FYI - For Your Innovation

SpaceX Starship reusability and Jensen Huang's departure reshape AI economics

29 min Listen to the original Episode page
SpaceXStarshipNVIDIAJensen HuangAI CapExLaunch CostsAI EconomicsNeo-clouds

Summary

Speaker 1 highlights SpaceX's Starship upper stage landing, claiming it could reduce launch costs to $100/kg, potentially removing orbital compute costs from economic equations. Speaker 0 discusses Jensen Huang's departure from NVIDIA, noting his aggressive investment in AI genomics and neo-clouds, and cites SemiAnalysis lowering AI CapEx forecasts by ~28%. The speakers debate AI demand drivers, with Speaker 1 citing a 97% year-over-year drop in intelligence costs correlating with increased spend, while Speaker 0 expresses caution about long-term willingness to pay.

Episode sections

  1. 0:44 – 9:54

    SpaceX Starship Upper Stage Landing and Cost Implications

    Speaker 1 discusses SpaceX landing the upper stage of Starship, describing it as the largest object returned from space and roughly 3x the payload of the Space Shuttle. He claims this reusability could lower launch costs to $100/kg from $600-$700/kg, with refurbishment taking weeks instead of 90 days, potentially removing launch costs from orbital compute economics.

  2. 9:54 – 19:55

    Jensen Huang's Departure and AI Investment Shifts

    Speaker 0 analyzes the impact of Jensen Huang's exit from NVIDIA, noting his role in aggressively investing in AI genomics, neo-clouds, and open-weight models. He highlights concerns about who will fill the gap in aggressive AI investment and mentions SemiAnalysis lowering CapEx forecasts by ~28%.

  3. 19:55 – 28:35

    Jevons Paradox vs. Sales Force Model in AI Demand

    The speakers debate whether AI demand follows Jevons Paradox (cheaper intelligence leads to more spending) or a traditional sales-force model (diminishing returns). Speaker 1 argues cost per intelligence has dropped 97% YoY, correlating with increased spend, while Speaker 0 expresses caution about companies' willingness to pay for AI beyond initial adoption.

Assets discussed

AssetKindStanceContext
SpaceXprivate companyneutralBut we will begin, we will begin with a discussion on SpaceX, some historic news um
NVIDIANVDAequitybearishSo, like net of it all, I think you can expect certainly less venture funding from NVIDIA, assuming Jensen doesn't come back uh into this kind of multi-omics type stuff.
Meta PlatformsMETAequityneutralUm, maybe offset by, you know, some US companies. I mean, you know, OpenAI has an open weight model, but Meta, Um, you know, potentially seeing an opportunity to actually go back

“Stance” is how the speakers talked about the asset, not a recommendation.

Claims worth verifying

Starship Flight 16 landed the upper stage, the largest object returned from space, roughly 3x the Space Shuttle's payload.

Unidentified speaker (speaker_1) ·0:44

How to verify: SpaceX official mission logs or press releases regarding Starship Flight 16; compare payload volume to NASA Space Shuttle specifications. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Falcon 9 launch costs are $600-$700/kg; reusable Starship could reach $100/kg.

Unidentified speaker (speaker_1) ·1:23

How to verify: SpaceX official pricing pages or industry reports (e.g., SpaceNews) comparing Falcon 9 and Starship cost projections. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Elon Musk stated refurbishment could take weeks, not 90 days.

Unidentified speaker (UNKNOWN) ·1:51

How to verify: Transcripts of SpaceX press conferences or interviews with Elon Musk post-flight 16. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

ARK previously modeled 90 days for Starship refurbishment.

Unidentified speaker (speaker_1) ·2:11

How to verify: ARK Invest research reports or publications from 2026-2026 discussing Starship operational assumptions. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Data center in space unit economics become viable at $300-$400/kg; at $100/kg, launch cost is negligible.

Unidentified speaker (speaker_1) ·7:55

How to verify: ARK Invest whitepapers or reports on orbital compute economics; verify against satellite manufacturing cost data. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

SemiAnalysis lowered AI CapEx forecasts by ~28%.

