Training Robots With Home Cleaning

As the AI boom in trouble as costs pile up and productivity questions emerge, Anthropic is now the largest AI startup passing OpenAI, and Robinhood will let your chatbot trade for you. That's coming up on a Big Technology Podcast Friday edition right after this. I'm just back from ServiceNow's Knowledge 2026 in Las Vegas, and the conversations I had there are ones you're going to want to hear. I sat down with their president and CPO Amit Zaveri on the platform strategy powering enterprise AI, Chief People and AI Enablement Officer Jackie Canney, and Chief Digital Information Officer Kelly Romack on what AI really means for the workforce. The technical leaders behind ServiceNow's NVIDIA partnership on shipping AI at scale and Ulta Beauty on deploying service now as technology across 1300 stores. If you want to know where Enterprise AI is actually headed, not the hype, but the real story, you can find these videos on my YouTube channel, search Alex Cantrutz on YouTube. Depending on who you ask, between 80 and 95% of enterprise AI projects fail. To get AI to work for you, you don't need more tokens. You need better people. A board pairs powerful proprietary tools with senior engineers who've seen it all. That combination means your project doesn't stall, doesn't drift, and doesn't fall. It ships. Whether you're a startup that needs to get to market or an enterprise with complex legacy challenges, a board delivers exactly what your business needs fast. Abord is your partner for AI transformation. Visit aboard.com and let's build something together. Welcome to Big Technology Podcast Friday edition, where we break down the news in our traditional cool-headed and nuanced format. We have a great show for you today. We're going to talk about the warning signs for the AI boom as companies start to question all the tokens they're spending on things that may or may not be shipping. We're also gonna discuss anthropic passing open AI as the world's largest AI startup. They're coming in close to a trillion dollars in their latest fundraise announced this week. And also, Robinhood is going to let you basically have your trading delegated to a chatbot. Is that a good idea? We'll cover it all. Joining us as always on Fridays to do it. It's Ronjohn Roy of Margins. Ronjohn, great to see you. Good to see you, Alex. We cannot have nice things . As we get into today , I and talking about AI and token maxing, it's just a reminder that we'll we can none of us can ever have nice things that are just given to us . Okay, well we definitely this is definitely a topic that will require a level of nuance that I don't think is being communicated in the headlines. So in the first half today, we'll do our best to at least tackle this with a degree of depth that I don't think has been shown in the conversation yet. Okay, here's the story. This is from the Wall Street Journal, capturing it all. Corporate America is starting to racian ? Ration . Ration. I should go with ration. Corporate ration ration is food. Right? All right. I'll I'll start that Corporate America is starting to ration AI as cost skyrocket. Use of artificial intelligence by big companies is exploding, and the soaring cost has some of them pumping the brakes in a way that could complicate AI's triumphal march across the economy. Some enterprises have hit their annual token budget in just three months or reported seeing their AI spending bills double or triple. Now corporate leaders are scrambling to bring down expenses by finding ways to ration AI use in their organizations, steer workers towards cheaper homegrown tools, and help them hone their skills to improve returns. Ranjan, I just want to kick to you. Basically what we're seeing is story after story of companies um that could include Uber, it could include Meta, it could include Microsoft, Doordash, that have talked about the spending of tokens has gotten way out of hand, right? Just the people are using these things to either get on the top of leaderboards or they're using high throughput models to do stupid tasks and just burning lots of money based off of it. And I think underlying all this is a question of wow, the you know, the revenue in this industry has run up really quickly. Um is it all a mirage? Is it basically a bunch of idiots token maxing to get to the to p of leaderboards where um where the actual value is much more minimal than what they're seeing and if that's the case, you know, could this all implode or slow down? What's your reaction? This is why I started this episode, Alex, saying we cannot have nice things because agentic AI, listeners know, uh is something I truly believe in and have seen the power of and get to work with every single day. But token leaderboards, token maxing at companies like Amazon and Meta already were making me feel a little uncomfortable. Then when you couple that with how that incentivized, soaring annualized recurring revenue, which led to more funding rounds and gets us into the situation we're in right now, I think at least I'm happy that it's being recognized , but I'm also unhappy with how quickly everyone is just kind of the pendulum is swinging dramatically the other way. And now suddenly everyone is saying AI has no value and no one has seen anything happen ing and this is all a mirage. So I think to kind of just start on the conversation of token leaderboards and what happened. Even last week I was saying this and again, for listeners, I work at a company writer focused on enterprise AI and have had a front row seat to all of this. What we were talking about last week, the last six months were this period of kind of unfettered experiment ation. And now we're seeing that kind of come to the headlines. And I've been saying this for a while now, actually, that no one was checking their actual Claude bills. Everyone was in the command line, you know, cranking out whatever they wanted in Claude Code and then codex. And now everyone's recognizing that, oh wait, maybe that's not how it should work. And I think it's a good thing. I think this was going to happen at some point. People should recognize there is a cost to all of this and that's fine. That's okay. You need to be thoughtful about how you build with AI. But I do think it's so much stupidity is coming out right now that it is it's kind of making a mockery of our indust ry. Well let's just talk a little bit about uh the magnitude of this. So this is from Axios going a little deeper. Um, it's talked about how uh Microsoft canceled most of its cloud code licenses and part over costs. Uh Uber CEO COO said AI costs are getting harder to justify. You also have Starbucks, um which had an AI program that users worked for automating inventory uh that it shut down and there is an