Anthropic’s $30B Revenue Surge, Amazon’s Supplies Crackdown, Tokenmaxxing Takeover

0:13Stephanie Palazzolo:Welcome, everyone, to The Information's TI TV. My name is Akash Pasricha. It is Tuesday, April 7th. First up today, Anthropic is out with some big announcements. Its revenue run rate is now higher than the latest figure that we have for OpenAI, and it is expanding its work. with Google and Broadcom. We'll unpack that news shortly. Next up, the information published Amazon is taking a hard line on the extent to which it will absorb tariff and oil costs. We'll talk to our e-commerce reporter who wrote about that and how brands are reacting to it today. Plus, token maxing has taken over Silicon Valley.

0:49Stephanie Palazzolo:We'll get you everything that you need to know along with some inside analysis. And we will wrap the show with a look at how Anthropic is adjusting its approach to OpenClaw. It's going to be a fun show, So let's get right on into it. Anthropic's run rate is now more than$30 billion. The latest revenue figure that we have for OpenAI is$25 billion as of February. That suggests that Anthropic is now bigger than OpenAI on its top line. Of course, we don't know what happened in March. Joining me now to break down the news is Mustafa Nimuchwala, a partner at New Enterprise Associates. Mustafa, welcome back to the show.

1:25Stephanie Palazzolo:It's great to have you here. Thank you, Akash. Good to see you. Okay, so Anthropic is putting up some big numbers here. What do you think? I mean, it's pretty unbelievable. They ended last year at nine. Now we're talking about 30. They've added five Palantir's in the first quarter of the year. And I mentioned Palantir because, of course, it's an AI company. And also, it's really the only high-growth company in the software world left in the public markets. And so if you just look at what the market cap of the ads that Anthropic has done this year would imply, we're talking about 1.7 trillion dollars of palantir equivalent market cap just added this year now of course palantir has very high free cash flow margins and anthropic is burning tons of money to build models a very competitive world etc but just to put into perspective you know the kind of growth they're having or put another way they're talking about adding two work days in a quarter work days lifetime they are one of the greatest software companies of all time pretty unbelievable.

2:23I think the question at this point for this year really is, what's the constraint for Anthropic? And I think many would point to the OpenAI aggression on aggregating compute. Obviously, many of the large deals that you guys have talked about over the last year or so that OpenAI and Sam Altman has been going out and making for compute, Dario, in comparison, has often been criticized for being relatively conservative, often even talking on Duarkes' this podcast about how the difference between being aggressive and conservative is potentially going bankrupt when the dollars we're talking about and the 10x growth a year that we're talking about.

2:58So this year's constraint is how much compute was Dario and the Adropix team able to get and what's that going to look like for the rest of the year.

3:05Stephanie Palazzolo:You talked about Palantir. I wonder of all the software companies that are public that are out there, are there, I mean, And is Palantir the comp for you? Which company do you look at as a suitable comparison? The truth is, there is no comparable. I mean, we're talking about, I think, one of a kind. We just need one to go public, and then we can comp the other one. Totally. And I think the growth is really the elixir of public markets. And I think ultimately, you just have to look at the growth of a company. And unfortunately, the real reason there's no comparable is because we don't really have high growth, at least in the software world.

3:44in the public markets anymore. I think you could look at some of the semi-cap equipment companies that are very high growth. The reality is that many of those trade on forward earnings multiples. So I think that wouldn't be much of a comparable given these companies don't relatively have earnings. But I think we're already hearing about tenders happening for Anthropic at one and a half X, the last round price, call it 500. I wouldn't be surprised if when they go public, it's closer to the trillion dollar range, just given the numbers that we're talking about and the growth rates that we're talking about.

4:13and realistically, given the enterprise business, the comparative efficiency from a capital and margin perspective that we're talking about as well.