Unidentified speaker (speaker_0) ·11:26

How to verify: SemiAnalysis published report or newsletter from November 2026 detailing CapEx forecast revisions. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Jensen Huang was forced out by the US State Department due to concerns over Chinese access to next-gen chips.

Unidentified speaker (UNKNOWN) ·15:13

How to verify: SEC filings (e.g., NVIDIA Form 8-K or proxy statements) citing regulatory reasons for executive departure; news reports from reputable outlets. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Primary source

Cost of intelligence (per token) has collapsed 97% year-over-year.

Unidentified speaker (speaker_1) ·21:02

How to verify: ARK Invest internal data or published research on AI token costs; verify against major AI provider pricing changes from 2026-2026. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

A 1% drop in intelligence cost correlates with a 1.2% increase in total spend.

Unidentified speaker (speaker_1) ·21:28

How to verify: ARK Invest regression analysis or economic models linking AI compute costs to aggregate industry spend. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Anthropic's revenue onboarding slowed in Q3 2026 compared to Q2.

Unidentified speaker (UNKNOWN) ·22:13

How to verify: Anthropic's public financial disclosures or third-party tracking reports (e.g., SemiAnalysis) for Q2/Q3 2026. For recent events, verify 2026-08-19 through 2026-11-19. Use another period only when the speaker explicitly names it.

Do your own research

  • Verify SpaceX Starship Flight 16 landing details and upper stage recovery status via SpaceX official updates or spaceflight news outlets.
  • Check SemiAnalysis's original report for the specific methodology behind the 28% AI CapEx forecast reduction.
  • Review NVIDIA's latest SEC filings (10-Q or 10-K) for any disclosures regarding executive changes or strategic shifts in AI genomics investments.
  • Analyze historical data on AI inference/training costs per token to verify the claimed 97% year-over-year decline.
  • Research the regulatory environment for US cloud computing exports to China to assess the validity of State Department pressure claims.
  • Compare current launch cost estimates from SpaceX and competitors against the $100/kg threshold for orbital data center viability.
  • Investigate the definition and current market status of 'neo-clouds' and open-weight model infrastructure providers.
  • Look for counter-arguments to Jevons Paradox in AI demand literature, specifically regarding enterprise budget constraints and diminishing returns.
Full transcriptexpand

0:00Unidentified speaker (speaker_0): What did Jensen represent to the broader AI market? So when you take him out of the equation, who steps in to fill that gap of aggressive investment in AI? So

0:09Unidentified speaker (speaker_1): Jevon's paradox, where it's like you get this stuff cheaper, you find more uses for it, and then you expand across your uses.

0:17Unidentified speaker (speaker_0): Hello everyone, and welcome back to another episode of The Brainstorm. Storm Today it is just Brett and I. Sam is traveling, so we have lost our minds. moderator, the voice of reason. But we will begin, we will begin with a discussion on SpaceX, some historic news um over the weekend, and then we will get into Jensen and everything AI. So Brett, why don't you kick us off?

0:44Unidentified speaker (speaker_1): Sure. So SpaceX, um, uh on it the starship number 42, and it's um flight, I think it was flight 16 of its starship booster and rocket uh landed the top stage of its orbital um rocket, which is um uh historic. This is the largest by volume object that we have ever returned from space. So it's you know, on a payload basis, something like three, three X, what the space shuttle was able to do. Um it's also um financially and strategically important for SpaceX. Uh we think that with a reusable top stage, which we weren't sure was actually going to be possible, um, they could lower their costs per launch to around $100 per kilogram. From right now, we think on the Falcon And 9, nine, they're at you know six to seven $700 hundred dollars per kilogram. And on the Starship, SARShip, if they couldn't reuse the top stage, they might have been able to get down to 500. Um, but this is a material lowering in their cost inputs. It's also can allow them to accelerate their rate of progress because he,

