SpaceX Starfall And Apple Price Hikes | The Brainstorm 138

Speaker A: This week we are talking SpaceX Starfall, maybe the last star named product from SpaceX. Memory prices going up across the board and AI competition from open weight models. Brett, are Starfall items going to come down and land on our heads? What is this product? Why is it so interesting?

Speaker B: Yeah, so Starfall is basically ah, a standardized ah, form factor for delivering um, goods from orbit. Uh, you know the cost of getting something back from orbit is much, much higher than the cost of getting something into orbit. At least in one piece you can burn it up on the way down. And so the Starfall containers are, call it pucks, but they are you know, on the order of 10ft across and a couple feet deep. Uh, and they fit nicely into SpaceX's Starship. And the idea here is you could have a manufactured, an automated manufacturing process that needs microgravity in order to work sitting inside one of these pucks and then you can the products back. And so people have speculated and some have demonstrated that microgravity could be good for growing crystals and doing pharmaceutical stuff and maybe some um, high end electronics, um, manufacturing. And it also means that you could um, you know, lob something around the Earth in half an hour to an hour and drop it onto a mountaintop. And that something could be uh, you know, basically micro shipping container worth of drones in a Starlink satellite or something. Department of Defense useful. Uh, and so this is a, you know, new to the world capability to deploy physical assets into a location on very short notice anywhere on Earth.

Speaker A: Yes, that is a scary thought. Orbitally injected drone swarms coming to mountaintops near you.

Speaker B: Yeah, I mean and it does seem, and this is um, you know, you know, a company, ah, we partnered with Mach 33s work but they, they you know, dimensioned that um, the cost to return things from orbit was 50 times the cost to send things up to orbit. So if you have to spend a thousand times to send things up, you spend $50,000, $50,000 per kilogram to get them back down. And this could reduce that cost premium from 50x to you know, single digit or sorry, low double digit percentages. Um, so if the starship can launch goods less than $100 per kilogram, we think it'll get down to you know, 50 to 70, then uh, you can maybe get things back for $100 per kilogram. So this is the, you know, a real performance improvement curve in an area where the unit economics and what people can find to use this for haven't been fully explored. Mostly we've been returning humans from Space. This isn't um, set up to do that, but you can imagine, yeah, getting anything like Amazon prime on steroids anywhere in the world could be useful for many military applications, but also, you know, human aid applications and um, you know, other areas where it's like I need this thing in this spot at this moment, um, or I need this thing that's in space that's, you know, doing some kind of manufacturing process back to my facility so it can be unprocessed. Um, you know, ah, that could prove useful in probably unexpected ways.

Speaker A: Well, that's what I wanted to ask you about. Is this the cost decline that, you know, we always talk about cost declines, tipping over, you know, these break even or unlock points that unlock exponential demand. And I think, you know, in space manufacturing has been one of those items where people have always hand waved. It was kind of like Earth observation, I would say, right? It's like, no, no, there's a huge market beyond government and defense, but it's actually like, oh, defense just became a bigger market for that. Like no one else is out there necessarily buying this. And for in space manufacturing it's always been pharmaceuticals. Oh, you could spin fiber incredibly well. No, like uh, eyes and things like that that need microgravity. Uh, is, is or are those industries about to materialize and have they only been, uh, held up because of the offside unit economics or do you think there's a lot to be done here and figuring out what the real application is?

Speaker B: I think it's really uncertain because this is a dramatic change in the cost structure. And so naturally you would bring things on side and the military application seems very obvious and very obviously useful. Whereas, uh, you know, how much microgravity manufacturing could we possibly need as inputs to other industries? And they have to be inputs that you are tolerant of there being a lot of latency because, you know, you're sending it up and you're getting it back. It's almost like, you know, contract manufacturing, you know, to someone external to you. You have to design an automated manufacturing system that fits in this thing. Like there's a lot of steps to even discover whether or not there's a market there. Um, and as Tasha pointed out, you know, the, the kind of size and the stealthiness with which they've approached this does suggest that maybe the form factor was requested from um, you know, the US Government in particular. And um, I, I do think given the way drone warfare is changing, um, this is extremely obviously useful in that way. Whereas in the other ways it's like yes, there's potential.

Speaker A: It's pretty clear. Aircraft carriers are great for fighter jets, but launching a drone from a aircraft carrier probably doesn't get it where it needs to be. And so this is a novel.