4:20Stephanie Palazzolo:Now, the comparison between Anthropic and OpenAI is interesting here. We have reported the information extensively about how Anthropic has been narrowing the gap in terms of its revenue run rate with OpenAI. Now we have this figure out. Again, we don't know what happened in March. OpenAI could have had a big March. We don't know. but the latest figure we have is 25 for them. And so Anthropic seems to be doing something right. We know it's the enterprise products. We know it's the API. We know it's the coding. Where does OpenAI still have a leg up, do you think? I think many things, to be honest with you.

4:57Ads are one. I mean, there's reporting about the rapid growth rate of the ads product right after they turned it on, and that's a pretty unbelievable, obviously, the intent that chatbots and these AIs are able to have because of the conversations that you have with them, paired with the reach that OpenAI has in the consumer world, paired with the fact that most people believe that Google and Gemini are not going to monetize Gemini in the near term because there's always a trade-off between monetization and growth, and Gemini prefer, based on reporting at least, to capture distribution first. So that's a massive leg up versus Anthropic, who's not really playing in that market.

5:33The second is the device side. Time will tell what unfolds on the hardware and device side, but we've lived in an Apple locked-in world, at least in the West, for quite a while. And that's a massive opportunity to rethink, but also just to get deeper on the consumer side and on the use cases side. The third would be kind of custom chips, as they often called ASICs. OpenAI has been working with Broadcom on that. And so that comparatively, of course, we'll talk later about Anthropik's TPU deal, but comparatively, they also have that custom silicon advantage that could play a meaningful role over time.

6:08Brand would be the fourth, and particularly people around the world know it. I think Anthropic is a bit of a better brand right now, I will say. I'm a little more skeptical on OpenAI's brand today. It's a good point, and I think that's really, this is the interesting thing, right? It's like if you ask anybody, if you just pull people around the world, most people probably outside of the enterprise, particularly outside of technology and finance circles, comparatively, probably more familiar with OpenAI. And I think there's a clip that Sam Altman, you might recall from the Super Bowl, had where he said more people in Texas use Chad's UPT than people in the world use Claude.

6:47That might have changed slightly, but I think on the consumer side, and only time will tell, of course, is consumer versus price battle. That's a good point.

6:53Stephanie Palazzolo:I mean, even as tech reporters, we live in sort of the tech bubble. And so I think you raise a good point, which is that among people who know the different labs, maybe they have nuanced opinions on open AI versus Anthropic. To the average person, I think you're still right. I mean, ChatGPT is still the way that people refer to it. So it's a good point. I will also say the advertising revenue is interesting. I think it's about 100 million, I think in annual recurring revenue. I think they probably would have thought it's getting a little higher. Still fast growth, but I think you could make the argument that the ads revenue could have grown faster For sure.

7:35Stephanie Palazzolo:I want to talk about the chip angle to all this, though. So the other part of the news yesterday was that Anthropic is expanding their partnership with Google and with Broadcom. Tell us about what the news was there from your end and then what your reaction was to it. I think this was largely known, but I think maybe a couple of highlights. You know, we're talking about the announcement was about 3.5 gigawatts. roughly people do call it$50 billion a gigawatt of CapEx. We're talking about$175 billion, all in including construction, power, etc. kind of build out. So it's a big announcement just in the scope of it.

8:17That's the GDP of many countries in the world. I think paired with that, the other important point is this 2031 commitment. Broadcom is locking in TP revenues all the way through there. I think there's been a bunch of debates on various sub factors over time about what does this look like, MediaTek versus Broadcom, which is another device design partner for TPU. And the rumors are MediaTek is way behind. So I think Broadcom is continuing to show not only in its dominance when it comes to TPU specifically, but also when it comes to custom silicon as a whole. Like I mentioned, OpenAI is working with them as well, as are a couple of the other hyperscalers.