1:51Unidentified speaker (UNKNOWN): uh

1:51Unidentified speaker (speaker_1): Elon, on inspecting the heat shield tiles on the on the on the top stage said we think we could turn this, refurbish it, and get it ready to launch again within a few weeks. That's that's much better than we. expected. We thought the first um starship to even successfully land, if it were to occur, would have taken 90 days of refurbishment. So their ship came in in much better shape than we anticipated. They landed. We actually didn't think they would even, even in the cases where they successfully land this thing, we thought it would take them six commercial launches before they could successfully land the top state. stage. Um, and you know, rather than taking three months to turn it, they think they can do it in a few weeks. So it suggests kind of cost, blended cost for launch for SpaceX, which is an important input into how much capital they have to um put up front to launch their satellites to generate revenue on, um, is coming in better than anticipated. Not to mention the fact that it was a totally mind blowing. mind-blowing and cool exhibition of engineering excellence. So that's the net, net, net.

2:55Unidentified speaker (speaker_0): Brett, what is the main hurdle to refurbishment? As in, you know, you're lowering the amount of time it takes to turn the starship around. What exactly goes into that?

3:07Unidentified speaker (speaker_1): Well, like if you look at the space shuttle, they would literally have to inspect every single tile and then handcraft replacement tiles for an any tiles that needed replacement and then and you know basically like do a custom kind of rebuild of the bottom of this uh of the space shuttle every time. Um so Starship is it's roughly the same as and you look at all of the kind of like return surface, all of the um protection layer and assess what needs to be taken off and built back in. And was there any kind of infiltration damage to the stainless steel surface under the tiles? And so, you know, conceptually you could um you might have to replace all of the tiles if you don't have very good tiles. And so then you would like basically take off the shell and put it back into kind To of the manufacturing assembly line and put them all on. What SpaceX has done here is one, they've made the tile tiles much more robust relative to the space shuttle tiles and designed them for manufacturer and come up with a mechanism by which to put them on relatively inexpensively. But if they only have to replace, you know,

4:15Unidentified speaker (UNKNOWN): I

4:15Unidentified speaker (speaker_1): don't know, 40 or 50 tiles, um, you know, then they could

4:19Unidentified speaker (UNKNOWN): go

4:20Unidentified speaker (speaker_1): through that pretty quickly. Uh and their aim is to get to something where it's like just really automated, or you do an inspection. and you know it's good to go. One advantage they have over the space shuttle is for now, there's no humans coming down in these things. So you can actually take better, you can take more risk because you're not putting any lives at risk. It's just your own capital at risk if you have to divert offshore and crash it. So I would expect that over time they'll they'll they'll like test the limits of how many, you know, how much damage the tiles can have and the ship still returns. There's some optimization of like, you know, if you're reusing the ship five or six times um and you have enough of them in inventory, um, then you actually are crossing some threshold where you're like, Okay, this one we know it's not gonna come back, but we're gonna launch it anyway, you know, and then have it incinerate on reentry. So

5:11Unidentified speaker (UNKNOWN): when

5:11Unidentified speaker (speaker_0): when they're inspecting the heat shield, how do they uh assess whether or not, you know, specific tiles can be reused again? And are the tiles meant to be reused for like how many do they expect how many reuses out of the specific heat shield do they expect to get per tile? Or for the tile specifically

5:34Unidentified speaker (UNKNOWN): I

5:34Unidentified speaker (speaker_1): don't know the answer to that latter question. I I you know I I think in the overall answer is, and this is call it general to Elon Musk Enterprises, is um they have, I'm sure, like a simulated expectation, but simulation runs into the you know fist of reality pretty quickly. And and so the the value they got out out of that previous starship landing and not exploding was actually being able to physically inspect and say, okay, how much, if we were to replace these, how many would we need to replace? Where would we need to replace them? And how can we kind of design so there's less infiltration in those areas? Same with this ship that comes back, as in they're going to, you know, look at the entire surface, um, assess where kind of they can make improvements, use the actual lived data to make additional improvements to um the ship. And it should increase their turn rate and ability to economically do this thing. And I think that's a certainly relative to the, you know, over the weekend there was also the United Launch Alliance's