Speaker B: Well, it also, it even changes like the way your drone could be. You know, if, if you imagine some of these like long um, kind of flight plan, drones that are expensive and need to navigate by themselves, um, that you can, if you can get something much closer you can do, you can handle a different form factor, much um, smaller, you know, and uh, more attritable. And so it kind of can change the entire um, kind of basis by which you project force in an interesting way. And I think that you know, the Ukrainian war has demonstrated that there's been a change in the technology of warfare and that doesn't happen that frequently. Um, um, but when it does it really changes like optimum strategy for war. Like think about the machine gun in World War I or tanks in World War II, um, where you've designed to a different domain than what you face and it totally transforms who could project power and in what ways they can do so and both the cost and the benefits of war. Um, we're clearly in a different domain here. And I think it's also pretty clear that the largest military on earth has not yet adapted to it. This is a super interesting tool for military that not anyone else yet will have access to, though one would imagine the Chinese are going to quickly develop something similar. Um, and it really um, transforms how you think about distance and ability to project power over distance quickly. Um, so I think it'll require an updating of all kind of like the strategy and even a ah, re updating of what drone warfare could look like on a uh, more regional or global scale. Um, as compared to the Ukrainian conflict where it's still kind of very much at the front lines. So we'll see.

Speaker A: All right, so speaking of projecting power, Apple forcing everyone to pay 30% more for their, for their products because memory

Speaker C: got not 30, not 30, 20, 20, 20%.

Speaker A: Um, Nick, you want to take this? Why, why is this happening?

Speaker B: Brett Wa and I, I disagree with the framing. It's not Apple forcing Everybody to pay 20% more. It's Apple being forced to force Everybody to pay 20% more.

Speaker C: It's actually our fault for using the models OpenAI and Claude so much such that the consumer skyrocketed dram price. It's actually, it's our fault. We should, we should all take blame everyone if we stop using the AI tools So much than our, you know,

Speaker A: it's Amazon's, it's Amazon's token leaderboard that caused Apple to raise prices on us there.

Speaker C: They are interconnected if you. And so to get to the real news here, Apple is raising prices on average around 20% across a number of different, of, uh, one of different SKUs of their product. And the reason is because memory prices have continued to rise. And you know, I think they're getting to a point where they can't continue to, to subsidize that cost to the consumer. And so they have, as of last week decided and announced that they're going to raise pricing as Apple did that. You had a number of other companies come out and say we're going to follow suit. I think everyone was probably waiting for one of the big players in the space on the hardware side to do this because there would be political pushback. I think someone has already come out saying we should break up Apple, uh, because they've done this to the consumer. I think it's aoc, I'm pretty sure. Um, so, you know, that is, uh.

Speaker B: How dare you try to remain profitable.

Speaker C: Yeah, yeah, yeah, exactly, exactly. Uh, so that is, yeah, it's happening and it is actually tied to the AI boom and everything happening on the demand side because DRAM prices have continued to run up. You can look at, you know, a number of different players in the space, um, that are, you know, very much supply constrained. So you know, the demand out there for DRAM has exploded, um, and the capacity to build DRAM has, you know, it takes a long time to, to, to create these, um, the capability to create more dram, um, and so as a result, supply and demand boom pricing is up, I think, you know, 60 to 90% depending on where you look. Um, I was watching the Gavin Baker podcast over the weekend. I think he was saying the floor pricing, um, because you have a floor and a ceiling pricing, when you lock in new contracts, just the floor pricing is up around 60% or you know, somewhere near that price. Um, so, you know, it's a tough time to be in the hardware space, uh, because a lot of consumer electronics have dram, almost all. Um, and so you know, these companies have had to eat that cost.