8:56And so Broadcom is this alternative to NVIDIA's locked-in ecosystem Broadcom did this custom silicon, which over time is how people build margin. The other thing that I think is really probably worth representing and talking about in this announcement is Gemini has not been, the rumors are Gemini has not been happy about this over the last few weeks because ultimately every single frontier lab is compute constrained. To be fair, so is Google Cloud and so are every single cloud provider, but most of them have made this trade off Azure, the CFO of Microsoft has made this point that they could have grown Azure faster, but they chose to use GPU elsewhere.

9:35The same thing is true at Gemini, except, you know, GCP ended up selling these TPU to Anthropic, which robbed, you know, comparatively speaking, Gemini of the progress they could have made in a very, you know, fast paced, hard battle that we talk about all the time in the AI frontier world. I think that to me is really the biggest thing. And only time will tell because I think Gemini and Google broadly has gotten a lot of criticism for not being aggressive and not being as AGI-pilled as the other frontier research labs.

10:04Stephanie Palazzolo:So I want to unpack this here. So from what you're hearing on the Gemini side, so what exactly are they unhappy about? Is this Gemini, I guess, you know, the AI lab within Google saying, hey, we can't get access to the compute that we need from TPUs? Is that the idea? What's the... Precisely. Precisely. I mean, if you think about it, right now in ComputeLand, everything is zero-sum. And the question is, what is the return on invested capital? What's the goal with which you're allocating compute? Right now, I mean, we just talked about Anthropics revenue growth rate. That is, if you just think about it very simply, it's compute resold at a markup with really smart talent that builds these amazing things to do with that compute, but it's directly tied to compute.

10:57And so Gemini, not only their distribution, but their revenue model, their modes, also talent is a big part of it. Every week that passes, every month that passes, you made the point about Anthropik's brand. These things really matter to retain amazing research talent. And if Gemini, every month that they don't have incremental compute or they sell it off to Anthropik, or for that matter, directly, now Anthropik has that marginal compute. What if that helps Anthropik as well? So not only does it hurt Gemini, It also may be hurting, helping their competitors set. And so a lot of these things play in.

11:27Stephanie Palazzolo:So from what you're hearing, this is an internal tension that is happening at Google is Gemini and the TPU team sort of not against each other, but I mean, fighting over the same resource. Precisely. And I think this goes to show a really important point of these kind of large big tech companies where they have multiple business units and also different business models. GCP and Thomas Curran, who runs that, is incentivized on selling cloud services. comparatively, Gemini and DeepMind specifically, right, which Gemini is a part of, is incentivized by making progress at Frontier AI Research and then having the consumer distribution and helping to offset the search business, which is a third and separate business unit, which is separate from Waymo, which is now an independent agency, which is separate, for that matter, from the custom silicon team, technically, which is only monetized by Google Cloud and then their operating partners like Fluidstack and others.

12:16So this, the next couple of years i think if you look back at how look look look at how google performs in a couple years looking backwards i wouldn't be surprised if this big tech aggregation political dynamic story ends up being a big part of whether they're able to work through it or whether that slows them down further and that was of course the the worry people had a couple years ago and so far they've been able to balance it but it's always going to be a tough needle thread okay and i just want to

12:41Stephanie Palazzolo:go back to something you said earlier you talked about media tech media tech of course is the low cost challenger to Broadcom in developing these custom ASICs. And we've reported the information that Google has turned to MediaTek as well to help them, in part, help them develop their TPUs. And so from what you're hearing, that's facing some challenges, that partnership. Yeah, I mean, I think the rumors are that MediaTek has not gotten their wafers in that TSMC, that the quality and the timing might be issues. I mean, if you just think about it from Google's perspective, just like we're talking about a lot in the ecosystem broadly, people are diversifying, right?

13:22You don't want to have single supplier risk. You don't want to foment too much. You don't have too much pricing power in your suppliers. You want to create competition. I think that's why Google went to MediaTek, even though Broadcom has built every single generation of the TPU so far. The other thing that's probably worth talking about is each incremental supplier you get may get you more capacity at TSMC as well, even though TSMC talks to, as they say, their customer's customer, which in this case would be Google. But Broadcom, and Google in particular, separately has also gotten a lot of criticism for, Google specifically has gotten a lot of criticism for not booking enough TSMC capacity.