6:44Amazon carrying rocket failed to get into orbit. Its upper stage engines seemed seem to fail in an unexpected way. or maybe, you know, its engines failed in an unexpected way. Um and there's just like the um they have so much less data. I mean they got data from this flight, but you know, they just don't have the launch cadence to gather the data to improve, to do more launches to gather the data to improve. And so um you know the fact that SpaceX is now going to get into a place where they're with less capital, they're going to get accelerated amounts of data from doing continued launches means that they should also increase their um, you know, capital efficiency and ability to know like where the limits are for the craft.

7:34Unidentified speaker (speaker_0): And this will be a question that will probably segue us into our next topic. which is Jensen out. Um, but how do you think about this relative to the opportunities surrounding orbital compute? What is this unlock when it comes to being able to set up data centers in space?

7:55Unidentified speaker (speaker_1): Yeah. We've previously talked about crossing the um around $300, $400 per kilogram threshold. is where the unit economics of a data center in space begin to pencil. Um, you know, that's subject to a lot of other unknowns, like cost to manufacture the satellite and cost of chips put in it, but that was a key cost input. And um, as you get down into the $100 a kilogram and below, that as an important cost input just falls out of the equation. And I think it's very likely that the limit. limiter on compute in space is no longer kind of launch or launch vehicle availability. That's the other like piece of this. If you're reusing the rockets, you know, they have a whole factory set up to manufacture these starship rockets, which nobody has ever really done in rocketry, uh, at least for manufacturing scale the way they have. And now they're going to begin to reuse them. And so their inventory of available potential launch vehicles, um, in terms of like launch mass to orbit, is just gonna explode. Uh and um so, in terms of if the competition in compute is one, if there is un not unlimited, but this huge demand for compute, which we think there is, the market is less certain of that, um, then then um kind of like

9:15SpaceX definitely has cornered the market on being able to do compute at scale. As in, it's highly likely that they'll be able to do a lot more and continue to grow it relative to themselves and others trying to operate on the ground, identify sites, negotiate with municipalities, get kind of like the the all of the complex negotiations in place to build these individual data centers. And so this is um seems clearly the way to manufacture compute at scale quickly. But let's turn to NVIDIA, Nick. I'd be curious to hear your opinion. Jensen is um now officially out. Lots of ramifications for like the entire industry, world, et cetera. What do you think about kind of like what it means for NVIDIA, what it means for? buyers of compute, what it means for the ecosystem and venture. What do you think?

10:08Unidentified speaker (speaker_0): Yeah, I think my big takeaway here is, you know, what did Jensen represent to the broader AI market? Like we can put, you know, just the discussion around NVIDIA on pause because I think the ramifications for the broader Our AI ecosystem are more interesting to speak to. Because I think what he represented was this agreement. aggressive investment approach to scaling AI. And so I think what people are rightly concerned with right now is what is the reflexive response by the industry when you take this central figure out of the equation that was investing in, you know, AI genomics plays, which was potentially providing backstops backstop for the neo-cloud business, which he essentially provided. propped up the neo-cloud business because his allocation to those companies over the traditional cloud providers supported the early growth of those companies. So when you take him out of the equation, well, who steps in to fill that gap of aggressive investment in AI? I think you're you're going to see, and you know, I sent you the the note this morning, semi-nounced. semi-analysis, which has been riding the uh call it the coattails of the AI trade and helping people distinguish right from wrong and you know where this market can head, just lowered their

11:26forecasts by I think roughly 28% uh for CapEx spend. And you know, you can delineate how that translates and what it means for AIR ARR for the entire industry. But I think, you know, people are trying to make sense of this. There's a lot, you know, of conversation around the CFO. Seems like her investment approach, which, you know, probably is what what you want from a CFO, is completely different than Jensen. And she's going to take potentially a much more conservative stance on AI. But you know, we'll we'll find out a lot tomorrow when when they report. Um And maybe we should bring Frank on next week. to discuss if there's anything pretty interesting that comes out of that earnings, which I'm sure there will be. But you know, me when I look at this I say, you know, there's gonna be some uncertainty in the market and that is gonna cause, you know, delays in not just, you know, spend for data centers, but how people approach AI generally.

