Limitless Podcast - This Week in AI: Anthropic Beats OpenAI, Deveillance, AI Farming

Episode Date: March 6, 2026

Anthropic has had a chaotic week, surpassing OpenAI with $20 billion in revenue and the launch of Claude Opus 4.6. We discuss OpenAI's response with the upcoming ChatGPT 5.4, financial struggles, and Anthropic's internal memo on Pentagon ties. We also break down the Deveillance privacy device and highlights AI's impact on agriculture and recent layoffs at Jack Dorsey's Block.------🌌 LIMITLESS HQ ⬇️NEWSLETTER:    https://limitlessft.substack.com/FOLLOW ON X:   https://x.com/LimitlessFTSPOTIFY:             https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQAPPLE:                 https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890RSS FEED:           https://limitlessft.substack.com/------POLYMARKET | #1 PREDICTION MARKET 🔮https://bankless.cc/polymarket-podcast------TIMESTAMPS0:06 Anthropic's Record-Breaking Week2:02 OpenAI's Response and New Strategies6:15 Dario's Controversial Memo7:41 Pentagon Drama and Its Impact8:52 ChatGPT Model 5.4 Teased10:47 Speculations on OpenAI's Hardware Device14:56 The Rise of Anti-AI Devices19:12 Privacy Concerns with Meta20:30 Exodus in AI Companies22:15 Innovative Uses of AI in Farming24:35 Weekly Roundup and Closing Thoughts------RESOURCESJosh: https://x.com/JoshKaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠

Welcome back to the Limitless AI Roundup. This week, a clear winner emerged from the Anthropic and Open AI drama. Anthropic broke records all week. They reached $20 billion in annual recurring revenue, up $6 billion in, I think, seven days. They crushed Open AI, dethroning them from the number one spot on both AppStores, Android and iOS. And they beat out all enterprise customers.

Their app downloads doubled as well in 10 days, just an insane week for Anthropic. But Open AI tried to fight back. They teased their art. upcoming 5.4 chat GPT model, which is going to be great, and a slew of other news items, including a brand new device that uses AI signals to jam and block devices, creating a new world of privacy around AI. We got to start with Anthropic. I mean, Anthropic, they've had the craziest come up ever. And it's weird because it feels like they had one maybe a month ago, probably around the end of December, early January. But I think the rest of the public now is

catching up for a completely separate set of reasons. So over the holiday, there's a was this huge thing that happened where all of the developers, they were home alone on their computer, well, home with their families, on their computers, spending a lot of time away from the workspace tinkering around with cloud code. Cloud code at that time was finally at a level in which it, I feel like people felt the AGI really, and developers were able to offload a lot of the developing and the coding to Claude Code. So developers switched over and they believed that Anthropic was number one as of December. Then I'd say, for me, it was probably around January where Claude Opus 4.6 or whenever that came out,

it started yielding much better answers than Chatsypti did. And they created artifacts, which I love. They had co-work, which I love. They were just shipping a lot of features that were superior to Chachypt. So then I switched over. And I know you've been using it a lot. And our producer Luke has been using it a lot.

And now in February into March, I think, is the third wave where they've got the developers, they've got the hardcore believers. Now they have the rest of the world. Because the drama that unfolded this week between the Pentagon and, between ChatGBT and Open AI and the war and Anthropic kind of getting rejected from the Pentagon for standing on its morals, but as a result, getting a huge whirlwind to become number one in the app store.

It seems like they have captured the entire market, the entire headspace, and a lot of these charts reflect that, right? Like, there's a lot of wins happening on this visual we're seeing. And it goes against what you would expect or what you would have thought would have happened. And last week was a crazy PR crisis for Anthropic. They got into a fight with the Pentagon. Trump called them like crazy woke company, and they got blacklisted by the government. Despite that, they're number one across both app stores.

And what that reflects is two things, I think. One, which is what you just described, they kind of crossed a chasm of intelligence where the models went from being this kind of cute little chatbots that you can replace your Google searches with to something that can actually be productive in your life. you mentioned that you and Luke are producer using it.