AI consultant that said uh one of their clients this is amazing the stat really made the rounds. An AI consultant said one of their clients recently spent a half a billion dollars in a single month after failing to put usage limits on clawed licenses for employees. Here's my question. I'm trying to figure out whether like this thing, this notion of runaway token costs is the exception or it's the rule. I mean, I imagine this half a billion dollars spent uh is first of all like un named. So would have liked to have gotten a name on that, although I understand why it's why it's tough for someone to put a name on such a claim like that. But it when you look at some of the other the other examples, you can start to peel away um, you know, some of these. So Starbucks, like it had this inventory tool, um, but it was a it was a visual intelligence, right? So I think that that was mostly a computer vision tool um with with Microsoft canceling its code closed code licenses. Yes it was financial, but also Microsoft has a competing tool. And so it sort of leaves you with with Uber and an unnamed source. So I I don't want to say that this isn't happening. Clearly it is, but I think I'd like to see a little bit more smoke uh to start to extrapolate to the entire industry ha you know being on fire as opposed to uh what we've seen so far. And I think there is this notion that like if you get one of these stories, it just blows up. Uh and you know, remember we've talked a little bit about how att itudes against AI are, you know, very negative right now. And in this type of environment, a story like this just kind of booms around the internet because it sorts starts to do some confirming of the preferences that people have uh for this stuff to go away. I I'm just saying that for the examples that we see, at least in regards to how prevalent this is, um, even though I've heard of many stories of companies with leaderboards, I would certainly um you know apply at least a tiny bit of skepticism here in terms of this being a widespread problem. Your thoughts? I think it is both. I think and I again I see firsthand, I I build things that I get to see work . But uh this is all also true and correct, I do believe. And I think let's take these one by one. Um, but again, overall the idea that there has been a lot of wasteful spending specifically with clawed code. And I think that's like the main culprit and that's what's coming up here because as a when you gave engineers unfettered access to cloud code and no one was monitoring anything and you could not actually see how much you were burning and you're being incentivized in many cases. Of course you're going to. That's like that's the entire system that you're setting up. But I think let's take them one by one. Obviously, the the Axios reporting that the consultant said that one of their clients recently sp ent half a billion dollars, $500 million , in a single month because they didn't put usage on limits. Now my favorite part of this is like there's not a lot of companies where that can happen and fly under the radar. There was also uh reporting that in meta's AI token leaderboard, one of the employees had used 60 trillion tokens, which would be actually be 90$0 million dollars at today's API pri ces. So I'm just saying if you're trying to guess who could it possibly be, there's there's a pretty finite, you know, the universe of companies that it could actually be, but well by the way, you know, it's sort of interesting, because you can actually take out a number that you might be thinking about, right? You can remove Amazon, you can remove Microsoft, and you can remove Google. And you certainly can remove Apple. Right. So it kind of leaves us with meta. You certainly can remove Apple on all that. So I I think but the so then you start to think, does that $500 million anthropic is getting to count that probably is six billion in run rate. So as their run rate is increasing, these dramatic levels, and like to me, that's one whole side of this conversation that is terr terrifying and is problematic, is the extrapolation of these kind of like big but still, you know, like isolated issues has been extrapolated into ARR and now fundraising. And we're going to get into anthropics fundraising this week. So I think that's a huge issue. And to me, the that it is crazy, the kind of like second and third order effects that one example like that can actually have in terms of fundraising. And suddenly the it's like the butterfly effect where suddenly a 27-year-old Korean is taking out a margin loan to buy more sand disk stock on the local stock market and then buying a Ferrari and there's reporting that like Ferraris are being sold all cause a meta leaderboard or unnamed company leaderboard, someone was just cranking out tokens. Like that to me, that's one of the most fascinating and crazy parts. Do you do you think I'm exaggerating the potential downstream effects of these kind of isolated issues? Or do you think are tied to the I do think you're exaggerating the downstream effects. Like there there's a specific problem. I mean I did title this episode Warning Signs for the AI boom for a reason. And uh I'm gonna get into the specific reason in a moment. But I and you know there could be and we're gonna also talk a little bit about the weird creative accounting and circular deals in a moment. But when you think about the revenue of anthropic, it's gone and this is uh uh the a the annualized revenue, so take that as it with for what you will. Um but e even if you know, you still had accounting tricks, you couldn't fake all of this. In january twenty twenty five it was a billion, May twenty twenty five three billion, june twenty twenty five, four billion, August twenty twenty f five,ive billion, October seven billion, December, eight to ten billion. This year, February twenty twenty six, fourteen billion, March twenty twenty six, nineteen billion, April thirty billion, May 47 billion. So again, like even if you were to discount like that, let's say that 500 million use is not just that 500 million. Like it there are moments in these cycles that you don't forget. And about a month ago, in two I think I had talked about this on the show a few a month ago, two separate instances with fairly high level technology folks that I'm speaking with, they are bragging about how much they're spending on Claude. Bragging. Like you don't hear people bragging about those kind of that like you're bracket, bragging bragging about operating expenses is not something that's typical in business. And that the fact that that was happening, and when you see that curve in terms of their revenue increase, it is real. It is like actual, I mean, it's not truly annualized revenue. It's or annual revenue. It's