12:22Unidentified speaker (UNKNOWN): I

12:22Unidentified speaker (speaker_0): think he was that big

12:23Unidentified speaker (UNKNOWN): Yeah, I think net that much. I think

12:25Unidentified speaker (speaker_1): well yeah. I think I think I I would distinguish between things that are strategically rational for NVIDIA to do and the velocity at which they're trying to pursue those strategies. And like as in, I would expect that a um

12:43Unidentified speaker (UNKNOWN): you know

12:43Unidentified speaker (speaker_1): a CFO leading the business might still conduct those strategically rational acts, but would do so at a more measured pace than he was trying to push. essentially the balance sheet. And so because the the one way to interpret his set of actions is basically he was doing everything he could to create um competition in other layers of the value chain. As in the infrastructure as a service layer, like he talks about a five-layer cake, I think it's a useful way to think about kind of like the AI kind of stuff. stack. There's other ways to think about it, but you have power, you basically have the um chips, you have the infrastructure as a service guys, then you have the kind of like AI platforms and then you have the applications built on top of them. And so if you think about where he was deploying resources, you know, he's occupying the chip layer. So he wants to maximize revenue or uh economic power to the chip layer. So that means he needs additional infrastructure partners, not just AWS and Azure, right? And so he injects capital into seeding these neo clouds that are going to be really good at accelerators to create more kind of buyers for his chips at that portion of the market. Then at the um the model layer, you know, he needs it to not just be an anthropic and OpenAI game. So he was

14:03kind of like um helping SpaceX to bootstrap. And he was um investing heavily in uh enabling open source or open weight models because that provides a competitive kind of layer against those core frontier labs at that part of the chip stack. Uh including, you know, um allowing you know infrastructure as a service um access to the chips in China is an important part of that creating these open weight models. Uh and so kind of the and then even at the application later, directly investing in kind of venture startups in the biology space where there's this likely very, very large addressable market, but it's more kind of speculative as to how we get there. But that could be a demand pull for chips because there's so much data in biology. Um and so

14:51Unidentified speaker (UNKNOWN): I

14:52Unidentified speaker (speaker_1): that's kind of how I think of how they were allocating. I would think a CFO would say, well, that's all strategically rational. It's just a question of how much am I levering up the balance sheet on the basis of the next 12 months of demand are going to be high enough across all of those different things, you know. Uh, and I think, particularly given the US regulatory,

15:13Unidentified speaker (UNKNOWN): I

15:13Unidentified speaker (speaker_1): mean, he was basically forced out by the State Department, I think is broadly the reporting and probably right, because they're concerned about Chinese companies getting access to these next gen chips even through kind of a cloud interface. Um, you know, that probably um dampens or maybe just clamps down on that channel for NVIDIA chips, which was basically selling in chips to willing buyers who would then turn around and create tools out of those chips that provide

15:39Unidentified speaker (UNKNOWN): you

15:39Unidentified speaker (speaker_1): know, open weights that provide inference demand for those chips because you're not having to pay OpenAI and Anthropic Source Margin. So, like net of it all, I think you can expect certainly less venture funding from NVIDIA, assuming Jensen doesn't come back uh into this kind of multi-omics type stuff. You can expect like a more measured financing of the neo clouds, potentially, though I think that's still strategically rational. You can definitely expect that there's going to be a difficulty for the at least Chinese companies getting access to next-gen chips to keep open weight up with the frontier. Um, maybe offset by, you know, some US companies. I mean, you know, OpenAI has an open weight model, but Meta, Um, you know, potentially seeing an opportunity to actually go back into open weight. way the meta has pulled back to that from that to some degree. Um, so I think that's kind of like how I think of the kind of um staggering ramifications. Um, and then there's I think there's this like so uh SpaceX, for instance, as a US buyer of chips, maybe they have more leverage over NVIDIA because NVIDIA