I've been using it nonstop. It is the number one AI model that I barely interface with chat GPT right now, right? And I used to use that every day. So it reflects not just our behavior, but a lot of the broader public. Not only that, but the app downloads have 4xed as well. But if you look at our friends over at OpenAI, their unsubscription, so the amount of people that have unsubscribed and undownloaded the app, has surged to 300%. So there is literally a clear trade between choosing OpenAI's chat GPT and Anthropics

Claw. And they're clearly one, but it's not just on the consumer level either. It's also with enterprise users, which is actually where Anthropic makes most of their money. Open AI earns most of their money from consumers through subscriptions and now advertisements. But Anthropic makes most of their money through expensive, large value enterprise contracts. And if you look over here in this chart on the top right, you'll notice that they have officially surpassed Open AI with that. So just an overall great and powerful week for Anthropic,

this chart over here from the information kind of shows that they still have a ways to catch up when it comes to pure revenue. Open AI is earning around $25 billion annual recurring revenue. Anthropic is earning $19 billion. Like I mentioned, it's up like $5 to $6 billion over the last week. But this doesn't really tell the whole story because Open AI is burning a lot of that money. Now, it says here that their cash burn is 115 billion cumulative burn. That's like for like the next four years or three years. But they're actually burning around $20 billion a year.

Whereas if you look at Anthropic, they're burning around $3 billion. So there's a clear discrepancy between the two, Anthropic being more efficient. So in theory, on paper, they're winning. That 3% burn doesn't make sense because if their revenue is $19 billion, they're not making money. Like, Anthropic hasn't made a dollar yet. So I'm not sure where the discrepancy in this is, but it's, clear that Anthropic is winning in a lot of ways. I mean, that that $5 billion that added to their run rate was in three weeks, and that's 25% growth on already the fastest growing

company in the world. And to your point on the different types of customers that they have, there's this pretty stark contrast between Open AI and Anthropic and the way that they're going about this were Open AI, they advertised, what, six months ago, the 800 million weekly active users, that number maybe hit 900, but it hasn't really grown a whole lot. And Anthropic has a fraction of that. And yet, Open AI is forced to introduce ads where Anthropic isn't because their consumer model is different. Their business model is different. I was reading this great paper that was talking about the average cost or the average revenue per user that the companies make. And Anthropics is eight times more. I believe Anthropic monetizes at about

$211 per user. And Open AI is only at $25. And that's thanks the strategy of selling for businesses. So they need significantly less customers to make more money. But now they're actually increased their customers too. So not only do they have the commercial business applications, but they just have general consumer trends that are going in their direction, and everything is setting them up for a huge win. I mean, OpenAI is set for $14 billion of losses this year, and by next year, Anthropic is totally set to stop earning cash. So there's a huge discrepancy in strategies. It seems like Anthropics is paying off for now. We'll see where that leaves chat GPT, but a huge week for the boys at Anthropic. Congratulations. This is like a massive win.

So you would think that Dario would be calm and composed and humble from all the success, but it was actually quite the opposite. There was an internal company memo that was leaked where he just goes off on OpenAI and the government and on all the drama that's happened. Some favorite examples here is he calls OpenAI's Pentagon deal safety theater. He accuses Sam of just gaslighting the entire Anthropic Pentagon deal, and he calls OpenAI's employees a gullible bunch. This really feels like Dario's venting

from the fact that he lost a $200 million contract with the Department of War, which is curious and interesting because his reasoning for that is he thinks that he lost the contract because he doesn't want to appease Trump's dictator-style leadership. That's like a direct quote somewhere here. And the fact that he just thinks that the government isn't conducive to his demands.