ARR. Um, but it's coming from somewhere and basically, people with every IT team in the every large company was given, and Cloud Code was a phenomenon. I mean, it was. It was like to truly bring a gentic to the world. Like it it and it is but how that got reflected into actual like numbers was i do think this directly reflects and then the downstream, the twenty-seven-year-old Korean taking out a margin loan to buy stock and then buying a Ferrari, I still think is directly correlated. There was a story about this guy that you read this week? I I'm actually conflating two different stories. One about uh 20-something Koreans taking out margin loans to buy memory stocks. And also apparently like the Ferrari dealerships have sold out inventory for the next two years in South Korea as well. I just put those two together in a convenient narrative. I recognize the conflation. Okay. Well at least we're honest about it. All right. So so here's what here's the kind of the punchline that I 'm getting to. Um to me, if there's there you know, I don't think the token maxing stuff is the rule. Like I don't think we're seeing forty seven billion of ARR that's just being wasted because people want to rise leaderboards or people want to brag to their friends how much they're spending. I would say that probably accounts for a percent of what's happening, but not all of it. Here's where I'm actually concerned. And I and I will caveat this by saying, I do believe in the underlying technology. I think it's making good progress. We've all seen it make progress, right? But in order for this to continue, it's going to have to show actual productive use and let's say it's 20% token maxing the other 80% of people being cautious and trying to be productive with it you know, within reason, not just trying to burn tokens for burning tokens sake, they're gonna have to see a return on their investment. This is from that Wall Street Journal story that we read at the beginning. For companies using advanced AI coding tools, only eighteen percent of spending on tokens is translating into shipped coding products that reach real users. According to Intelligence AI, a startup that aggregated data of more than 2,000 companies using AI using advanced AI for coding. So here's where where my concern really comes in. You know, let's say you're talking about the eighty percent that's not token maxing. And by the way, I think that's a low number. I think it's probably higher than that, of people trying to actually use these tokens to accomplish things. 8 2% of that use is not translating into ship products that reach real users. And that to me is the issue. Now it was bound to happen that in this moment where there's gonna be some frothy spending because of the potential, that you are gonna see some waste. But when 82% is wasted, that to me is the bigger concern than the leaderboards and the token maxing. And that's where I think we're gonna start to see, you know, if we see a real reckoning with the generative AI boom, it's going to come in that area in particular. And that to me is a flashing red warning sign. Your thoughts? Okay. I'll I'll give you that is more of a concern. The token maxing is kind of funny , sad , but I agree. It's not it's still isolated and this is the bigger issue. Where I, even though again, in this conversation, all of this I have found very problematic, and I think there's like just the the hype and the scale which with with which everyone has kind of like chased things during this cycle has terrified me. This is an example where this is new technology. And if again, why I started this episode with we can't have nice things, if everyone was allowed to simply start building, understand what's working, not be overly pressured, spend some money, see like you know,, take what works and then build more on that. Cause like when you say 82%, what should happen, this is only four to six months old, call it. So that 18%, then you reinvest in whatever kind of work is being done there. And it's actually would any kind of standard business process that's not an unre asonable path. We see it ourselves again, like the reason I truly believe we can get there is like even things that I have built with our customers, we see like, okay, you start with one process and it's like taking X number of tokens and then oh wait, instead of using a giant CSV file as context in the Gentic workflow, let's turn it into a JSON. Actually let's chunk it into multiple different JSONs and only call like the getting that technical, but like that dramatically would reduce the actual tokens consumed in the by 70, 80% in a process. So like this is the stuff you should be doing. This is stuff we are working on and trying to do, but then all of this hype around it makes it makes it incredibly difficult. So I think like I I do agree that's probably even more real that 82% is wasted. But I think in any normal business cycle, that would be fine because any new experimental technology, that's part of the, that's part of the process. Right. And and I think this is definitely happening across the board. Um, and it brings me to this like uh story in Business Insider that Uber's COO says it's getting harder to justify the money spent on AI. Um this uh the chief operations officer of Uber, Andrew McDonald, uh said that based on talks with Uber's senior engine engineering leaders, that higher token usage did not translate into proportional increase in useful custom consumer features. That link is not there yet, right? He said. I think maybe implicitly there is more getting shipped, but it's very hard to draw a line between one of those stats and okay, we're now actually producing 25% more useful consumer features. Um there was a pretty interesting reaction to this online. I think that like the AI critics um, you know, pushed it out there being like, see, it's all bullshit. And then the AI boosters like pushed it out there and said, honestly embarrassing for Uber and this is a skill issue, not a technology issue. But I I would just kind of draw it right down the line here, which is that if this is a problem that's happening for Uber, it's a problem happening for many other companies. Right. If Uber can't figure this out, uh then you would imagine that uh that line again about the 82% of tokens being wasted isn't just some like AI consulting firm , you know, sort of pulling a number out of its butt. It's something that's actually, actually real. And to me, that's that that to me again is sort of where you could see if you were going to see like a real pullback, it would be this sort of reckoning of like, oh goodness. Um, we're using these bigger models uh and you know, we're not seeing the impact that we were hoping for. It is Uber and I guess I mean in terms of like dramatic