16:46doesn't have this Chinese channel to sell into anymore. Um, and um certainly. he had all these supply chain relationships below him that he had secured with handshakes, where, you know, Google is out there being like, hey, you could package our chips. We're a more reliable partner now because who knows what's happening with NVIDIA because maybe they're not going to as aggressively spend. And so it does shake up kind of like AMD MD probably doing the same set of actions. It shakes up kind of the degree to which NVIDIA had aligned all. of the supply chain to allow them to continue to accelerate. And so it probably does make it more likely over the medium term that NVIDIA's share of kind of like the cutting edge diminishes. I mean, I think that's like the full set of things here.

17:31Unidentified speaker (speaker_0): I think what will be interesting too, and you you've talked about this more from the the lens of SpaceX, where you know sp because SpaceX is able to bring compute on faster than most of its other competitors, they're the likely company that ends up finding, you know, where demand ultimately sits. And I think because you're now and and this might be more of a US specific story, but because you're now going to take all of those chips that were supposed to be allocated to China or were finding their way into China and it's all going to, you know, presumably stay in, you know domestic territory in the US. I think a lot of these companies are going to get their hands on the allocation they had seemingly been wanting. And so I do think it will be interesting to see you might get a faster answer into what like true demand is there for you know GPUs. I you know, I think one thing to watch is GPU rental pricing. Like a lot of, I think outside of maybe power bottlenecks, you should have a lot more come online, at least in the US, because those chips, if there are willing buyers, they they will, you know, be sold to them here, not make their way across the the sea. So I I'm curious to

18:40Unidentified speaker (UNKNOWN): see how

18:41Unidentified speaker (speaker_0): that plays out. And and again, you know, when when you are in a a moment of scarcity, your your indication for what you want, you're I think we were listening to that Howard Marks call this morning with everyone on the TV.

18:54Unidentified speaker (UNKNOWN): You always

18:54Unidentified speaker (speaker_1): overorder because it like

18:55Unidentified speaker (UNKNOWN): the price

18:56Unidentified speaker (speaker_1): is high and you don't want yeah, of course.

18:59Unidentified speaker (UNKNOWN): Right.

18:59Unidentified speaker (speaker_0): So now we're

19:00Unidentified speaker (UNKNOWN): gonna find out the

19:01Unidentified speaker (speaker_0): flip side of that and like where true demand actually is, because I don't think you have that main figurehead being you know the leading force forced to like I think what Jensen was doing was that foreign video for you know you know going to TSM C and saying here's what we need and you know making it a huge number so that they remain 80% of the market. Now if you have a CFO step in and say, well we actually only need this amount because this is you know what we're what we're seeing, I think that changes you know everything in in a very short amount of time.

19:33Unidentified speaker (UNKNOWN): And

19:33Unidentified speaker (speaker_0): you'll find that

19:34Unidentified speaker (UNKNOWN): But

19:34Unidentified speaker (speaker_1): see then then then it distills down to like those does does OpenAI anthropic and

19:41Unidentified speaker (UNKNOWN): you know

19:42Unidentified speaker (speaker_1): SpaceX's kind of call it AI software revenue collectively, you know, do they continue to accelerate at anything close to the historical pace of acceleration? Or was basically differ,

19:56Unidentified speaker (speaker_0): I think. Which

19:57Unidentified speaker (UNKNOWN): is

19:57Unidentified speaker (speaker_0): interesting. Not interesting. I mean, we're usually on the opposite side of things, but I do I'm more cautious around, you know, ARR and demand, at least in the next year, than you are. But curious, you know, you you should say your

20:12Unidentified speaker (speaker_1): Well, like I think there's two ways to one, kind of your caution comes from the fact that, you know, Ford companies are saying to you, like, hey, look, yes,