Now, there's a bit of. back and forth on this because Open AI to their credit exposed all the details of their deal with the Pentagon. And in my opinion, it actually seems pretty reasonable. Like, listen, it's, you've got to let things that are contentious go to a court of law. That's the way that the country of the USA is set up. And Darrow was trying to clinch onto more power that would put Anthropic as the decider as to when and what they use AI tools for. And for a case of national security, that's a bit of an issue. But a lot of that is kind of interesting. because some new news broke this week,

which is like Anthropic is renegotiating with the Pentagon. So not only did the Pentagon drama get Dario and Anthropic to the number one spot across all app stores, but now they're just retigning the deal that gave them that spot in the first place. I'm getting tired of following all of this drama. My God, it just keeps going back and forth. I know. It's a lot to wrap my head around.

But, I mean, there has been some interesting developments that are probably worth noting around the world of chat GPT, which is their teasing of the new one. already 5.4 sooner than you think. So Chat Chhabit, they've been in this cadence of releasing models much faster than usual. And in fact, this 5.4 teaser came less than an hour after the 5.3 instant model was announced. So it's clear that what they're doing is kind of reversing their strategy, which was just large model release after large model release into these incremental things. I think they've released five models in the last seven months, something like that.

Like they're starting to increase the cadence. And the plan is to continue doing so. There's a few interesting things in this leak that has come out about the model that we have outlined here, right? Like a larger context window at least? Yeah. So they've doubled the maximum consumer context window. Anthropic had the lead with Claude Opus 4.6 at 1 million context. And it looks like 5.4 is going to have 2 million context. To put that into context, there's a lot of me saying context here.

You can basically put an entire company's code base or novel into the prompt, and it'll understand it in one single inference or prompt request. So that's pretty impressive. But I think the real take from here is rumored to be persistent state, which is something that made OpenClaw, the personal AI agent project, really, really valuable. It was valuable because it remembered anything and everything about you. And with ChatGBT, GBT and Claude sometimes, you need to remind it. You need to say, hey, by the way, this is my preferences.

With an update like this persistent state on memory, you wouldn't have to do that at all. And the other thing I liked here is they're going to be using S-Rinds. which stands for static random access memory. The typical memory type that is used in GPUs today is HBM high bandwidth memory, and that is in large demand. It's what's resulting in a lot of these consumer electronics and GPU prices to absolutely escalate

because there's not enough HBM. S-RAM is a different memory bucket, which allows you to inference any kind of AI prompt extremely quickly, which is going to be a huge benefit. This is enabled by OpenAI's partnership or rather aqua hire of the company known as Cerebrus, which they signed, I believe, like, a month or so.

I don't know, it all moves so far. So, yeah, that's the highlights if this is true. The final thing I'll say is they've released like maybe five models in the last seven months, as you said, but they're all version increases. They're all 0.1 increases. So we've gone from, hey, we're going from chat chitb-5 to chat chit-t6 to 5.1, 2, 3, chat chit-td-d-d-in instant.

So it's a new kind of like paradigm of model releases. I think I may actually take the under on the context window bet of it being 2 million. I think chat GPT is very much playing catch up. And for them to go from, what is it, 250,000 context window or 500,000 context to 2 million, it seems like a big jump. So maybe I'll take the under on that one. But excited to try the model, we'll be covering as soon as it comes out. We also need to cover a follow-up to the craziest rumor. The thing that's been driving me nuts, it's been keeping me awake at night, is what the hell is this device on Joe Gabby's ears? on his table. For those of you that didn't watch your episode a few weeks ago, there was a Super Bowl

ad that got leaked from OpenAI announcing the dime, which was their new hardware device. It's the first hardware device in collaboration with Love From and Johnny Ives design team, the guy who has famously designed iPad, iPod, MacBook, all of the devices that we use every day. Now, what we're looking at is Joe Gebia. He is co-founder of Airbnb and also chief design officer of the United States of America. This is a government official participating in the signoff. So the first, the first The first instance of this was featuring an actor from Secession. I forget his name, but this favorite... Alex Sondis Scarsgard.

Oh, dude, you're on it. Alexander Scarsgard. That's the device. He's got the things around his ear. He's got this puck on the table. And then there was a leaked video, which is what we were just sharing on screen, of Joe Gebia, using this same device. It has the same ear cuffs.