innovation from a company. Ronjon, they're a tech company. That's what I'm trying to say. If Uber is not seeing the productivity, is Lowe's seeing the productivity? Well, Home Depot? I mean if Uber can't figure it out, are you going to Home Depot's chatbot's actually pretty good? If you use uh I think it's called like the orange apron, I was kinda stress testing it, but the the that's a separate thing. I I think we're gonna see a pullback and we are gonna recognize but but structurally I mean that this is the thing. Structurally when you go to your entire engineering team and say, go use as much as you can of this when there's no clear direction. And this is over a four-month period. It's over a four month. I mean, in the grand scheme of things, that's not a ton of time. And if things aren't being done strategically and everyone is just kind of doing whatever they can, obviously you're not gonna see any kind of like, you know, like concerted uh progress ROI, whatever it is. I think like to me, this just kind of captures a whole thing. The Starbucks one too. I don't know, like uh that you'd we'd mentioned before from the reporting. This was such a perfect example. Like everyone's like, oh, they're scrapping AI. New CEO came in in September 2024, wanted to do an AI kind of press release. You'd mentioned computer vision. It was basically like, you know, taking photos and then trying to analyze those photos and it like analyze the the milk, syrup, the sauces and an inventory count. That's a hard problem to do in kind of an uns unstable environment like the real world and in Starbucks of all places, which is a pretty chaotic place often. And the idea that you're going to solve that versus demand planning, inventory forecasting, like these are pretty advanced AI fields. And like , I mean, there's many, many companies out there. There's many. But so this gets that one to me does get cherry-picked as it's a press release type thing. They take on a really, really complex problem. And of course it doesn't work. I feel the Uber stuff s feels similarly where it's just like use a lot of AI and then of course in four months you're not going to see dramatic improvement. One more bit of of nuance here that I think we should share. You know, you remember when the Uber there was somebody from Uber that said like, oh we hit we blew through our entire twenty twenty six token budget, you know, in in like two months or three months and everyone went crazy over that. that And's sort of one of the things that we're seeing here is like, you know, executives are are coming to grips with how many tokens they're spending. Well, Simon Wilson had like a pretty interesting perspective on this. He goes, Some of the most widely cited of these stories appear quite over blown to me. Given that Claude Code really only got good in November, it's entirely unsurprising that a budget set in 2025 may have failed to predict demand for that tool in 2026. I think that's a good point, right? That we all know that in November, December is where these models took a leap and became much more useful for things like autonomous coding. And so we should expect to see many more stories of companies hitting their budget allocation or their bud their token budgets uh early and trying to figure out what to do. But in in some ways that could even be like an indication that the technology is working well. What do you think? I guess I mean but that that's exactly what I was saying, that this is all so rec ent that it's uh like it's clear to me that trying to get any like long-term learning from just three to four months, this exactly like I think is is not correct. But but I I I do think that I don't think that actually shows that there is value. I think it just shows I I mean these are I think these are all great tools and as we said like I think uh it requires learning it's gonna require like completely changing organizations and the way they work. That's gonna take time and it's not gonna get fixed in four months. So I don't think again, anyone who has used these tools, you can just sit there, go down rabbit holes, crank tokens, and end up nowhere. Plenty. Like and we've all tried to build some random app that doesn't really go anywhere and it's fine. But if you're both pressured to do that and like that's being done at a large scale, we're seeing the results. I guess there is this kind of third way here, which is that um these tools are actually quite useful. Uh, but engineers have figured out a way to just courta kinda um set their jobs on autopilot and kinda hang out and build the same features that they were test. So the tokens are doing their job, the engineers are gaming the system, the higher ups aren't seeing any productivity um and everything can sort of be explained as like uh the engineer has hacked their way through the system with this brilliant new technology. Which I wouldn't be completely surprised uh by you know, to learn that that was the the real story. I I support that. I support that. Okay, so bottom line from this segment, I'll speak for myself. To me, you know, the core thing to read from this story is basically um the eighty two percent. AI's gotta not waste eighty-two percent of the tokens in order for this boom to continue. And that to me is the real warning sign here, um, those type of numbers. If it doesn't figure that out, there's gonna be some serious problems. Your thoughts? And I will you're go ahead. I will push back that that eighty two percent actually I believe could be a reasonable thing in any early stage of like a pretty dramatic new technology. And I'm saying like the world changed in November 202 5 and it's like uh that 82% is okay. If it's a reasonable thing, it's if it's still thought of as let's experiment, let's learn, not let's just plow money into it, which is what people did and and imagine that everything is going to be different. So for you so for you the real problem is the twenty percent that's being token maxed, assuming my eighty twenty breakdown is No no for me is we can't have nice things that if everyone just approached this in a nice responsible way and just built and just learned and took took what wasn't working and discarded it and took what was working and then invested and doubled down on that . Like any other thing, we would be okay. But it's as we get into the next few segments, we're not okay. Yeah. And I think that this is something that we need to get into as well, which is that the financing still looks a little wonky uh here for AI. And I think there was a long discussion of the circular financing uh and that kind of went on pause for a while. And we've seen these amazing cloud revenue numbers over the past few quarters with Google and Microsoft and Amazon all booking insane profits