Thinking Machines Co-founders Head to OpenAI, Zuckerberg’s AI Warning | Jan 15, 2026

0:13Welcome, everyone, to the information's TI TV. My name is Akash Pasricha. It is Thursday, January 15th. Today on the show, we are talking about TSMC's big quarterly results, and we'll speak with our Asia Bureau about why the company has become so dominant in the chip manufacturing story. We'll then have the latest on all the drama over at Thinking Machines Lab, and we're also going to break down the busy week that Meta and Mark Zuckerberg have had. We're then going to break down ByteDance's valuation. We is undervalued on the secondary market right now. And finally, despite all the layoff chatter that we've been hearing, one of the biggest software companies in the world is ramping up hiring.

0:56Stick around to find out who that is. It's going to be a fun show. Let's get right on into things. TSMC reported its quarterly results. The company is surging on all fronts. I want to bring on Channer Liu, our Asia reporter, to break it all down for us. Channer, welcome back to the show. It's great to have you here. Thank you for having me. So what were the headline numbers from TSMC's quarterly results? Well, TSMC delivered some exceptional results today. December quarter, revenue up. Net profit jumped to a record high. For the fall last year, revenue surpassed$100 billion for the first time.

1:36The biggest revenue source for TSMC is the advanced chips, of course, which is the 7nm and below, which now account for over 70 % of the revenue. And the driver behind this is pretty clear. It's chip for AI. It's not just the processor like Nvidia's GPUs or Google's TPUs, but also the supported chips like we can saw in the data center like those for the networks. So TSMC, because of all of this, TSMC now has become more confident in the sustained AI demand. They are planning record capital spending of like 52 to 56 billions in 2026 this year, and projecting nearly 30 % revenue growth for this year.

2:23So$56 billion in CapEx, what is all this money going towards? Most of the car packs will be for buying advanced chip making tools for AI chips, meaning that TSMC will use the money to expand the advanced chips production capacity. But the capital spending announced now won't help, I would say won't help meet today's demand. It's for 2027, 2028 and beyond. In the near term, TSMC has to get a little bit more created. They are boosting the productivity at the existing factories and repurposing some of the factory space and equipment from older facility to support advanced chip production. So I want to pivot to talking to a story about you wrote this week, which is actually, it's on the same topic, which is TSMC's dominance in the chip manufacturing ecosystem.

3:21The premise of your story was really looking at all the ways in which the world relies on TSMC. and I wonder if you can explain to us why is TSMC so dominant? I mean, what is the secret sauce? How is it that they have become seemingly the only company in the world that can manufacture these chips as well or as efficiently as every company in the world might want them to? I would say, first of all, and maybe the most difficult one, difficult reason to explain is their technology. We probably need a little bit more, like one or two hours to explain why the technology is so good. But I do have high respects to Android.

4:00I mean, look, we have the whole morning, but it's late for you. So you go as long as you want. So I think, well, technology-wise, I would say they have good technology. That's why they can achieve higher yield, which means that if you place a chip-making order at TSMC, the cost would be much more lower than you place your chip order to other places. And besides technology, TSMC's dominance also came from their business model set by their founder, Maurice Chang. They only manufacture chips. They don't design or sell their own chips. It means that they would never compete with their customers like Apple and Nvidia.

4:39So everyone can trust them with their most secretive and sensitive design. So for example, other advanced chip makers, TSMC's competitors such as Intel and South Korea, Samsung, electronics. They design their own chips, but also try to manufacture for others. Now, you talk in the story, though, about how despite their dominance, the fact that they are not designing their own chips and they're relying on contracts from these chip designers, that actually is a little bit of a limitation for them insofar as how far in advance they can forecast the facilities that they need to build. Yes, they are.

5:16I think I remember CC way, which is TSMC's CEO. Obviously, he was in a very good mood today. But he did mention and admitted to the analysts that they have been very conservative in terms of the capacity, obviously because of the business model and also because they are the one, they're probably the only few companies that literally need to pay like tens of billions of dollars to build the capacity and have to make sure they can self-order capacity. now if i am nvidia or google i'm looking to have a company make my chips if i can't get capacity at tsmc and i'm just using those two as an example but where else could i go to to get these chips made i would have to say for those two big companies maybe maybe a bad example i i know but it's fine like for nvidia they probably said like oh i want like like 10 chips but tsmc people said like, okay, you want 10 chips, I give you eight.

