Novartis Buys Excellergy for $2B, Anthropic Vs. Pentagon, The Mansion Section | Diet TBPN

0:00John Coogan:So why is no one talking about Accelergy? I love these titles. Very niche company, was in stealth mode for five years, but sold this week to Novartis for two billion smackaroos. That's a good outcome. Novartis, of course, is the$300 billion Swiss pharmaceutical company, and they just paid$2 billion for basically a five-year-old company named Accelergy. sellergy. It's a great outcome. This is an interesting story because it's very much like the biotech playbook done right. They bring in the right CEO. Tyler, is$2 billion exit in five years from your point of view, is this good? Is this good? That's pretty good, yeah.

0:40I agree.

0:43John Coogan:But this isn't in the sweet spot for typical tech news, but there are some interesting data points. Amir over at The Information said, leading indicator of AI adoption, drug companies like Novo Nardisk, pharma has long been at the forefront of machine learning. Ozempic Maker says AI agents are shortening its clinical trials. People think that it's going to a prompt and saying like, cure cancer. It's not. It's like you have a 10 ,000 page document that you need to send to the FDA. And if there's a comma in the wrong place, they can just send it back to you and say, there's an air like we don't want to deal with it right now.

1:19John Coogan:You're back at the back of the queue. Is that real? You can get dinged for crazy, crazy stuff. Like I don't know exactly what I can share, but I've heard of multi billion dollar companies getting dinged for not showing like they will list out, OK, we use this centrifuge to run this assay to understand how this chemical separates. But they didn't say they didn't list out who the suppliers to that supplier are. They didn't explain the entire supply chain. And there was like very little guidance on it. Like little things like that come up all the time and reading through all the documentation. Of course, most of this, when you're talking about Novo, it's like they have a direct line of communication with the FDA.

1:57John Coogan:So it's different. When some type of public company software CEO comes out and says like, everything is accelerating because of AI. Oftentimes I look at the business and if, you know, if revenue isn't accelerating dramatically, you think like, okay, this guy just wants an AI narrative. He's just reaching for an AI narrative here. But when I see a pharma company do it, no real context on pharma personally. So I'm like, it must be true. Yeah. I mean, I was working on FDA filings like four or five years ago, and there were so many Word documents that needed to be part. It was the same application for multiple variations on the product that the FDA was going to review.

2:34John Coogan:And so we actually wrote Python scripts to programmatically interact with the Word documents, to basically do like a very advanced version of find and replace. It's like perfect for AI. And it just speeds things up a little bit. You submit a week earlier, you have, you know, you can submit more things, more variations, you can give more information, compile more information, investigate other literature, do deep research reports, what else is out there, let's review all the IP faster, summarize things, find different threads to pull on. Little things like that just squeak out like an extra five minutes here, an extra day there.

3:10John Coogan:And it all adds up to acceleration and progress. It's not like an overnight, all of a sudden, everything's completely changed. Maybe that happens. But at least right now, there's clearly, you know, a benefit from overnight success. But this was an overnight success, at least for Alex Egel, Ted Jardensky, Luke Pennington, and Jeff Harris, who took academic research, which they've been doing at labs at Stanford and burn into what is basically a VC backed biotech company, a drug company. So the goal was simple. What do they actually make? They wanted to make a better allergy drug. So when people have allergies, they, you know, you have a peanut allergy.

3:45John Coogan:If peanuts come in, your allergy cells like flare up and you need to fight that back. So the current, the current blockbuster drug in the category is called Zolaire. And when someone has an allergic reaction to a food, allergy drugs act quickly to treat the reaction before it does lasting damage to the body. So the key immune molecule at the center of many allergic reactions is called IgE. And Accelergy's lead program, EXL111, targets IgE. So the goal is to work faster, quiet the pathway more deeply, and ideally last longer between doses to fight allergic reactions. And so these will make that less severe.

4:21John Coogan:So Redtree Venture Capital seeded the company in 2021, but they stayed in stealth for about five years. The team announced their first big fundraise, a$70 million Series A, led by Samsara BioCapital. This is a pretty typical flow for biotech companies. University of Science first, then they bring on secondary specialist capital, someone with experience in biotech. It's less common to go to the big names that we know, the Sequoias, the Founders Funds, the Andreassens. They do some stuff in bio, But very often, if it's coming directly out of a lab, going to be IPO'd or acquired in a few years, that's a pretty traditional biotech VC playbook.