from their clouds divisions. Uh but at the same time , they've also all of them have invested in AI labs, Microsoft in uh OpenAI and Amazon in OpenAI and Anthropic, you know, tens of billions of dollars and Google and Anthropic and also in its own divisions. And then when the uh you know when these startups spend these massive amounts uh on uh the computing that sort of has been tied to uh the deal, you know, the computing from their funders, it gets booked as revenue, and then they show profits, and then, you know, the cycle moves about again. So Rajan, you brought this up to me in our uh in our text messages Well I think again, going the the going back to why we can't have nice things, you can't build responsibly and take a deep breath because of this insane like infrastructure that that's been built around circular funding. And I it it reminded me, I was trying to find, I think it was almost 18 months ago on the show, we were talking about it was like one of the early stage, I mean, and it was still six billion at the time, but where it was explicit that a lot of this is in cloud credits. I think it was Amazon first going into anthropic, um, and us kind of half joking like, oh, I bet you it's just all AWS credits, and that's a and they're gonna call it funding. But now as the numbers have gotten bigger, I mean Microsoft invested 13 billion in OpenAI. A lot of that was just spent back into Microsoft. They recognize it as Azure Revenue. Then they also recognize a paper markup in OpenAI as well. Like every one of these, I didn't the scale kind of shocked me because I saw these numbers that last quarter Alphabet parent company of Google reported $62.6 billion in profit. Google is a cash machine, they always have been. $28.7 billion of that was a paper markup on anthropic. Amazon 30.3 billion in profit, 16.8 billion, same anthropic paper gain. So again, we all know this. Nvidia put in 100 billion into OpenAI. OpenAI is going to then be committed to buying NVIDIA chips. Like overall , this everyone has known this. It's like right the there's been reporting all along the way , but we haven't really really worried about what does that mean. And I think we're starting to get to the point of actually start what happens if there's a slight downturn. It's like the reflexive nature of all this money that's moving in a circle uh is terrifying. And again, like that's why we talked about this last week. I think everyone is rushing to go to op uh IPO. So this will just be in the hands of retail investors as opposed to these companies and their investors. Let me play a devil's advocate here. I mean, all right, let's say you're Google, you have your big investment in anthropic, you invested it when it was like a billion dollar valuation, now it's a nine hundred billion dollar valuation. Why should you not mark the uh the markup in your shares value as profit No no you uh you you have to. I mean you you you actually have to. Like so I I don't think they shouldn't, but uh it is still I mean it's still the case that that is half their mar the nearly half of their overall profit. So I think like that's a I'm not saying this is like an unethical thing in any way. I think I guess uh if we're gonna say anything unethical, it's it's not even unethical. I think like I don't think there's like a cabal of tech executives sitting around talking about how can we juice our overall valuations and then build this circular financing system. Actually, what's fascinating to me is I think this really is like a good example. It's still such a small circle of people . Like in the grand scheme of things, it's in the hundreds of people maximum across all of these companies that are kind of at the forefront of this, the poster children for this. And probably in s rel atively similar conversations and social circles and conferences and sitting at the same tables at industry dinners or the Trump White House, what have you. You know, like so you can feel how the conversation in that group thing can kick in. And then suddenly if the guy over there is tossing in an investment that actually goes right back to them in cloud credits, why why wouldn't Right. Well you get that, but then you also have this promise of future spending, right? So it's not just like Microsoft putting money into open AI and then open AI putting it into Microsoft. It's Microsoft putting money into OpenAI and then OpenAI promising to uh you know, to basically buy c computing services from a Microsoft or an Amazon. I mean think about this. Uh this is again from a tweet that you shared. Microsoft had four has forty-nine percent of its sixty-two six hundred twenty-seven billion future backlog tied to open AI. Oracle has fifty-four percent of its five hundred and fifty three billion pipeline depending on open AI alone. There's no guarantee that open AI is going to come in and actually spend that money. So when you look at the financial health of these companies, um I don't know, you know the AI could be a blessing, but also this dependence on an open AI could also be a curse. Well, not just a curse, but I mean you take even again, maybe Apple comes out of this just golden, that they're the the only ones who like somehow manage to weather by purposefully stay out of it. I think like those that that kind of forward looking spending, and this is all in the next three to five years that everyone has kind of like built these contracts, it all assumes an end demand from the rest of the world outside this small group of companies. And like , are they going to be interested? Is it going to work as magically as everyone has been promising? And again, that 82%. Like, I think that's why this is caused so many alarms. Like and then also I mean, one of the other interesting things to me is and w we can get into the IPOs, but I think what was SpaceX 's estimated fundraise? It's like ighty86 billion.. E Eighty bil billionlion . Eighty plus billion. It's between I think I saw like Josh Brown on CNBC called it like three asteroids coming to hit the Earth. Like the amount of capital that's gonna be required from the rest of the world outside of sovereign wealth money and dragonier and altimeter and whoever else like to actually fund these three IPOs. Like where is that money coming from, other than the South Korean twenty-seven year old taking out the margin loan for all right? But let's talk about that though, because this is fascinating, by the way. So uh very briefly we should cover this. Do you I mean, it's not just these handful of companies anymore. This this boom is spreading to the memory uh chip companies. It's from the Wall Street Journal. AI has made memory chips more valuable than oil. Memory is now worth more than oil. Staying that way will depend on how much the notoriously volatile