20:22Unidentified speaker (UNKNOWN): this

20:22Unidentified speaker (speaker_1): AI stuff is super useful, and now we're figuring out how to efficiently extract, call it. the capability out of it without as intense a set of AI spend. Um, and so what does that mean? It means instead of like giving all their engineers access only to the frontier and the engineers just use the frontier and crank out tokens, uh, we're um aggressively developing our own harnesses so it shifts to the most efficient model for the task. Um and if you look at the ramp data on a um on a like a blended weighted average adjusted for intelligence token basis, like the cost for intelligence to these companies has collapsed 97% year over year, basically. My um confidence in part arises from the if you look at that same data, you know, there is a clear correlation between when they get cheaper per um intelligence per like intelligence-weighted tokens, um, these firms tend to spend more in aggregate, not less, blended across. As in their token

21:28demand for every 1% drop in basically um intelligence uh uh costs, uh you end up with a 1.2% increase in. the total spent. So Jevon's paradox where it's like you get this stuff cheaper, you find more uses for it, and then you expand across your uses. And so that's the the high level like way I would think about the industry. Uh and I understand that people don't believe that. As in And they they just don't. And and they don't there's a like and it

22:01Unidentified speaker (UNKNOWN): some

22:01Unidentified speaker (speaker_1): of it's feels but also there's an important, like backward-looking reason they don't. They look at anthropics revenue in particular and they say, Well, you onboarded a bunch of revenue in the second quarter

22:13Unidentified speaker (UNKNOWN): on

22:13Unidentified speaker (speaker_1): a dollars basis, and then in the third quarter, you onboarded less. So if we just assume, like with a typical The software business, I have a sales force, I say, go sell this product. They go out, they're gonna going to identify the best prospects. first. They're going to go to them, and those best prospects they'll like land them and they'll land this much revenue. And then they say, okay, now I'm going to the next best set of prospects. They go out to that. It's like smaller, slightly smaller, slightly harder to sell. And so same sales force lands like this much revenue. You'd naturally say, Well, what's coming next? It's an even harder set of prospects. So then you would have it,

22:48Unidentified speaker (UNKNOWN): you know,

22:48Unidentified speaker (speaker_1): increment down again.

22:50Unidentified speaker (UNKNOWN): If

22:50Unidentified speaker (speaker_1): you use that to like say, well, this is what the industry is going to do, you might end up with, you know, collectively like less than a trillion dollars in spend by 2030.

23:01Unidentified speaker (UNKNOWN): Like

23:01Unidentified speaker (speaker_1): that's what you, it would just like increment down that way. Whereas,

23:05Unidentified speaker (UNKNOWN): you

23:05Unidentified speaker (speaker_1): know, if instead you say, look, on a dollars per compute basis, this is what they're monetizing at, and this is how much compute they have lined up. And if you look over time, they've been monetizing. better on a dollar per compute basis over time because they're delivering more intelligence per unit of compute. And then you look at the planned compute they have on coming online, then you end up with this, you know, high single digit trillions in revenue by 2030.

23:32Unidentified speaker (UNKNOWN): And

23:33Unidentified speaker (speaker_1): so it's really a question of is it a sales force tell selling AI or is it the intelligence itself? of AI that's selling AI? And I think it's a different beast. Like as in, I understand why a software as a service analyst would be like this is the way that software businesses work. You have to have somebody go in and convince people to use

23:52Unidentified speaker (UNKNOWN): it.

23:52Unidentified speaker (speaker_1): And then they use it and that's kind of it. Because this is in a utilization based, so it's really usefulness-based. usefulness based. I think that like the dynamics of how this sells in.