Speaker B: But this is, this is what's a few things that are interesting about this to me. One, if you go to Apple's like Mac Mini and Mac Studio, people are buying those explicitly to use them for running local AI. Uh, and they're, I think they already raised prices some. But also it's like they're back ordered out to like 10 weeks or something. But this is for a category of products that is not being used for AI or certainly not optimized for it. Uh, and uh, it's capitalism doing its work. People are soaking up, um, the globally scarce memory chips and then wildly underutilizing them on consumer devices. Therefore you have to pay a higher toll because those memory chips can deliver more productive use in keeping the context, keeping track of the context for my AI model that needs to be doing stuff for me. And so you have to pay for that, um, basically your underutilization of your consumer devices, um, a fairer price for this globally scarce resource that you're using. And it's kind of like remember when the um, the Iranian war, um, began and people were pointing to all the helium coming out of Qatar and saying uh, oh, this is going to pose a problem for semis, um, because you need helium and semi production. But if you look like the helium goes to more things than kind of like semi chips, it also goes to party balloons. And it's pretty clear that kind of like you're going to raise the price on party balloons before stopping producing computer chips. And if you want to have like party balloons you can pay more for, but you could just reduce the demand in party balloons in order to keep producing these computer chips. Well, you could like dampen demand for the MacBook that somebody, you know, doesn't even need, uh, or that you know, they are electively upgrading, uh, in order to reserve that um, kind of memory for better use. And that's what's happening here. But it might be bad for Apple's consumer business and certainly Sonos business and all of these people that are consumer facing with you know, um, kind of these inputs that aren't being fully utilized in their channels.

Speaker C: Well, and it's also, I mean if you think about where the new demand is coming from, it's coming from the hyperscalers. You know, they're, they're creating these massive data centers. Huge component of it is on the dram, high bandwidth memory side that's driving up pricing and they're paying inflated pricing. Assuming that, you know, this doesn't last forever and pricing at some point comes down because demand tapers off. I think that's one potential avenue. The other one is demand does not taper off and pricing continues to rise until you're able to bring on more supply. But my understanding, Brett, maybe you know better, but it takes years to put together the ability to fabricate new memory like it's not like you can just spin up uh, uh, a manufacturing plant for new memory.

Speaker B: Yeah, yeah, yeah. So this is in response to essentially call it two years back. You know, the memory companies underinvested in memory, not anticipating the demand spike that they would have. Um, inevitably in these kind of inputs, uh, they will respond to this price signal by investing in capacity. And while this time seems different, demand evaporates. Yeah, yeah, and this time seems different but like inevitably at some point they will overinvest and so in the cost of, you know, or at least their margins will collapse though, albeit it could be at a different pricing level depending on how long the um, the, the cycle continues. Uh, and uh, you know, these will not be forever up into the Right, it's not like memory has suddenly become secular business, whereas before it was cyclical. It's just this is a wildly large cycle. Uh, and so let's, let's talk on,

Speaker A: let's talk about that front and whether this time is different demand. We made the joke about Amazon and the leaderboard. Meta's probably taking their leaderboard down. Nick, I think you said that, um,

Speaker C: they've all taken them down. Shopify too.

Speaker A: Coin, Coinbase just posted a chart showing token use declining because they're moving to more efficient models. Right?

Speaker B: No, no, no, no. It showed token use continuing up but their spend declining as they're getting more efficient with spend. Um, yeah, uh, right, um, and that, you know, I'll paint kind of like the night nightmare scenario for the frontier model providers, right? Which is the OpenAI's and Anthropics of the world. They've injected uh, tens of billions of dollars into R and D, basically training models up to painfully um, educate kind of their users. This is how you use these models in productive ways. Users begin using them and say, hey, these are very useful, um, but actually the bill is a little higher than I was anticipating. And meanwhile on 6 to 12 month lag to them are um, mostly Chinese companies who are also um, kind of uh, basically at scale using anthropic and OpenAI's models not for useful direct work, but to figure out how they work and basically distill out all of their great answers into open weight models that um, no longer have to carry the margin burden to pay back that R and D spend, uh, and instead can be priced just at cost over infrastructure as a service to run them. Uh, and so the likes of Coinbase say, oh well I can use kind of like this model, it was cleanly built by an American company or this model that was partially stolen from an American company, but it's much lower price and I can get, not frontier class performance, but for a lot of my workloads enough. And so therefore, I'm going to aggressively shift spend to these open weight models, uh, and save myself a chunk of change while getting the same output performance. And so then the frontier class models have just basically invested all their R and D dollars on behalf of the world without an ability to recoup. That is the. The. I'm sure that's the narrative that will lack. People will latch onto here. Uh, and it's definitely, um, you know, at some set of companies. And the. The coinbase chart is indicative. It's what's happening. You use these to, like, for the. Only the. The very smartest things you need to do. Um, and you can turn.