chip industry can make recent changes stick. The world's three largest memory chip makers, Samsung Electronics, SK Hynix, and Micron Technology, now carry market capitaliz ations of more than a trillion dollars each, that's 22% above the combined market cap of the world's three most valuable oil companies, even with Saudi Aramco weighing in at its own nearly $1.8 trillion I mean, just talk a little bit about this boom in memory trips, right? It just seems like um the AI companies are buying up all they can get, and they've made a, you know, sort of secondary or tertiary group of of sh of shareholders, those that have had uh the memory chip stocks very rich while also making like life difficult for uh anyone who owns a camera and wants a new memory stick because OpenAI has bought it. Your thoughts on this? Well th these are my second and third order effects. In the that that 500 million on the token maxing leaderboard earlier, here's where we ended up. This is what I'm saying. It's like, again , the idea that the more and like this is something I thought about for a while. And I mean, as AI produces more data and content and stuff that needs to be stored. It was always inevitable that memory would be more important . Um but to see two companies re ach market capitalizations of more than a trillion dollars. Three. Three companies. Three companies, sorry, three companies. Like, come on. That's bananas. That will fall. It has to fall inevitably, right? I mean, not investment ad vice but that has to fall but but this is where i had uh friends like in in the investment community the question they keep asking me it's funny like again i work in enterprise AI . They ask me, what is the next bottleneck? Like that's the conversation is not, oh, you know, like how's adoption? Like what are you science question? What yeah, what are you like thinking about how companies are gonna leverage AI? Are you seeing the ROI from a like our business operations improving? It's literally where do you think the next bottleneck is? Because that's that is where we are again in the cycle that the mania is so strong and everyone missed if you weren't if you didn't you missed the memory chip bottleneck what's going to be the next i mean is it like it's like cooling uh for data centers and stuff like that. Like there's all these kind of like really minute or niche areas that serve the entire value chain. And everyone is trying to look up the next one because they've seen what's happened. So is that a healthy market? I don't think so. No. Do you? But on the other hand, like maybe there's real economic forces there of like maybe you actually do need all these memory chips to make this I'm just glad I got a four terabyte hard drive like last fall when I was trying to export all my Google photos, but never successfully did because it they make it a real pain. But I 'm Soon enough, Ron John, you might be able to. All right, um, let's go to break. But before we go to break, uh good news, Ron John is going to be joining us in person for the Big Technology AI Summit. In San Francisco on June eighte,ent weh have just a few tickets left, so if you have any interest in joining me, Ron John, opening app president, Greg Brockman, uh Semianalysis, President and CEO CEO Dylan Patel, uh Aaron Levy from Box and Arvin Turnivas from Perplexity, Lauren Good from Wired. It's gonna be a great day. Definitely should join us. Um the day's gonna run one PM uh till five and then we'll have a wine reception with some food on the roof. So um it's really coming together nicely. Just a few tickets left. Hope that you will join us there. Um so just go to summit.bigtechnology.com and uh we'll see you on June 18th, back right after this. Summer always changes how I get dressed. I want pieces that feel lighter and more comfortable but still put together. That's where Kints comes in. They focus on well-made essentials and breathable linen and soft organic cott on. The kind of basics you will keep coming back to. Everything at Kintz is priced 50 to 80% less than similar brands because they work directly with ethical factories, cutting out the middlemen, so you're paying for qual itity , not brand markup. Personally, while I'm moving between air conditioning and summer heat, a lightweight zip-down sweatshirt from Kintz has been a life changer. Elevate your summer wardrobe. Go to KINS.com /slash big tech for free shipping on your order and three hundred and sixty-five day returns. Now available in Canada too. That's Q Uin C E dot com Slash Big Tech for free shipping and 3065 day returns. Kins.com slash big tech . Most leaders know how work is supposed to happen, but when it comes to how it actually gets done day-to-yda acrosss tool, teams, and handoffs, they're mostly guessing. That's exactly the problem Scribe Optimize was built to solve. Trusted by over 80,000 enterprises, including nearly half of the Fortune 500, it gives leaders a live view into how work is really happening across approved business apps without interviews, manual process mapping, or extra effort from the team. And because it's continuously analyzing real workflow activity, the insights stay current instead of going stale the moment a process changes. You can see which workflows are happening, where time is going, and which tools are involved. It automatically surfaces top issues, explains why they're happening, and even recommends ways to fix them with estimated time savings. And importantly, it's built with privacy in mind. So activity is only captured in admin approved business apps and user level data is anonymized by default. The kind of visibility that used to take months, now it's just always on. If you're ready to stop guessing and start seeing, visit scribe.how slash big tech. That's s c r I B E dot how slash big tech. Last month, Mary's team did something unheard of. They closed the books before Friday. No late nights, no missing receipts. So what happened? RAMP happened. RAMP automates receipts, enforces policy in real time, and gives teams total visibility. That's why it's used by 40,000 companies like Shopify and CBRE. Finance te ams, want your weekends back? Get two hundred fifty dollars off when you join at ramp.com slash trial. That's ramp.com slash trial. And we're back here on Big Technology Podcast Friday edition with Ron John Roy of Margins. Ron John, uh the day has come. Anthropic is bigger than OpenAI. This is from the New York Times. Anthropic tops open AI to become the world's most valuable AI startup. On Thursday, Anthropic punctuated its ascent by officially passing OpenAI as the world's highest flying AI startup. It is going to it raised sixty five billion dollars in financing that values it at nine hundred billion pre-money. Uh what do you think about this? I mean is this