It has the same puck on the ground. And it is leaving me kind of just spinning in circles as to what this thing is. Like, clearly it's something. clearly there's people in on it. A government official and a professional actor are both in on this. SIEF. What is this? I'm so confused.

Please tell me, what is this? I think the answer is obvious at this point. It is OpenAI's new device, and they are all under a strict NDA that they can't talk about it. Now, as to why Open AI hasn't officially confirmed this, there's a lot of speculation around this. My current thesis is they built the thing

and then wanted to fine tune it to make it even better. And two, they don't have the supply chain to scale it to millions of users tomorrow, right? Which is, they're aiming for like 40 million, 30 to 40 million within the first like couple of months. I don't think they have the infrastructure to do that. So I think they've delayed it for now.

The other reason why I think it's actually real is Joe Gavio's response to this post where he calls out the guy that videoed him and said, hey, come say hi to me next time. By the way, for those of you asking about the device that was in my hand, it's a Mugi gel ink pen. Of course, tongue-in-cheek humor here. doesn't address the fact that he was wearing this novel new pair of headphones and this puck-shaped

device on the table, which tells me that I think he's under NDA. They're all in on it. Like, the video was not an accident. That was, that was a marketing stunt. By who I don't know, it feels very Johnny I've coded. It has like the stainless steel materials. It's very elegant. It's futuristic. It is, it feels like something that he would produce and that the world would want to use. So we'll see. Josh, let me ask you this. What do you think the odds are that, you know, Open AI announces one of these device products. I'm glad you asked. Thankfully, we have Polymarket to determine what the viability of this is.

So the market that we're looking at right now is what kind of product OpenAI will announce in 2026. And we actually have the probabilities of the type of device that they will announce. And we've heard rumors of all of it. We've heard speakers. We've heard glasses. We've heard hearing devices. We've heard pucks. And what it looks like based on this is there is a 50% chance, which is higher than all of the other markets, that it will be earbuds or headphones.

Which is really exciting because those over the ear, earbut things that they've been using, they look really slick. And I mean, both of us, we have AirPods in. We use AirPods all the time. I would love an AI first alternative. That would be so cool. Yeah. It's also up 11% since Joe Gabio's video went viral. Interesting.

I wonder if there's some insider trading going on there. I think there is. But there was also this second polly market, right? Yes, which is when is this thing actually going to come out if it does? And unfortunately, it appears as if the answer is not any time. soon. It's showing for March 31st of this year, 2%. December 31st of this year, 34%. So the odds are not really in our favor that the device comes out soon, which is a bummer, but I suspect it's coming. And when it does, I really hope it's what we're seeing on screen in these teasers. And again, I mean, this might not be

coming soon, but there is something that is coming soon. And it's kind of the anti-AI device. And it's this thing called, what is it called? Spectrue 1. And it came out of no. I saw this video going viral on X and I was like, wait, people have been building this stuff. It's pretty cool. It's super refreshing to see. So basically what this gadget or device does is it can sit in your pocket or it can sit on your desk. And it creates AI generated sound signals or audio signals that block any nearby microphones that are listening. Now, if you're wondering who the hell are carrying microphones on them, what's this little thing that everyone hones?

It is your mobile phone, it is your Amazon Alexa, it is your Google Home. They all have mics that are actively listening to you, and that information gets fed back to these companies, which can be used for targeted ads. It's the reason why when you talk about something, you suddenly get a specific ad on Instagram, surprise, surprise, that is advertising that exact thing for you. So what this device does is it is pro-privacy. It fights back against the machine and blocks out all those signals so you can finally have a conversation in peace and privacy.

Do you think Instagram's listening? Yeah, absolutely. Wow, that's a crazy thing. You don't get served ads. I get served ads, but I'm not sure they're listening. I think they just have a really complex profile on who we are. Besides the point, we should talk about why this is interesting, right? And it's like, okay, you picked up the phone.