14:00And I think when we talk about this at every part of the stack, this is where Google gets our criticism for not being enough AI pilled, because they didn't get enough TSMC capacity, they sold off TPU to Anthropic, they're not making, they haven't made certain investments, that they could have made otherwise in the distribution side. So MediaTek is just a way for Google to have more diversification and more paths to capacity at TSMC, and also over time potentially at different types of TPU that they may build as well, so they can do more workload specification.

14:28Stephanie Palazzolo:The last thing, Mustafa, that I wanted to talk to you about is Anthropic very much touted its multi-chip workload in its press release. It said, hey, we're running on TPUs, we're running on Tranium, and we're running on NVIDIA GPUs. I think back to the deals that OpenAI has done. OpenAI has done a deal with AMD, of course, with NVIDIA. Again, I'm forgetting the matrix here of, you know, which lab has done a deal with which chip company. But the question I have for you is, to what extent do you think Anthropic is doing a better job of diversifying its chip workload? Or is it very neck and neck right now?

15:09Stephanie Palazzolo:I don't know the answer. It's a really good question. And I would maybe say that the first frame, at least today, to look at it from is less diversification and more compute is so constrained that every kind and any kind of compute you can get is a big advantage. As a result, Anthropic has realized this pretty early and decided to do the work, which is often just handwriting kernels in some cases. It's kind of making some of the NVIDIA architecture pieces more flexible sometimes it might require writing more software sometimes it might require porting existing software and once you get in the flow of this clearly anthropic feels pretty confident about this because they're the largest customer of amazon's tranium they're the largest external customer of google tpu they're one of the largest customers in the world uh now as nvidia as they announced at gtc and you keep going you wouldn't be surprised if they see other custom silicon coming over time so to your point on diversification i think the diversification from a market structure perspective probably will be important over time but it's really this competence that they're now building internally across low level mid-level different parts of the infrastructure stack on how to make custom silicon performant and of course at rapid scale that will be really helpful to them over time whether it's for custom silicon or new kinds of silicon or heterogeneous compute across training and uh inference which only time will tell how much the fragmentation and inference looks like and in the private world that's what we've been talking a lot about historically is that inference is easier to sell custom silicon into versus training.

16:43And so you might have much more fragmentation of the many startups that you might have seen in the inference native world.

16:49Stephanie Palazzolo:Great. Well, Mustafa, I want to thank you for coming on. That is Mustafa Nimutwala, a partner at NEA here on TI TV. Okay. Tariffs and oil prices are making it more expensive for brands to sell goods on Amazon, as my colleague Anne Guillen reported this morning. The company has taken a hard line to protect its margins. I want to bring her on to talk more about what she found in her story. And welcome back to the show. It's great to have you here. Hey, Akash. Let's talk about the macro here for a minute. So you've got oil prices that are tough right now. You've got tariffs. How are brands doing with all this?

17:29Yeah, I mean, I think you laid out kind of the two main things that brands are really spending a lot of time thinking about. We're about a year out from Liberation Day when a lot of these new tariffs went into effect last year. And I think at that time, a lot of brands were just kind of trying to wrap their heads around, you know, what does this mean for my business? How is this going to be something temporary? Is this a permanent reality that I have to adjust my business to? And so now that we're a bit removed from the introduction of a lot of those tariffs, I think brands are starting to realize, you know, this is sticking around for the foreseeable future.

18:11So, you know, we're going to need to adjust. And also, you know, this is the time of year that a lot of brands are starting to or have been negotiating with Amazon about, you know, their wholesale terms for selling Amazon inventory. and then, you know, throw in the Middle East conflict and a lot of uncertainty about the price of oil and shipping costs. And, yeah, brands have a lot to sort through right now and a lot of different new costs and price increases to work through.