Speaker A: I just want to. Yes, I want to clarify here, because the. The stacked bar here is spend. So token use is up, but spend is down, Right?

Speaker B: Yeah, that's what I said. Yes, exactly.

Speaker A: You said the opposite.

Speaker B: Oh, no, no, no, no. I mean, I said the opposite.

Speaker A: I thought Brett said. Okay, yes.

Speaker C: Okay, we're on.

Speaker B: On the same page.

Speaker A: On the same page.

Speaker D: Yeah.

Speaker B: Ah, yeah, yeah. Um, so. So one. Do you believe the narrative, Nick? Do you believe that that's the state of play and that's where we're headed? Where kind of the open AIs and anthropics of the world are in huge hairy trouble because open weight catches up and that's it, Game over.

Speaker C: See, I don't know that it's.

Speaker A: No, Nick. Say yes or no.

Speaker C: No. No, I. I think yes. If you were to say, like, yes, is this happening? Like, I don't. You know, it's. It's not very often you see CEOs come out and say, hey, we're completely reversing course on the thing we were just doing two months ago, right? Like, every company in the world was saying, we want our, you know, employees spending X amount. And then the next company would come out and say, we want it to be two times their amount. Right? Like that was happening just two months ago. And then I think everyone saw the bills and how crazy it got and said, hold on, what productivity lift are we getting? And how measurable is that? So I think 1. What needs to happen is companies need to figure out how to measure this productivity lift. Like, is it an. Is it employee, you know, sentiment? Like, our employees happier because they have to work less?

Speaker A: Probably.

Speaker C: Like, that's probably the main immediate benefit, right? Your work just got that much easier. And so all your employees are like so much happier because they can spend half their day watching Netflix. Yeah, but that's not, I have the

Speaker B: exact opposite problem by the way.

Speaker C: But yeah, I'm saying the, the, the majority of, of workers, okay, not, not the 1%. The second is you have to see meaningful lift in your revenue, you know, and then your bottom line. And that's going to take I think a bit longer as you have to build out new products. And so I think all these companies said we're going to slow down on spending, we're going to try to do this as efficiently as possible. We're still going to want to use AI in as many places throughout the organization as possible, but we need to do this in an efficient manner. Then you have, you know, the government coming in saying, hey, OpenAI anthropic, you can't continue to release your most performative models. We need to get a uh, clean look at this beforehand. And so the performance of the model and Brett, you and I have gone back and forth on this, right? Like that's what would take them, you know, continue to have spend increases if the performance is X amount better. Now you have new models released. And it to me, you know, going back two years ago, like Mythos or Fable for us normal folk, um, and then OpenAI just announced I think three new models, right? Like you're not seeing that same hype, you're not seeing that hockey stick in performance. So there is somewhat of a tapering off at least from a, from a, from a user perspective of like how performant I can be with these models. So now I think what we're entering in is the product stage. So you need to be able to wrap these products in a harness. You need to be efficient with that product. You need to have companies develop true use cases for these models, not just employees playing around and doing one off tasks. And so that I think is actually going to work against the frontier model companies because if you can swap out the back end or swap out the model for something that's more performant, cheaper costs, could potentially be open source, could be something fine tuned, something run locally, internally. That does not bode well for the model companies that want you to spend aggressively as possible on um, the most frontier capability because that's the most costly.

Speaker B: So I do think they're in, I don't think actually the model companies don't necessarily want you to spend as aggressively as possible on the frontier, um, but they do want you to spend gross profit maximizing to them which could be across a mix of models. I think the landscape is obviously very fluid. It's also. It's important to distinguish between you know, large enterprises that are running Palantir instances that then they kind of like stage models into and have um, basically like staged access for employees to models that they can switch back and forth. Uh, and kind of. And it's not just large enterprise, I should say large and technically sophisticated enterprises. Right. Versus kind um of like.

Speaker C: I think those organizations skip Palantir. Like the companies I've talked to when we've asked how are you using AI? The ones that I would qualify as in like saying oh you have like a technically savvy, amazing engineering team despite you not being labeled as a tech company, are all saying hey we're fine tuning, we're you know using this, we've built our own harness that we can swap out the models on the back end. And I don't think they're actually going to the palantirs of the world. I think like legacy companies that have extremely complex business.