The phone is like clearly a very noteworthy thing to kind of defend yourself against. But all of these new devices that we're seeing, the pins, the puck's, the earbuds, even the glasses, those are going to be constantly observing. So this feels like the first defense tech consumer AI hardware device. And it's super fascinating juxtapose against this other story that we have, which is what you're exactly defending against. And it's meta workers saying that they're seeing disturbing things through user smart glasses. Now, my question when I read this headline is,

how on earth do they have access to that data? Okay. So there's a lot of layers to this story. So let me try and walk you through this. So Meta released a slew of different AI-powered glasses. Their most famous one being the Rayban displays last year. And the product kind of flopped initially, and now it's like super popular. They're actually shipping 30 to 40 million more units this year alone.

So there's a bunch of people out there, millions of people, that are wearing these glasses. The issue is, when you switch on these glasses, you sign a terms. of service agreement with Mehta saying, yeah, whenever I have the camera on, all that information that it records, it's now Mehta's property. And they can own it, they can use that video feed to do whatever they want.

They can use the audio signals to do whatever they want. The problem is some people forget to turn their camera off. And that's where the issue begins. So what Meta does is they've ingested all this information, this video. They send this video footage to outsourced workers in Kenya under a company called S-A-M-A, SAMA, and what these reporters, or rather what these workers reported, is their explicit footage that is on all of these things. They see people doing extremely private things, looking at bank account details, which they shouldn't be able to see.

And so it brings up the point around, should we be signing these consent forms, should companies like META have access to this type of information? One final Easter egg on this, Josh. guess who got into or who was also associated with a major conflict with this company, Samma, around privacy of data two and a half years ago. Don't tell me it's Samma, Sam Altman. It is. Is it actually Samma's engaged with Samma? I'm not making this up.

2021, go check it out. Yikes. Okay. Well, clearly this is a trend that is a little bit scary. And I guess it's a testament to what the expectations of these companies are, right? It's like, meta needs to train these models. It needs the data. It has no choice but to take it wherever it can get it.

And unfortunately, that's you. And that is the people who are using these classes to help the training data set. I'm surprised there's no opt out. Maybe there is an opt out. I guess we'll look at that and report back. But this comes on the back of some other big news. We had a couple of big exodus this week, just like total destruction of companies starting with the Quinn team.

And it's a shame because we're really big fans of Quinn here. Quinn, for those who aren't familiar, really popular open source Chinese model. It's doing amazing work. Just this week, they released a model small enough and powerful enough to run on an iPhone 17, which is really cool. But unfortunately, it seems like they kind of fell apart. Yeah, it was the best and worst week for this Chinese AI lab, who was founded under the Alibaba company. So they're kind of owned by Alibaba. Three of their key leaders were let go this week because there was a restructuring where the Alibaba CEO said,

I kind of don't want you guys to open source the models anymore. I want to take this under my wing, and I want to plug in an ex-Google Gemini guy instead. So yeah, three of their key leaders have left. It's a massive loss because it was their best week. They dropped four models, as you just mentioned, all which fit on either your phone or laptop, which was a major win for open source and local AI.

Coming after a week of Apple releasing their five new devices, where you can also run models on their devices, I was really excited about this. But unfortunately, yeah, we are moving towards a more close source world. Yeah, that sets a bad precedent. And then for, I guess, a different reason, there was another exodus from the Block team, which might be setting a different type of precedent as opposed to going close source. They're actually just leaning down the company.