6:17Like they can never, they are never going to offer as much as they requested. So for any companies like this, if they cannot secure enough capacity at TSMC, they can totally turn to other chip makers like Intel and Samsung, like the one we just mentioned. But the transition can, well, the transition is rarely seamless. You know, while Samsung and Intel, they also have advanced chips production capacity, they are trying to catch up with TSMC. They often struggle with lower yields at the most advanced chip productions. The yield here means the percentage of usable chips per wafer. So lower yields means higher cost to produce chips.

7:01And also switching foundry is not like switching from different serial brands. It requires redesigning the entire chip's architecture and to fix different manufacturing technology, which can take a year or two. So if TSMC really has a stronghold on the market, you mentioned the$56 billion in CapEx, are they building more factories? What is their expansion plan? Yes. The quick answer is yes, they are building more factories. Well, not fast enough to solve the immediate crisis. GSMC is building new facilities in Taiwan, Arizona, and Japan's Kumamoto. But these all take years to come alive. And there's also a global shortage of the chip making experts that needed to operate those complicated chip making machines in those factories.

7:56So, and it's not just about building the front end shipmaking facilities. Those are, and also the advanced packaging is not enough. For example, there's one Arizona factories by TSMC has started to produce NVIDIA's most advanced Blackwell chips. But TSMC still has to ship them back to Taiwan for advanced packaging. So even if they're making them in the US, the bottleneck still remains with the facilities in Asia right now, which is that they have to package them there and then we're back to where we started, really. Yeah, exactly. Great. Well, it's a company that I know that the world is watching and I want to thank you for coming on and sharing with us all that you know about it.

8:43We will let you get some sleep because I know it's late where you are. That is Chenna Liu, our Asia reporter here at The Information. Thank you so much for having me. Okay. There was a little overnight drama in Silicon Valley after Thinking Machines Lab CEO Mira Mirati said the company had parted ways with co-founder and CTO Barrett Zoff. Not long after, OpenAI's FijiSimo announced that Zoff, along with fellow Thinking Machines co-founder Luke Metz, would be returning to OpenAI. Zoff and Metz both left OpenAI in late 2024 to help set up Thinking Machines Labs. Here to break down what happened and what it signals about the AI talent wars is our AI reporter, Stephanie Palazzolo.

9:29Stephanie, welcome back to the show. Hey, great to be here. So what do we know? This is juicy. I know, I know. We're kicking off the year in the right way. Just a lot of AI drama. No, but I mean, essentially what we know for sure is pretty much what you just said. Again, last night, Mira Mirati, former OpenAI chief technology officer, now CEO of Thinking Machines, she posted on X that they had parted ways with Barrett. An hour later, Fiji Simo, who's the CEO of applications at OpenAI, said that OpenAI had hired Barrett along with two other Thinking Machines researchers, including another co-founder of the company.

10:11and Fiji basically said, you know, these talks have been going on for weeks, which implies that this is something that had been in the works and maybe Barrett and these other folks were thinking about joining OpenAI at least for a while. So Barrett and Luke, what were their roles at OpenAI when they were around there the first time? Yeah, so both of them were very senior researchers at OpenAI. So Barrett specifically was the lead for a team called PostTraining, which is basically the final step of the training process where you get the AI models ready to be released and to be put into products.

10:49And then Luke was another senior researcher at OpenAI that had been there for quite a while. He's actually part of the kind of original team that built what ended up becoming ChatGPT. Okay. And remind me, I mean, these are two very important functions. PostTraining, that is that key battleground area that has become very competitive, right, for the models between Gemini and ChatGPT. I mean, this is a pretty important role. Totally. I mean, post-training has been a very important part of the training process, as you mentioned, and has especially become important in, you know, the last year or two.

11:24So it was a very big deal whenever Barrett left OpenAI in the first place to go to Mirror Startup. And, you know, it's now a pretty big deal that he's coming back. And just to confirm, we don't know anything about the reasons here just yet, right? This is where there's got to be some more reporting. We've got to find out what's been happening. That is the big question. I think right now there are two kind of conflicting narratives coming out. So, you know, on one hand, we have reporting that Mira had told staff at Thinking Machines that Barrett had basically been fired for, quote unquote, unethical conduct, which, you know, doesn't sound great.