18:51Stephanie Palazzolo:Okay, so now let's get to your reporting. Is Amazon helping with this? So we've seen, and, you know, from talking to several advisors that work with a wide range of these brands and help them negotiate with Amazon, you know, a lot of them are telling me that Amazon, you know, hasn't been very receptive to paying suppliers more for their inventory to help cover some of these increased costs. So, you know, that's leaving brands to kind of foot that bill on their own. And so that's leading some brands, you know, particularly brands that are only in a, you know, brands or companies that are only in a handful of categories that maybe don't feel like they have as much leverage with Amazon.

19:37They're starting to question whether it's worth it at all to be an Amazon supplier. So we're seeing more of that, you know, start to start to be something that brands are considering over the past couple of weeks and months.

19:49Stephanie Palazzolo:So I just want to say this. So you have the brands who their input costs have gone up because of tariffs and oil. And so they would want Amazon to basically accept higher prices for – basically they would want Amazon to pay them more to store the product as inventory. Yeah. So a lot of coverage of Amazon sellers comes from outside sellers, which are responsible for about two-thirds of the sales on Amazon's marketplace. But about a third of Amazon's retail sales come from inventory that Amazon buys in bulk, basically, from these wholesale suppliers and then resells. So those are the suppliers that we're talking about now.

20:37Stephanie Palazzolo:The resellers. Or the wholesale suppliers who sell their inventory in bulk to Amazon. And so these kind of negotiations where they work out how much Amazon is going to pay them for their inventory, that's really kind of the key, you know, point of control that they have. Because after that, you know, a lot of, you know, things like the ultimate retail price at which the good is sold, you know, Amazon is in charge of all of that. Whereas these outside sellers, you know, the other two thirds of sales, they have a lot more control over their inventory, over their pricing. But that comes, you know, at an additional cost, obviously.

21:21So these are sellers where, you know, they're basically wholesale suppliers to Amazon. Okay.

21:27Stephanie Palazzolo:So what's the net impact of this for brands? What are they doing about it? So like I mentioned, a lot of brands are starting to wonder whether it's worth it at all to be an Amazon supplier anymore, kind of given this squeeze that they're feeling. And so that would either require them shifting to this outside seller model, which, as I mentioned, you know, gives them, it does give them a little more control over how their products are sold and priced and all of those things. But that does come with additional costs that, again, you have to pay Amazon for. So you have to pay them for storage, for shipping, for advertising, for all of those things.

22:08Or there are some brands that are starting to wonder if it's worth it to sell on Amazon at all. And just, you know, focus on other, you know, whether that's brick and mortar sales, other online marketplaces. I think brands are starting to kind of reach, some brands are starting to reach their breaking point with Amazon a little bit.

22:27Stephanie Palazzolo:Right. So, you know, and this is sort of the inside baseball of the Amazon reseller, outside seller ecosystem. I mean, if we just take a step out for a second, why should we be paying attention to this story? What does this tell us about Amazon's strategy? How could it ultimately affect consumers as well? Sure. I mean, something that's come up, you know, in talking to several of these advisors who, again, they work with a wide range of brands. They, you know, pretty much all of them have been in the Amazon and e-commerce space for a long time. You know, a lot of them have explained to me that historically Amazon's focus was on having, you know, a broad selection of goods.

23:08And, you know, when I asked them for a comment on this story, Amazon told me that was still the case. But what I'm hearing from some of these advisors and people working with brands is that, you know, Amazon is less tolerant of some of these unprofitable products than they used to be. And whereas in the past, Amazon would sell a product at a loss in order to make a good impression on shoppers, to make them feel like Amazon had a wide variety of goods, and to ultimately get them to come back and shop again. Again, we've seen Amazon still pursue that strategy in categories that are really important for them right now, like grocery.