4:58John Coogan:And so by February of this year, Accelergy dosed its first phase one patients. So they gave the drug to subjects. And it was obviously looking good because the Novartis deal signed a few weeks ago. How well does that phase one trial need to go for somebody to immediately come in and be like, yep, here's$2 billion? Probably really well. Yeah. But it is like standing on the shoulders of giants. Remember, the research was done in 2020, 2021, probably earlier than that. And it's building on an existing pathway. There's some interesting stuff going on. So Novartis. This guy is such, CEO is such a chad.

5:33John Coogan:Yeah, insane. So yeah, I was like, I didn't want to diminish his accomplishments by reducing it to a playbook. But this isn't his first rodeo. So the CEO, Todd Zavodnik, he joined in 2025, sells the company in less than a year. And he's seen the story twice before. When he was at Zeltic, Allergan acquired it for$2.4 billion. And when he ran Dermavant, he sold that company for$1.2 billion in 2024. So he's seen a unicorn outcome two years ago, and he gets a new job. He's like, I was going to take some time off, gets back in the arena and immediately sells the company. So clearly a great operator.

6:11John Coogan:and no one was really surprised by this at least in the stock market novartis is flat join his next company yeah that's a good idea yeah just follow this guy around three billion dollar acquisitions be like back to back i would i think it's a good i would love to be a janitor at this company yeah i would love to be a janitor here i'll actually take all stock i'll take all stock and i will clean the toilets and get coffee if that's what it takes because i know what's going to happen. I've seen this playbook before. Anyway, Novartis, you know, the stock is not really moving on this news because this is very expected.

6:45John Coogan:This is part of their playbook. Just last year, 2025, they made$1.7 billion in sales of Zolaire, the allergy drug that's currently the blockbuster drug in the category. And this is a very logical next act in that category. So key Zolaire patents are about to expire. So a next generation IGE asset is strategically valuable right now. Novartis' CEO has previously outlined a pretty broad acquisition playbook. It's focused on bolt-on deals in core areas. So when they have a business up and running like immunology, like treating allergies, they have all of the sales and distribution. They have an existing product portfolio.

7:24John Coogan:The business is working. Add things on to extend the viability of that business line. And Accelergy fits squarely inside of that plan. Now, the$2 billion acquisition price is not all cash up front. It is split between an upfront payment and there's milestones based on development and commercial targets. So if all of a sudden they flop in phase two trials, probably some of that gets clawed back. It needs to actually work and all of the promises that they may have to play out. But they have Novartis behind them to propel them through that. And it feels like a pretty standard playbook. So good stuff there.

7:58John Coogan:And in general, it's just like it's cool. Anyone who's dealt with allergies, either personally or with loved ones, it's always a hassle. uh you know everyone has a story of like someone was on vacation they got a drink that had a nut in it and someone you know had an allergic reaction and anything that can treat that is obviously good so another arrow in the quiver of modern medicine which is great to see breaking news huge leak omg did you guys know that anthropic was training a better model this whole time that's this is shocking according to unemployed capital allocator yeah Of course, referencing a leak, a scoop.

8:38John Coogan:I wonder if this will trigger the team at DeepMind or OpenAI to consider training a better model. It feels like if they're going to have to respond. So you could see multiple companies training new models based on this leak. Yeah. Yeah. An interesting story. It's more from a marketing perspective and how you actually roll out and manage vibes and hype. Yeah, to me, it doesn't seem improbable that there's been a lot of spud talk on the timeline this week. You've got to start pumping Mythos. Mythos talk. The most powerful AI model ever developed. It's a good tagline. And very, very painful for all the SaaS companies out there that just sell off whenever Anthropic does something.

9:23And it's a vicious cycle now.

9:25John Coogan:Prinz wrote a little summary of this. Anthropic has been testing a new model called Mythos with certain customers. They call it a step change in AI capabilities, including dramatically higher scores in coding, academic reasoning and cybersecurity. I want to know how it will do an ArcGIS V3 because all of the models are under one percent completion on that benchmark. It's the one benchmark that is not saturated. It's not it's very closely guarded. You can't get into the tests, the actual test set. You can only see the examples. and it clearly is, I still love the benchmark and the team over at ARK for what they're doing.

10:01John Coogan:So this is part of a new Capybara series of models which are larger and more intelligent than Opus. It's more expensive to run and it's not ready yet for general release but they are excited about it. Mythos pricing bets coming in from 0.005 seconds. $100 in,$250 out. subs repriced to$50,$250, and a$1 ,000 plan might be coming. Who knows? What do you think? Well, let's pull up this picture of a capybara just to put this into context. Is it a mythical creature? No, a capybara is real. So what's the connection between capybara and mythos? So they're normally quite docile, but they can get violent.