mostly symbolic or is this the day that uh Anthropic truly has passed open AI as the top uh AI lab? I wouldn't count out OpenAI just yet, because I wrote about this in 2022, I think probably spring. There is the altimeters of the world, the dragoners of the world, Sequoia as well. Like a lot of the late-stage financing game of pre- IPO doing one last round essentially as a signal to the market of this is the value worked very well for a while back in 2021 and 2022. I think that's what's happening here very clearly. And and again, it's like I mean, do you remember the days when the valuation associated with the fundraising round was actually a highly guarded secret? Back in that happened. Twenty fifteen. By I think even like twenty seventeen, eighteen, it wasn't a reported thing. It was like reporters would have to like dig to get valuations. Like you hid that on purpose from like a competitive standpoint. And so like I mean, we saw this a lot. I think that's what's happening here. I mean, you have this incredible growth and then you wanna get out to market and you wanna signal to the market, here is where we believe the market, the valuation is so when you go to IPO that you can actually try to actually kind of get that pricing like anchored to this because now in the conversation, the anchor is what is it, 900 , 900 billion. Yeah. Yeah. I mean I think so just to sort of zoom out a bit, I mean, I do think this is a remarkable moment for anthropic, which a year ago there was I mean that they were their last valuation was like 350 billion uh 350 billion. A year ago, if you would have told us that they were gonna be worth more than open AI at this point, um, I think we would have been stunned. And it's largely come on the back of Cloud Code. So they're clearly the hottest company in AI right now. They do have this sort of straight shot to being the most successful IPO of the bunch. Um and I mean it maybe excluding SpaceX , uh, but I wouldn't even say a SpaceX is an AI IPO. And so I don't know. I I think that this is an important milestone. And it just shows how locked in a battle OpenAI and Anthropic are right now. I mean, OpenAI really re needs to respond with this Codec Super App and Google, my goodness, I mean doesn,'t really seem to be factoring. Uh I I mean, obviously they are with the with the um cloud business and that cloud business is growing, uh, but not in this sort of uh autonomous coder in the way that uh openai and anthropic have really rode recently. I I actually what's even crazier to me is that uh openai had even raised what was it a hundred twenty 22 . 122. And only got to 852 post money. So like, yeah, they uh they got some work to do. However, I don't know. I'm just waiting for the S1s. Like uh Space SpaceX . Yes, SpaceX . What to their credit, the S1 came out. It was and we talked about this in length at length last week. It was almost shockingly not good. And still everyone is just it's 1.8 trillion. It's that that that is see again, the value of anchoring. It's a way, it's one of those like uh, you know, like mental principles or whatever it is like the anchoring effect just say it just say it over and over and everyone assumed right now anthropic 900 billion that's the valuation because the same people have been investing in it the whole way through and are the ones poised to cash out if people believe that are saying that it is. It is so. That's a nice conspiracy. I mean, you're right. It's not a conspiracy. It's like true, fake it till you make it. Silicon Valley style. No no but i uh it's not a conspira it's like it is just fine. Damn it, that's what's happening. No no it it's just no no, it's just fine it. Like why wouldn't you do it? Why would like what possible reason would you not if you set the valuation? Because that's a typically a higher valuation is supposed to be against invest ors, right? Like you're gonna fight for the lower val uation. So you have more of the company. But in this case, you have every incentive in the world to juice the valuation because that means that it gets anchored there and then you go out at a higher value like then you realize that uh at the expense of the retail market. So why wouldn't you? You know what would really be interesting if they went public and then uh everybody on Robinho od, which has uh Claude and Chat GPT controlling their trading, just went full hog into the IPO because uh we had a headline, very interesting headline this week from the Wall Street Journal. Robinhood lets customers use AI to trade stocks and make credit card purchases. Robinhood is launching a new feature that lets customers hand their trading and credit card purchasing decisions to their favorite artificial intelligence tools. Robinhood users can link an AI agent like Anthropics Claude or the coding agent cursor to separate uh to a separate dedicated investment account. There the agent can access the dedicated funds and place trades as directed. For instance for example, users might instruct their agents to root out risks created by being overly concentrated in one part of the market or monitor a basket of promising semiconductor stocks. Notably you cannot trade options this way yet. Uh oper ative word yet. Ranjan, I'm sure you love this. I mean, this is in the the grand scheme of the things and the kind of like this is where we need to go. So thank you, Robin Hood, for agentifying trading for all of your millions of customers so that they will somehow end up with SpaceX in their portfolio because it's the only way this story can go. And I think I'm actually, I don't know. I'm I'm happy about this . Not not truly happy, but like from a pure narrative standpoint and uh as the script should be written, Robin Hood kind of like finishing this out is perfect to me Can I make an argument in favor? I mean, I wouldn't put your entire portfolio in this thing, but to give it a few hundred dollars and say, can you sort of uh come up with a strategy based on these principles and go out and trade for me. And it can like, I mean, most 95% of day traders lose, but what if you sort of had an AI day trader that was going out and in sort of synthesizing so much more information than you possibly could and paying attention to the second by second and minute by minute shifts. Maybe that could work. I don't know if that's a behalf. No, no, no. I'm gonna okay. I'm gonna come around. I'm gonna come around and say I do think there is I agree like actually probably an agent should be able to trade better than a human, like a an everyday person who's not like really focused on this, actually setting some parameters and letting it go. I'll I'll give it to you. I think again, this is again in my my new Grant's theory that we just cannot have nice things. This actually makes total sense