more attractive, and they keep trying to flop onto land and roll over to the other shark-infested lake. It's nuts.

Yeah. I mean, it does not make a ton of sense to me, and we're only a year removed from them spending literally billions of dollars on various AI research projects.

That spending is fine, to be clear. I think it's going to end up going into their lake. I think they're going to be fine.

Well, no, totally. They will end up in the right spot at the end of all this, but the theory and the messaging around the signings a year ago was, “We want to develop some of the leading models—”

Right.

“—and be in the same space.”

“We're going to develop your own personal superintelligence.”

Right, exactly.

Exactly. And here's the thing, by the way: They could eventually do that. It might be the case that they actually end up getting so much data about you.

I've been waiting. They absolutely could, any day now.

No, the great thing about being a consumer tech company is that you basically get all the data. You're not dealing with enterprises that are fighting with you about data-retention agreements. It's just a fact that you get all the data. For better or worse, that's just the reality of the consumer internet. That's why Facebook succeeds: It collects so much data.

So you're already in a great place in terms of doing that, and you could, in the long run, back into, “Oh, wow, it turns out Facebook is a personal assistant for everyone.” That is not the vision. You can't lay that out as the vision. You have a very tangible, very real way to apply this, which is advertising. And by the way, they're doing this. Their ads have gotten way better.

Mm-hmm.

There's future technology on the horizon that uses LLMs to, instead of matching ads, predict the next ad, and then it's just there, and predict your content, and it all gets tied together. There's lots of complexity that goes into it, but the line from them making way more money because of AI is very clear. They're going to march down it.

Yeah.

Just talk about it. Be happy about it. Embrace it.

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.