12:03And then on the flip side, we see from OpenAI that it sounds like from their perspective, they had been in discussions with these researchers for weeks to potentially come to OpenAI. And there's even some stories last night that said that Fiji told OpenAI staffers that basically Barrett had told Mira earlier this week that he was talking with OpenAI and then he was pretty quickly fired just a day or two later. So there's kind of a lot of questions around what exactly happened, Was this just a researcher that wanted to make a move to a different company and then got fired for it? Or did he actually do something wrong?

12:40And that was kind of the true reason behind him leaving Thinking Machines. Okay. So where does all this leave Thinking Machines Lab now? Yeah, I mean, I think this does depend on how important some of these researchers actually were. Like, yes, I mean, these are, you know, these are very high profile, very experienced researchers. but we don't know the kind of exact details around what exactly were their contributions at Thinking Machines, how kind of like critical and vital were they to like the success of the startup. So that is kind of an open question. But we do have to say, though, like it doesn't look great, you know, like a startup that is extremely highly like it's very expensive valuation.

13:24I mean, its last price was was 10 billion. But then in the last couple of months, they've been in talks to raise at a crazy$50 billion price, which is pretty wild considering that the company has only really released one product. And that has, you know, I think those are questions about how that product has kind of gone so far. So it doesn't really look great to have such a young startup losing pretty important researchers and a number of co-founders in just its like first year of existing. And, you know, even investors I spoke to last night were like completely shocked, very caught off guard.

14:01They really did not see this coming. Does this say anything to you about the AI talent wars, or does this really feel like an inside baseball story of 20 people that are really just playing musical chairs around these three or four different companies? You could call it the original open AI group that has spread their wings, now they're coming back. Like, I don't know, it doesn't seem like it's poaching, although maybe it is. What do you make of all this? Yeah, I mean, this definitely ties into the broader theme of this crazy talent war that's going on. But I think this case does seem pretty specific to thinking machines and open AI.

14:43It does seem like there's more to the story here and perhaps specific reasons behind these people leaving that have to do, you know, with thinking machines or what it was like to kind of work there. So yes, I'm sure that they got paid pretty handsomely to go back to OpenAI, but it does feel like there's more to the story that's pretty specific to this one case. Great. Well, Steph, I want to thank you for coming on. I really appreciate it. That is Stephanie Palazzolo, our AI reporter here at The Information. Okay. Meta has been very busy this week between its layoffs, its big vice chair hire, and a report that it is looking into doubling production of its Ray-Ban smart glasses.

15:24There was also the big announcement around MetaCompute. I want to bring on Gil Luria from DA Davidson to give us some analysis on how he's putting all of this news into context. Gil, welcome back to the show. It's great to have you here. Thanks for having me. So, big picture, Meta is making all these changes. You look at the stock price for the past year, I mean, it's flat. And yet, the ads business is still surging. there are all these changes to move towards the ai opportunity what do you make of all this yeah i mean meta is the least expensive of the mega caps they're trading in the low 20s there's nothing else in mega cap land that's not trading in the high 20s or above and and that's a very that's really just because mr zuckerberg is trading is treating meta like his own company Like he still owns it, which he does.

16:17And investors are just uncomfortable with that. He has this unbelievable business, right? His digital ad business is growing just through 25%. It'll probably grow more than 20 % this year at very high margins. If he just let Meta be that, the stock would be a lot higher. But he's not. He's investing very heavily still in reality labs and then investing even more heavily in building AI data centers that he's not even reselling. At least for now, he's just using to do R &D. right and so investors are just uncomfortable the fact that you have such a great business but you're growing expenses and capex even faster than revenue and that's just not what other companies are doing right now I was going to ask you this question at the end but I'm going to go there directly how long have you been covering metaphor now officially two years but really since before their IPO exactly so so you've been following the company for a long time and the question that I wanted to ask you, Gil, is we see this version of Mark Zuckerberg, which is making big bets and not accepting the status quo of I'm just an advertising business.