Speaker B: But that's what I'm talking about. Okay. It could be like AWS is wrapped around it. I'm basically saying take companies that are actually kind of like empowering their employees by saying hey, you get access to this um, kind of API gated, um, almost like you know, custom tuned for our org set of um, AI models versus companies that um, are giving their employees access to cloud code or Codex as an endpoint and those employees can use it as they see fit. Those are very different sets of enterprises. And I think that the reason that Cowork took off and the reason that Codex works is it really is a um, um kind of like product led growth rather than selling through the CIO's office. And the employees appreciate it because it delivers them productivity immediately. And uh, like coinbases of the world are you know, incredibly technically sophisticated. In his post he talked about the way they continue token usage but while um, kind of like reining in spend was just changing the default that people went to. So if they went to an open weight model as the default but you still give them access to the frontier. It turns out for you know employees are using these things to be like you know, when is my dentist open? Next week? Uh rather than actual work. Yeah. And so um, kind of like giving people either the default or giving them I think ah, another stance that companies have take unlimited budget on a more efficient model whereas you have budget capped on the frontier so that people are being, you're using your employee to be like more thoughtful about when they're using what model for what um, does result in more efficient spend ultimately like in some ways anthropic needs something like this to happen because they are so compute constrained as uh, in they need people to be more selective so that their kind of like frontier class capability can be deployed against frontier class things only so that they can demonstrate how they shine. And then where I take kind of issue with both my own narrative that I laid out and your stance, Nick, is even at the frontier, at least within a harness, agentic models are wildly dumb and terrible relative to what we're going to see even six months or a year from now. And this is like a philosophical belief I have, I cannot give. I think you have to keep very close reigns on a model in order to get output that is good, that you can trust and that actually has any sort of taste at all over any time horizon. You give it a 15 minute task, it comes back with something that's like 80% done, that you kind of appreciate, but you have to do a lot of refining to get it close to something that is actually externally deliverable and that is a huge continual gap. So it's not like uh, they've basically productized this in a way where people can see the potential, but actually for the endpoint worker it is worthwhile to use, but not yet transformative from a productivity perspective, which I mean like a 10x, maybe it's a 2x, but it requires a lot of battling and figuring out kind of the gaps in order to get that 2 or 3x. And so um, long as we're in that space then um, one the kind of model builders can fill that gap with their R and D. Um, they also can spend R and D dollars not on delivering higher class models, but on taking the models they have and um, um, distilling them themselves to be more efficient. So OpenAI's newest model release, it's like they have something that's fable class at the frontier, but they're also delivering the current model for half the price. And they just released the current model. You know, I think it was three months ago, it's not that long ago. Uh, right. And so kind of the they are for not the coinbase of the world, but for the broader set of users who are like hey, we're going to give our employees codex that halves the cost per perform per performance for that entire set of employees in those organizations. Uh, and so then it puts less pressure on budget while getting same output without Having to be like, oh, now we're going to rip and replace and swap over to here. So I think that there's a, um, if I had, I think one, there will be this narrative of oh, actually we didn't need all this expensive AI, people are going to get as many tokens for less cost, which will then be narratively costly for OpenAI and anthropic. But I think it's wrong that actually we need to be burning a lot more tokens because we need to be doing a lot higher value added, uh, tasks and people will continue to use the Frontier and kind of like OpenAI and Anthropic will also recognize kind of this market pressure and try to make the routing better in their models where they're actually actively routing to their own internal models rather than forcing people to be like, oh, now I'm going to try to go to China for this stuff, um, so that people can get good performance per cost. Whereas if the world is putting up leaderboards like use as many tokens as possible. Anthropic has been like, we are happy to serve that need. You know, now that the world is putting up leaderboards of like, this is like a dollar per task rather than like, yeah, exactly. Then I mean particularly OpenAI, which has the R and D budget to spend, is going to be like, look, this is where we sit on the frontier and this is how we auto route your queries so it, so it like achieves the task at least cost to you. Uh, and they'll compete in that space as well.

Speaker C: I think.

Speaker D: Yes.