24:03Unidentified speaker (UNKNOWN): will

24:03Unidentified speaker (speaker_1): play out very differently. And I think we'll see like strong evidence and like what does Inf Topic's fourth quarter look like? What does their commentary look like as they're coming to market? But

24:13Unidentified speaker (UNKNOWN): yeah. I think it's

24:14Unidentified speaker (speaker_0): some some in-between of the two where you know the sales force is important to growth, but I do believe in what you're saying. My bigger question is just, you know, yes, there are maybe infinite. numbers of use cases to which you can deploy AI against. But my question is when you distill down, you know, all of those use cases, what are the ones where companies are actually willing to pay top dollar? And I think what we've seen over this last year is like the diffusion of AI into those categories was extremely fast because it is a very intuitive product for people that understand it. That's why coding a adoption, you know, AI coding solutions, you know, that technology was adopted extremely fast. But I think what happens now is you get into, you know, a mode and a use of the technology where you're going to figure out whether or not it's actually worth to spend money on it. As in, you may still be using AI across your marketing and legal and X, Y, and Z.

25:14but how much are you actually willing to pay for it is the big question. And I just I think you have to have a few quarters. You have to you have to understate like put it this way you can't have your AI spend continually outgrow your revenue growth. So there is a gating factor

25:31Unidentified speaker (UNKNOWN): you can of which if

25:31Unidentified speaker (speaker_1): you're offsetting it if you're offsetting it with cuts and spins somewhere else.

25:35Unidentified speaker (speaker_0): Yes.

25:35Unidentified speaker (UNKNOWN): But yes.

25:36Unidentified speaker (speaker_0): Yes. That is that that is but I do think that takes potentially longer than the market expects because everyone's looking at this year and saying, well, this growth was staggering. And if we we play this out and you know, there's a new model that comes out with a potential new use case and we see this next inflection point, then yes, you can get into this low single trillions of dollars in revenue. And I just think there's going gonna to be bumps and hurdles along the way that the market may or may not be pricing in at the moment. I I think it's gonna be, you know, you'll find out if what I'm saying or what you're saying is right or wrong within like the next two quarters. Like I think it plays out, you know, given what's happening with NVIDIA.

26:14Unidentified speaker (UNKNOWN): Like

26:14Unidentified speaker (speaker_1): I think

26:15Unidentified speaker (UNKNOWN): I think what what will happen is is

26:17Unidentified speaker (speaker_1): is financially it'll look like I'm right, but the market will still disbelieve Believe it.

26:21Unidentified speaker (UNKNOWN): That's

26:21Unidentified speaker (speaker_1): where I I think

26:23Unidentified speaker (UNKNOWN): I think that's

26:23Unidentified speaker (speaker_1): what's gonna happen. We will

26:25Unidentified speaker (UNKNOWN): see.

26:25Unidentified speaker (speaker_1): We will see. Yes.

26:27Unidentified speaker (UNKNOWN): All right.

26:27Unidentified speaker (speaker_0): All right. Thank you everyone. Uh we will be back next week. ARC Investment Management LLC is an SEC registered investment advisor. ARC, 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. ARC, Wolf Financial and Public make no representation or warranties regarding and expressly disclaim

26:50Unidentified speaker (UNKNOWN): any

26:50Unidentified speaker (speaker_0): 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 out of or in connection with this podcast or its contents. 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 ARC, 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 ARC and/or show guests and are not endorsements. by ARC of any company or security or recommendations by ARC 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 ARC's current views and assumptions and involve known and unknown risks and uncertainties that could cause actual results performed. performance, or events to differ materially from those expressed or implied in such statement. ARK assumes no

28:09obligation to update any forward-looking information. ARK and its clients, as well as its related persons, may, but do not necessarily have financial interests in securities or issuers that are discussed. Certain information was obtained from sources that ARC believes to be reliable, however, ARC does. not guarantee the accuracy or completeness of any information obtained from any third party.

This page was produced automatically: the audio was transcribed with nvidia/parakeet-tdt-0.6b-v3+sortformer-v2.1 and interpreted by Qwen3.6-35B-A3B-UD-IQ3_S.gguf. Transcription and interpretation can contain errors; always confirm against the original episode. Nothing here is investment advice.
All insights