17:37Has that changed at all over the time that you've followed the company? You know, has his focus, is it the way of running the company? I mean, has it been the same all along or has it shifted recently? I think we've had some back and forth, right? There's times where he does feel responsible to shareholders. We had famously the year of efficiency a couple of years ago. During the year of efficiency, the stock more than doubled. Because that's, again, all investors want is, can you just be responsible with our capital? And when he does that, he gets rewarded handsomely on the share price. But to be fair to Mr.

18:17Zuckerberg, he sees a huge sense of urgency right now. He feels like on two different fronts, there's going to be a winner take all or winner take most. And he wants to and needs to be one of those winners. So one is the next hardware platform, right? We're going to go away from a handset to wearables at some point. And he wants to win that because otherwise he'll continue paying Apple and Google the 30 % tax. And then on the frontier AI model, he feels like if he doesn't win there, he's just going to be relegated by those winning models and lose the biggest incremental economic opportunity out there.

19:02So that's in his mind, he's doing the right thing. He's in founder mode and he needs to win. Now, to your point earlier about the news that's come out this week in terms of some layoffs at Reality Labs and there were other stories about that. Maybe he's willing to take that back a little bit, gesture towards investors that he's not just incinerating their money, but we'll only find out when they report and we get the magnitude of savings and the magnitude of how much he's willing to scale back ambitions on those two fronts. But again, it's very understandable. He thinks those are the two biggest winner-take-all markets, and he wants to be that winner.

19:40Now, if you look at Reality Labs right now, the business is doing, I think, in or around$2.2 billion for the last 12 months in terms of revenue. It is obviously a fraction of what the ads business is. But I wonder, as you look at this opportunity for Reality Labs, how big do you think that business could get? So in the short to medium term, we're just talking about the Ray-Bans, right? and maybe the glasses with the little projector, the$800 projector glasses. We're mostly talking about the$300 Ray-Bans. That is not a very big market. But it's how you get to the ultimate market of, again, the next hardware platform.

20:24Because eventually, he'll be able to fit all the compute that we have in our handset into the glasses. And at that point, it's going to be as big a market as Apple's has. and that's what he's really focused on. The problem is the$2 billion of revenue is coming with about$4 billion of loss every quarter. That's the problem is that you don't need this much loss to build$300 Ray-Bans or even$800 projection glasses. You could actually do those for a lot less. And by the way, Google and Apple have their own effort and Samsung have their own effort to participate in these markets. They're not losing money the way he is.

21:07So he's really investing well beyond where the platform is now, again, because of this notion that at some point in the future, five or 10 years from now, that's going to be the platform that everybody's going to have. And when you have that, you can be the one that charges the 30 % tax on every application. Now, all of this talent that he has spent a lot of money on acquiring these giant contracts, What signals would you look for in terms of understanding whether or not those investments paid off? And I ask this question because a lot of that talent was researchers. And we know that researchers are really, the thing to tie them to directly is how good the models are.

21:53And we know that the Lama family of models are an open source family. They've had some mixed reviews. And so So are you looking to see that, hey, maybe the next family of LLAMA models, maybe it really excels on the benchmarks? Is that what we're looking for? Are we looking for ultimately the next wearables product to be a big hit with consumers? You just said the market wasn't that big. How will we know if this research investment will have paid off? No, that's exactly right. Right. The next frontier model from Meta needs to be competitive. You were being kind when you said mixed reviews. The last iteration of Lama - I'm a nice guy.

22:36I'm a nice guy, Gil. I'm not as nice. The last iteration of Lama was a colossal failure. That's why he fired everybody and spent billions of dollars hiring a new team, is that the next model better be good. Because at this point, between what Google, OpenAI, and Anthropic are doing, they're pulling away. It's not like they're stopping in their tracks with the current generation of frontier model. If the next model from Meta doesn't at least catch up, they're in a lot of trouble. And by the way, they probably wasted$100 billion or more doing that between the CapEx, the salaries, and the operating expenses.

23:17We're at least$100 billion into trying to develop a new model. It better be good. Maybe this iteration won't be better than the other ones. but it would have to at least somewhat catch up for us to believe that the iteration after that would be better. Now, we don't expect it to actually be open source. It'll be closed, okay. Yeah, just because of where the market is headed. And by the way, we're not hearing good things about how things are going inside of Meta's Frontier Lab. It's in the organizational, cultural issues there don't go away just because you're paying somebody $14 billion. Those issues sometimes get worse when you pay somebody for it.