and I like support the concept of it, but Robinhood releasing it at this moment , I just it's I can't see a good outcome of this, but in theory it makes sense and maybe this is the way investing. I mean, this is what like the betterments of the world and more of the kind of like algorithmic wealth planning we're supposed to do. And it's just the next generation of that, I guess. Yeah. I mean my hot take on this, and I wrote about this in big technology today, is that everything is going to go this direction. And it's not gonna wait until you get to Robinhood. The second you start researching stocks in Chat GPT, we're gonna get to a place where ChatGPT or Claude will offer to build a strategy for you . We'll have access to your bank account. We'll ask if you want to portion some money towards, you know, giving this um this strategy a try. We'll come back to you and be like, here's how it's going. Do you want to invest more? Um and I think similarly with everything that you chat with these bots about, they are gonna just try to intuit what your next move is. They don't want you to go to the Robin Hood. They want you to stay, you know, within their chat experience and let their computer use bots and agents go out and finish the rest for you. That's my perspective. No I think they already are in some ways.. Yeah And we're seeing the very early stages of it where like uh more often than not now, Claude, but also Gemini have seen been chat GPT not so much, like it'll go build you an entire interactive experience when you're like, dude, just like what time is the Spurs , uh the Spurs Thunder game or something like that? Like here's an entire HTML dashboard and a website that can do this for you. So you're already seeing it try to go above and beyond. And uh yeah, the more access it has, I can imagine the more it'll try to do. But if it's good, it should do a good job of that and help you anticipate Yeah, this will happen. I mean one interesting thing that I did this week was I was uh trying to research if I had some um like uh special number you know associated with my business, like a like not an EIN, but like something to to that nature or to that degree. And I asked Chat GPT, hey, do I have something like this? Like would a company like mine have something like this? And it goes, well Wait, I saw it. Oh I I said yes. I said yes. And then it went in, it found my incorporation documents and it said, Nope, you don't have this. Better than Gemini, I'm sure. Yeah. Oh I should go I should connect it and ask that my c question what's my I have to say Gemini. I also did it and I mean again, like take the privacy concerns uh into consideration. Which are But I was like, oh how much did I pay for this flight? And it like went into my Gmail and it got the ticket price. It is It's crazy. Yeah, but then it will take the next step for you, right? And and you know how sometimes it will draft an email in ChatGPT? Once you connect that Gmail, it's only a matter of like, let me draft this in Gmail. All right, forget about drafting it in Gmail. How about I go and send this to you? You know? Is that the case Uh I mean I just think yeah, it's sort of it's effectively my duty to test this out and come back to our listeners and readers. Well that was more of a segue and a lead-in too . And therefore Go ahead. Would you be willing to allow a stranger into your house to clean your house in exchange for them recording that entire cleaning session and providing that real world data to an AI company. Because that's happening right now. That is happening right now. There's this AI training startup shift wants to clean your home for free, but the they will have record cleaners as they scrub vacuum dust tidy wash and use that footage to train robots. So it'll be an actual human coming. They'll clean your apartment for free. And in exchange, they get to uh get all that data and use it to train robots. And now they, as part of this announcement, they uh indicate that like somehow they will I'm sure they will, uh scrub all personally identifiable data or anything private. Uh they want. Which I mean, come on. difficulty of that is actually like I mean it's that's a very difficult problem to solve in itself. There's gonna be a headline like in two years from now. Apartment free apartment cleaning AI startup stores videos of people Every state Yeah, even all the robot vacuums have that. I mean I personally wouldn't but I'm sure people hold on no hold on but that would assume this is an actual human coming into your apartment with a camera. If you are still having an intimate act during this person cleaning your house, that's on you, my man. I mean yeah you may you may go to jail for that but uh that's totally on you. I'm just talking about if you have like some like contracts laying out that are getting scanned or something. Gotta think bigger, Ranjan. I think that's my advice to you here is you gotta think outside the box. So so one thing I'll say, I actually had tweeted, is this real? And the founder had actually responded , yes, we've already served some over the last few weeks, and globally over ten thousand contributors collected their skill demonstrations. I found this very fascinating because he calls them skill demonstrations, the house cleaning, because it's an actual human going and demonstrating a skill like vacuuming to a camera, um, which already was just, I don't know, fascinating. in itself But I also realize that probably means you only get one cleaning, right? Been demonstrated in your space. Wait, would you we're we're running out of time. Would you would you do this? Like are you tempted to allow them to come in and clean your place? I am tempted from purely like I just want to see the person who comes into my house. I just want to like do they have a camera on their back? Is it like over is it a GoPro? Is it like is it a rig? Is it a rig of cameras? Is it like one of those uh like virtual reality motion capture suits? Uh street view in the middle of your living room. I I'm more curious about that. I would like to talk to them. I still am a little on uh it's less the data privacy I'm more concerned about you giving ChatGPT your Gmail than this random shift guy showing up in my house. But by extension, I should definitely invite the shift guy in for a call. I mean it's all gone anyways, so you might as well let the shift guy in. Yeah. I mean just on the w on the chat I mean, ChatGPT has my email, Google has my email, every tech company has my email. You know how many places I've signed in with Google? You know? No no you don't check every single permission that you're given when you do sign in with Google in detail? No. No, I don't know. I'm just joking Just like when the shift cleaner comes in and I have to sign something.