All right. So, you did mention the enterprise efforts at Meta. I'm going to read this note from Tun, which was a counter to what you wrote earlier in the week criticizing those enterprise efforts from Meta. Tun says, "Folks, I understand the conventional wisdom. Consumer and enterprise markets are wildly different, and it's tough to execute on both, even for companies as large as Google. However, I'm not sure that entirely holds when selling models. Anthropic is going from $1 billion to a hundred billion dollars in mostly enterprise revenue in a couple of years without starting with a traditional enterprise go-to-market strategy. Same is the case with OpenAI recently. If Meta is confident in its model trajectory, doesn't it make sense to at least have the option of monetizing its models in multiple ways other than consumer agents alone? Additionally, because of their enterprise momentum, both labs can overpay per megawatt of compute since they monetize that capacity better than anyone else. I think 50% of the incremental capacity coming online in the next few years will go to those two labs alone. However, if Meta's monetization of compute for consumer agents turns out to be lower than the other two labs, which have significant enterprise revenue while achieving parity on model performance, Meta will at least have the option to compress the price per megawatt for those labs so it can continue building and bidding for additional capacity. What do you think, Ben?"
No, read his last paragraph.
Okay. Meta can always prioritize consumer over enterprise or vice versa depending on revenue per megawatt. I think this provides downside protection for all of their compute and also an internal clearing price for allocating compute. What do you guys think?
I don't dispute any of this. I don't dispute the issue of Meta's revenue, Meta's data. Like, if it makes a coding product that is compelling, why wouldn't other people like that? The problem with this email, and the problem — I was in a long debate with someone about this the other day, just listing out reason after reason why Meta should do an enterprise product. Okay, Terrone, nowhere in this email, and nowhere with my interlocutor...
Interlocator. [laughter] Is that what it is?
I just ask one question: Why would any enterprise buy from Meta?
"Will it work?" [laughter] is the operative question.
It's not "will it work." No, it's more specific: Why would an enterprise buy from Meta?
Because I was talking about the enterprise strategy working. Because it's cheaper, potentially, and nearly as performant as the frontier options.
Well, but Terrone says Meta can always prioritize consumer or enterprise. Do I want to sign up my company for a company that is a consumer company? And by the way, this is just option value for them.
Precious whims. My future depends on whether Meta succeeds. Every justification for this is sociopathic in nature. It is about Meta's needs. It's about Meta's monetization. It's about Meta's ability to bid for compute. It's about Meta's need for a competitive model. Meta, Meta, Meta, Meta, Meta. For a business transaction, you need a counterparty. No one can articulate a reason why any enterprise customer — if you're an enterprise and you sign up for Meta, I'm questioning your leadership. It's not their business. It's not their focus. They're doing it for Meta reasons. There's a long history of companies in tech doing things for their own reasons because it makes sense for them, and it doesn't work because their motivation is not the customer. It's not meeting their needs. It's not doing what they want. Say what you will about OpenAI and their shift to enterprise, but they shifted. They're not doing stuff for the consumer. They're doing stuff for the enterprise so they can be competitive in the market. Yes, of course, model quality matters, and you could get to maybe a world where Meta's model is so much better that people use it anyway, but then you could just sell it without a whole enterprise team. I, by the way, am on board with you — go back to my letter to Mark Zuckerberg — sell compute, that's fine. And actually, I endorse the internal clearing price mechanism that lets you know, are you using — like, instead of just blindly building compute, let's monetize.
Is the consumer effort actually working relative to...
Exactly. It's a good sort of limiter on what they're doing, but no one is articulating...
And so the problem is that you can have all these reasons why they need to do it. "The cost — like, Ben, you're underestimating how much this compute is going to cost. You're underestimating how much XYZ." No, I'm not underestimating. I'm asking: Where's their customer? Who's it going to be? Why on earth would they buy from Meta?
Mhm.
What are the downsides of them going this direction, though?
And meanwhile, you're looking at this unbelievable opportunity in the consumer space. And so people come back to me and they're like, "Oh, well, you know, Meta's total revenue this year compared to the projections for AI compute is very small."
That's not the opportunity. Like...
It drives me crazy in this AI debate. You see this in place after place. People will look at one area of the market as being very dynamic and changing rapidly, and then assume everything else is static and nothing changes. Right? We had this debate — like our dorm room debate about jobs in the future. Like everything has stayed the exact same except for AI goes crazy. No, that's not going to happen. When one thing changes, everything is going to change. In this case, let's be very clear what we're talking about in terms of the opportunity. I thought we were clear a week ago. Let's say it again.
It's not Facebook's current ad revenue.
It's number one, Facebook's current ad revenue getting way better. Obviously, that's kind of obvious, I hope. And it's obvious. And by the way, you can see a future where with LLM advertising, you know, it's like — did Facebook probably could have just made your data available in Meta AI? No one seems to care. Real validation of the "no one cares about this except for people in the media and San Francisco," but they didn't. They made a promise that actually might be working out. The future is like your own personal ad engine in your Meta AI that is just predicting what you want to see. You like all LLM-based. It's not the old two-towers matching approach. It just knows exactly what you want to see and puts it in front of you. And they actually move to this world of totally personalized advertising. It's just you and your LLM and your Meta AI showing you exactly what you want to see. Totally untouchable by anyone else. Like, that's possible here. So that — but this is all sort of Meta stuff. But what happens in a world where suddenly Google-type advertising doesn't make sense? Amazon advertising doesn't make sense? People still need to sell products. What is the one advertising that's still going to work? Boy, oh boy, the pricing power for Meta in that universe.