10:44Let's look at this picture. Okay, watch out. I mean, look at the chomp on that.

10:51John Coogan:This is something we should read into. We should say, what do they mean by this? You know, it's clearly some sort of vague post. They're telling you something that, you know, friendly, cute, but also potentially the end of the world. Do you think we could contain it? Do you think if we had one in the studio, it would overrun us? Or do you think we could keep it locked up? I think. That's pretty horrifying. I think four or five of these could take you down. Yeah, median weight is between 60 and 174 pounds. 174 pounds. That's bigger than my dog. Wow. That's a big capybara. That's what I'm saying.

11:27They have these at the Santa Barbara Zoo.

11:28John Coogan:So this is the metaphor. You know, like these models are getting incredibly powerful. You need to contain them and think about safety. And you need to think about how do you keep the capybara safe? You need to lock it in a box, in a cage. It cannot be allowed to run free. Yeah, I think just back on the pricing, I think like this is generally right. So basically 0.005 seconds is saying like it's just gonna get 10x more expensive because right now I think it's something like$10 in$25 out people are ready for people are ready to pay more though Like they're paying a bunch through the API The subscription like technically it's the same price, but the limits are going down So it's okay.

12:06The tokens are getting more expensive. Yeah, and as you see like there's like this insane compute crunch Like I assume that models are going to get much more expensive as they're getting bigger.

12:14John Coogan:Yeah, it is interesting. I wasn't expecting to be in the regime of like subscriptions as long as we have been. Like just talking to Ben on our team about how he sort of bounces around from one subscription to the next. Like he'll max one out and then move over to the next one and then move over to the different one. And he's like maybe more price sensitive than model sensitive. I would always assume, I'd always assume that most businesses would just be on an API consumption basis. So the reason, like, I'm not hitting the APIs is because it is, like, basically 10x more expensive than using the subscription.

12:51John Coogan:The subscription. So you do most of the work in the subscription, which is just, I mean, that's economically rational. I guess what I'm saying is, like, I'm surprised by how many people who are in a business context still make that decision and how much of a price war we're in, how much capital matters here. Well, let's pull up this video because I'm not sure you guys would be making the same decisions with the Capybara family of models if you knew that Capybaras were capable of this kind of thing. They can swim? Wait, this is a human. Is this AI? Oh my god, that's really attacking her. Wow. This is why they're calling it their most powerful model ever.

13:37John Coogan:We need to slow down. We need to slow down. All right, we can turn it off. It's horrifying. It's horrifying. Yeah, we need to slow down. I hope they keep this with the safety team for long. I think we need all countries to come together and agree to control the capybara population because that is truly horrifying. Tay Kim has a very independent Tay. He's the biggest NVIDIA stan, and he makes a lot of good points. But he's making the argument that Anthropic needs more GPUs because he's sharing some posts here. To manage growing demand for Claude, we're adjusting our five-hour session limits for free Pro Max subs during peak hours.

14:16John Coogan:Your weekly limits remain unchanged. And Take Him is coming on the show Monday. I'm very excited to talk to him. We'll ask him what he thinks about NVIDIA. Yes. And so a lot of people are getting into polyphasic sleep because of this. So Claude has different limits at different times. So when the limits are in effect, that's when you need to be sleeping. Tyler, have you gotten into polyphasic sleep yet? I'm not there yet, but I mean, the problem is getting worse. I told Tyler that sometimes I indulge in polyphasic sleep after the show. oh, I'll sleep for an hour and then I'll go about my day and then I'll sleep at night.

14:56John Coogan:And he said, that's just a nap. It doesn't count. I guess people really are doing the polyphasic sleep thing to monitor their agents. You would think you would know one person that has done that and stuck with it. Because I feel like a bunch of people have tried that over time. Yeah. And then eventually. I mean, if you're trying to get around the limits, polyphasic sleep doesn't make sense. You just need to basically be nocturnal, right? because you need to be like inverse what like San Francisco hours are, right? Yeah. So the peak San Francisco usage, you're going to be like - Yeah, or just move overseas.

15:26John Coogan:You could do that. Yeah, but the polyphasic sleep doesn't really help, right? No, no. No, it seems like it's much better to just, again, you know, have a compute desk or a prompt desk that is geographically distributed evenly across the globe. So when your American employee stops the prompting, goes home for the night, then your European colleague steps up and starts prompting on the same project. And then your colleague in Asia can pick up when Europe goes to bed, something like that. You know what I mean? So you have 24-7 coverage. Yeah, but that's still like during one of those periods, it's still like the peak hours, right?