23:56What are we hearing? That they're not necessarily having success with the development of this new frontier model. And maybe it's happening quietly, but the scuttlebutt in the industry is that the next generation of the frontier may actually not be compatible. Last question for you. We have talked on this show about the relationships that all the various tech companies have and are building with NVIDIA. It is obviously a very important company in the ecosystem. We haven't yet seen any headlines about any big partnerships between Meta and NVIDIA, at least explicitly. And one of the predictions that we've had from a few of our reporters here at The Information is that we might see that this year.

24:44We might see NVIDIA and Meta get together in some official way. How realistic do you think that is, and are you expecting something there? So yeah, your reporters have done a great job. Anista and Anita have done a great job reporting on NVIDIA and Meta and Google's TPUs. And we would expect Meta to diversify away somewhat from NVIDIA by buying more TPUs. And that's part of the story that they broke. But realistically speaking, everybody's still buying NVIDIA. Everybody's probably buying as much NVIDIA as they can because NVIDIA works. They know how to use it. All their scientists know how to use it.

25:20it's available. You could plug it into the data center. You know it works. It's not, oh, will this next generation of TPUs be very good? Will the next generation of AMD be very good? Jensen's chips work. The next generation is going to work even better. You know that. And so NVIDIA will continue to be the lion's share of AI compute equipment for the foreseeable future. but all these companies, all of NVIDIA's customers want to diversify away. They don't want to be beholden to one vendor. So there's this aspiration to diversify away. And right now, the main way to do that is to use or buy Google TPUs.

26:03Great. Well, Gil, I want to thank you for coming on. It was a great conversation. That is Gil Luria from DA Davidson here on TITV. Okay, what is ByteDance worth? That has been a big question as the company remains one of the largest companies that has yet to go public. My colleagues Anita Ramaswamy and Jing Yang published a column today suggesting that the$330 billion price tag that ByteDance is trading at right now on the secondary market is undervaluing the business. I wanna bring on Anita to walk us through her analysis. Anita, welcome back to the show. It's great to have you here. Hi there, Akash.

26:43Good to be back. So$330 billion is too low, you say? Yes. That is the point of view. Why? It's too low looking at ByteDance and some of its comparable businesses. I mean, just to take a step back, ByteDance is a huge business. It's digital advertising focused. It has apps including TikTok in the US, Douyin in China, and a bunch of other social media apps. And then in addition to that, you have a virtual reality business. You have all sorts of things in this giant company. And it's basically just now gotten to the point where its revenue scale is similar to that of Meta platforms. But Meta is trading at a much higher valuation multiple than ByteDance is if you sort of forecast out some reasonable projections for this year and next year's revenue.

27:31So the point to take away here, Akash, is ByteDance has grown really fast in the last couple of years. but a lot of regulatory issues, especially in the U.S., have contributed to TikTok trading at a discount to some of its peers. Now, you say revenue is around the same range as Meta. So how much revenue are we talking here? We're talking the$200 billion annual revenue range. Obviously, ByteDance is private, so we don't know for sure, but there has been some reporting on their quarterly revenue. And if we sort of extrapolate that out, that is a revenue pace of around$200 billion for 2025. And do we know anything about the growth rate of the business?

28:06Yeah. So compared to last year, the Q2 revenue growth for ByteDance was around 25%. And it was a little bit faster last year. So there has been a little bit of deceleration in the business. But if you look at where ByteDance is growing and where Meta is growing, very similar growth rates and actually ByteDance is a little bit faster. So Meta is one company that you could pick as a comp. Are there other companies that you look at? And we're only talking about companies based here in the US right now. I wonder if you looked at companies that are trading publicly in Asia. Yeah, it's a really good question, Akash.

28:44I mean, Meta is obviously the one that comes to mind for a lot of people in the US if you think about TikTok and Reels and just sort of the usage and consumption patterns of something like ByteDance's products. But in Asia, Tencent is a really good comp to look at as well. And Tencent also trades in the range of six to seven times next year's revenue, which is around the same multiple that Meta is at. Now, if you look at ByteDance in comparison, once you do that projection, if you assume this year, ByteDance is going to grow at 25%. And when I say this year, I mean 2025, the period for which they haven't reported yet.