Block, which is a company run by Jack Dorsey, who is famously the founder of Twitter, also known as NowX.com. He founded a second company named Block, and it is the company responsible for popular apps like Cash App that I'm sure a lot of people have heard of. The company had 10,000 employees as of last week. As this week, the company now has 6,000 employees. 40%, 4,000 people got laid off. And Jack, he released a pretty heartfelt message saying, I'm sorry. Like, if you're affected, they're giving you 21 weeks of severance pay, six months of health care, $5,000 towards transitioning. There's a lot of really generous severance, but the, like, ultimatum kind of

came down to the fact that a lot of these people aren't required for the job. And I'm sure some of this is downstream of overhiring that happened during COVID. But a lot of this is from, yeah, this part that you're highlighting now, you just just the intelligence tools. And what happened they did they laid off this company or they laid off all these employees? The stock was up 20%. The stock market loved it. So it sets a precedent for, I'm sure, what we're going to see to come. I mean, stock markets or investors love company cuts. and it's because companies are leaner and in this new AI world, not only can they make the cuts,

but they can become more productive, which means more money to investors in returns. So I can see why he made this decision. I'm a little skeptical as to whether his claims of AI intelligence tools being good enough to replace a lot of employees' work is actually true. I saw something going around in response to this news, which was Jack Dorsey had just burned a bunch of money unnecessarily over the last couple of years, and now he had to be forced to make cuts.

But in other news, in more fun news, I came across the wildest post yesterday, Josh. If we read this, it goes, an invidia-powered farming machine uses AI vision and precision lasers to eliminate weeds in milliseconds without herbicides. Now, that all sounds like farmer talk until I dug in a bit more, and this thing is super awesome. So what you're seeing on screen here is a combine harvester equipped with none other than Jensen Huang's Nvidia GPUs. And it powers a camera which stares at the ground, finding all the weeds that aren't your crops that are affecting and sapping away from your crops. And it zaps them at 600,000 weeds an hour. That's 5,000 weeds per second, for those of you who are super curious.

And it reduces the cost of spending on herbicides by 90%, which means after a year of using this $1.2 million machine, you save a bunch of money. How crazy is up. This is ironically, like, my favorite use cases for AI. because again, we've had so little innovation in the physical world in a meaningful way. And what a cool use case is like, okay, yeah, bring it to the farmers, zap all those weeds, don't use the pesticides, don't harm any of the plants, just knock that thing out. And the fact that, like, to think that these things could be running like GB200 server racks inside of them is pretty badass. I know it's probably not the case, but it's a really fun and exciting use case of AI that impacts the food chain.

This is a big deal. For those of you who aren't watching this, by the way, we're not exaggerating. We mean zap literally. There is smoke and flames coming up off the ground where these lasers are firing. It's absolutely insane. So I went down to rabbit hole just to kind of round this story up. This isn't the only AI machine powered by NVIDIA's GPUs. There is an AI cow milking machine.

There's also a scarecrow which shoots lasers at birds that land on the crop fields, which eliminates 80% of bird interactions, which in defrauding. fears with crops and crop growth. The result of all of these machines being used is a 30x increase in crop production if you use AI. There you go. There's another use case for GPUs. AI affects everything, not just software. Bullish on farming, bullish on AI, bullish on Vivida and not bullish on weeds. Rough day, rough week, rough month to be weed. Apologies in advance. Sending condolences, but that wraps up the weekly roundup. If you've made it this far, if you've listened to the other three episodes we've published this week. You are now fully caught up. You can go touch grass.

You can go outside, hang out with your friends, or share with your friends all the things you've learned because there is nothing. There's no stone unturned that we have left this week in covering the world of AI. So thank you so much for joining us again for another great week. In addition to sharing with your friends, rating of five stars helps a lot on either like wherever you listen to podcasts, like Apple Podcasts, Spotify, leaving comments. We love those. EJA, any final notes before we head off for another weekend? Yeah. So Josh and I in The Limitless Brand has, has had probably its best ever week on X. So if you're on X,

come and check us out. Our profiles are linked in the description, as well as some banger tweets from both of us. Definitely go check it out. We also post all our content, so full episodes as well as newsletters, which we dropped on Wednesday and a new one's coming out today on there.

So X is really where we converge a lot of things, so definitely go check that out. But as Josh said, yeah, please give us a rating, YouTube comments, we're reading all of them, we're responding to all of them. And finally, we transition to four episodes a week. So even more high-quality content for all of you who have been wanting it. Been an exciting week, and I'm excited to see you guys on the next one.

We'll see you guys next time.