Speaker C: The nuance here is that it could be a short term, and it probably will be a short term headwind for these companies. But long term, if you believe in AGI and you know that we're going to continue to move towards a more agentic workforce, then this is going to be a blip in terms of, you know, what the end state of the market looks like. I think the question for us and for, for everyone looking at this is, is a six month lag or slowdown in token spend enough to throw the entire market into a tizzy? Like we're talking about dram prices, all of that is benchmark off of. Well, token spend is just going to continue to increase. Like all of these data center, um, like you know, agreements and leases that are, you know, spanning out five years is all based on the assumption that anthropic's ARR is going to go from 9 to 47 to 56 to 60 by the, and 100 by the end of the year. If that starts to taper off or looks like it's going to taper off, that could just be enough. Just given this market is moving at uh, like 150 miles an hour to throw everything in disarray for a short period of time. And you, you find out who was caught off sides and who overextended themselves. And you know, this, you know, one of the other things we've been talking about is this like round tripping of finances and deals behind the scenes. Like all of that could end up coming, you know, to fruition with just like a slowdown. Like I, I look to. And not like Look, I think OpenAI is an amazing company and there was, you know, rumors last week or I think maybe a spokesperson for them said or I think it was a rumor. I don't think it was a spokesperson but it was like they don't want to um, they're delaying their IPO to 2027 because they don't think that this is the right market environment to IPO at a $1 trillion valuation. I don't know what market environment you need to, to IPO right now, but I would think that this is pretty favorable given what we just saw with SpaceX. So to me that suggests like there is something happening behind the scenes.

Speaker D: Right?

Speaker C: Like we were getting ARR updates every two weeks from these companies. It's been two weeks, we haven't seen another one. Right. Like there's definitely a slowdown. You know, like if you were just

Speaker A: like moving at 150 miles an hour

Speaker C: and we slowed down to like 125.

Speaker A: Where's my.

Speaker C: But I, but I, but like you like you have to assume that like the market was thinking we were going to continue like maybe we were increasing to 175 miles an hour and we've just slowed down to like 120. And if we go down to 90 miles an hour to use this horrible analogy, like this really could be enough to throw like the market into a complete disarray in my opinion for a

Speaker A: short period of time.

Speaker B: And then I think, yeah, I think, I think, I think there's a question of like if you, from the model perspective, provider's perspective, right. OpenAI just raised what, 125 billion. Like why did they raise that much money? It's to secure future compute, right? And they don't sign those contracts on a like do you trust me this week basis. They signed the contract on a multi year. They're considered a good credit customer now because they have a Huge balance sheet basis and if yes if there is you know if they're on the verge of signing one of those deals and this kind of stuff happens which then causes them to have less negotiating leverage in that deal or not able to sign that deal then they have to hold back. The reason you have such a big balance sheet is because so you're not forced to sign a deal and you have to do it that week. It's so you can be strategic about you know where your leverage is um on a time basis to sign and secure that future compute. Um so I'm not sure that it directly impacts them in a material way. Um and narratives matter as in like I think that if it were yes a 1 year or 18 months of oh as it turns out we overspent on all that stuff we um then that would cause probably some strategic reshuffling in the immediate term. I would expect that Claude and Anthropic which has um, I think demonstrated the highest cost per task and actually is net slower through cowork than Codex will lose marginal growth share to OpenAI out of this and that people will take a hard look at XAI and cursor um as a result of this and the open weight models will continue to gain um you know chair for people who are on the API layer of like toggling switches between models and that's the I think the API business is always, it's always the business I've been less comfortable with from a um defensibility perspective because if you're designed into an API you can easily switch whereas the seat based or ah subscription business which is basically like a seat plus um per token usage um is much stickier because then you're getting kind of qualified in for a specific harness to your org, your employees are getting used to using that specific application uh um and you begin to build custom skills and maybe you let it have access to your calendar and email and so it becomes much more tied in with how the organization works and that uh, is less extricable. Um the other thing to think about here is like the Chinese companies have mostly worked by distilling the work that the frontier model providers have done by systematically asking them questions and you take the information out of the model into your model that you're building offshore uh Anthropic and OpenAI are um figuring out ways to combat and limit their access to to that uh and so it there's no like there's no natural reason why somebody has to pay the R D dollars for open Weight M models. Now that could be the Chinese government, which has plenty of money. Um, but uh, you know, they have to like figure out smart, motivated people to do it when kind of OpenAI and Anthropic are paying very high salaries to smart, motivated people. And so one way to do it is you basically like crib from your competitor and then you present it as if it's fresh. But if your competitor like begins to conceal their paper, um, then it actually might be that the kind of like the open weight models fall behind measured over the next year and the year to come. There's no, um, you know, guarantee that open weight continues to compete. And given it's a harder model to finance, um, absent state sovereign support, um, I, I think that's a fair assessment

Speaker A: assumption on that front. And we don't need to go this, maybe we'll tee up next week. Uh, but China, I remember five, six years ago, it's like they were paying crazy salaries to all biopharma individuals. And it's like, clearly that has worked to some degree. It's like I see tons of articles and demonstrations that like their output is, you know, hockey sticked upwards. It's like the same thing could be true for this next leg of growth, uh, with AI.