16:08John Coogan:Yeah, because these peak hours are based around U.S. hours. I was more. So there's actually two things going on. There's one where it's like you're waiting for your agents to respond. And if it's going to take hours for a prompt to return, you should go take a nap because you're better off checking in on a prompt every two hours for the whole day for 24 hours. and in perpetuity, as opposed to doing a two-hour prompt, watching TV, come back, do that eight times, and then sleep for eight hours. That's four more two-hour sessions that you could have been letting it cook and interacting with it. That's separate from trying to get around these rate limits.

16:55John Coogan:Last March, like a year ago, people were like,$50 billion,$100 billion, dollars,$200 billion. Is this stuff that useful? Is it slop? Is it actually going to be that useful? Are these things toys? Like, yeah, they're reasonable for this thing, but what's the real economic value? Certainly it can't be more than like 20 bucks a month or$200 a month. And now there's like lots of people that are, maybe they're only paying a few thousand a month because they're using a bunch of subscriptions, but like they're pretty close to justifying like the Jensen like 250K a year per engineer thing. Like that doesn't feel that far away from companies just saying like, yes, I give you the permission to spend$250 ,000 a year on AI inference.

17:38John Coogan:Be as efficient as possible as you can with that budget. But you have my approval. I believe that with that 250, you will generate more than that in value. Yeah, I think the big thing is like there's going to be more use cases that have a similar level of product market fit as CodeGen that use tokens on the level that these CodeGen tools do. And we haven't fully identified those yet. There's some early evidence that general agents will drive that. But like I would say, an example is like it seems very obvious that models will be very useful for making financial models. But do you need to make 500 financial models per day?

18:27No, right? Yeah, that is interesting. And so there are areas where the models are going to be incredibly, incredibly valuable, incredibly important, used by everybody in any given field. And yet they won't necessarily generate.

18:40John Coogan:There are some people that would need to make 500 financial models a day. I think Ro Khanna makes hundreds of trades every single day. And so if you're doing a proper full DCF, pulling every SEC filing for a company, and you're doing that hundreds of times a day because you're trading. Or he might have hundreds of individual positions. Exactly. And you want real-time models. Sort of like rebuilding a Bloomberg terminal for yourself. I think Bloomberg exists and maybe should work there. The news is that Anthropic has been granted a preliminary injunction re the Pentagon supply chain risk designation in California, but is allowing stay for one week.

19:19John Coogan:I think a lot of people, like the tech community broadly, came away with this prediction that this wouldn't hold, that this was, you know, that whatever you thought about how you work with the government and the Ben Thompson thesis, there was, you know, the supply chain thing wasn't really going to stick. It wouldn't necessarily hold up in court, but we'll see where it goes. Dean Ball says, this is a devastating ruling for the government, finding Anthropik likely to prevail on essentially all of its theories for why the government's actions were unlawful and unconstitutional. One of the things she mentions is the huge range of amici briefs.

19:55John Coogan:I don't know how to say amici. Briefs supporting Anthropic, by the way, zero. Hey, finally, a word that John can't nail instantly. You have an incredible mind that we did find. We found my limit. We found my limit. On a personal note, says Dean Ball, some friends and allies of mine on the right who have been angry at me for my own words and actions and all of this. Anyone who thinks I spoke out for personal gain or trivial reasons against an administration I served in is crazy. Yet people forget that Dean Ball served not in the first Trump administration, but the second Trump administration. And he was like, look, I wrote the policy thing.

20:32John Coogan:I worked on that. Like, this is not the right path. And so it's always it's a me and it's the plural of amicus. It's this is brutal for me because I studied Latin and I should be able to nail this because I know it's a Latin word. Anyway, it was a hugely costly decision for Dean Ball, but this judge's ruling shows why I did it. This is a staggeringly illegal act by the government. That is why I'm particularly honored to have been implicitly quoted in the ruling for calling this what it was when Secretary Hegseth made his initial announcement an attempted act of corporate murder. He's standing with the corporations.

21:08John Coogan:This case continues, but Anthropik has scored a very large win here. The real victors, however, are all red-blooded Americans who are, as the founders would have said, jealous of their liberties. I love it. Well, congratulations. And hopefully everyone can just go back to building good models. You know, all this nonsense in D.C. has been a huge, huge side show, side project, side quest, as the race for compute is so much more important. The race for capabilities is so much more important. The other scoop is coming from Axios. Sam Altman apparently told staff he tried to save Anthropic in the Pentagon clash.

21:48Which he apparently thought was somewhat ironic, given that said competitor openly blasted him on any podcast that they go on.