Speaker B: Well, certainly if like, part of the reason that worked with Biopharma is because China also has like, uh, significantly streamlined its clinical, um, trial process, particularly relative to the US So it's not just paying individuals. It's like, ah, you can get a trial started, finish the trial, get it submitted and get to market more quickly in China. And um, at least so far as we can tell, it's not that they're cutting corners in a way that means that the trial data is no good. US companies are looking at it and saying, hey, this is like meaningfully good work. So if the US government continues this stance of like, AI models are not allowed to be delivered right now there's a select group of customers that get the same performance for half the cost with OpenAI's newest model because the government has said, no, you're not allowed to release that to the general public. It's the same performance. This is like, yeah, it's great if you're the person on the invite list, but it's net bad for everybody else. If that extends out and continues and cludges up kind of like the US company's ability to deliver products to end customers, then that, you know, China is not going to hold back, like, yeah, everybody.

Speaker A: Even without that, even without that, I think there's a parallel. Right. Like China's ability to deploy energy and build data centers is far better than the U.S. right. It's like they do in a year what we think the US Needs to do over the next five years. And so it's like, I think similar in the way that you just laid out the biopharma case. The same could be very true for AI.

Speaker B: Yeah, though, um, getting, I mean energy is a more, you know, fungible good. Um, and so it, I think it's really not the long pole in the tent in terms of actually delivering frontier intelligence capability. And you could have energy free and you still have to get chips and you still have to figure out how to package the data into the chips and efficiently deploy them over a long time. So not energy free, but like, uh. And um, you know what, we'll get the energy from space with SpaceX and that'll be available to US companies before it's available to Chinese companies. So maybe that's the vector by which we maintain a lead. But I think net, it's confusing out there, folks. It's confusing.

Speaker A: Stay safe, don't drink peptides, don't inject.

Speaker B: And. It could be if it really gets geostrategically call it, uh, techy, uh, that Chinese data center that's distilling US intelligence out of US models gets a delivery from Star Fall that's like, uh, a, you know, set of drones that goes in and unplugs all of its servers one by one. So that's the place we're going.

Speaker A: Exactly the what would happen. All right, we'll see everyone next week.

Speaker D: Ark Investment Management, LLC is an SEC Registered Investment Advisor. Ark Wolf Financial and Public are separate, unaffiliated entities and do not have a relationship with respect to any company marketing or selling the products or services of the others. Ark Wolf Financial and Public make no representation or warranties regarding and expressly disclaim any responsibility or liability for any acts, omissions, losses or damages of the other entities or their affiliates, officers, employees or agents. No company shall be liable for any indirect, incidental, consequential or special damages arising

Speaker C: out of or in connection with this podcast or its contents.

Speaker D: The information provided in this show is for informational purposes only and should not be used as the basis for any investment decision and is subject to change without notice. It does not constitute, either explicitly or implicitly, any provision of services or products by Ark and investors should determine for themselves whether a particular investment management service is suitable for their investment needs. All statements made regarding companies or securities are strictly beliefs and points of view held by ARK and or show guests and are not endorsements by ARP of any company or security or recommendations by ARK to buy, sell or hold any security. Historical results are not indications of future results. Certain of the statements contained in this show may be statements of future expectations and other forward looking statements that are based on ark's current views and assumptions and involve known and unknown risks and uncertainties that could cause actual results, performance or events to differ materially from those expressions expressed or implied in such statement. ARC assumes no obligation to update any forward looking information. ARC and its clients, as well as its related persons may but do not necessarily have financial interests, insecurities or issuers that are discussed. Certain information was obtained from sources that

Speaker C: ARC believes to be reliable.

Speaker D: However, ARC does not guarantee the accuracy or completeness of any information obtained from any third party.

Speaker A: M.