Brett Winton argues Anthropic's $1.4T IPO valuation is cheap given $80B revenue and 350% growth despite election-driven supply constraints.
Summary
Brett Winton of ARK Invest discusses Anthropic's upcoming IPO, estimating its revenue at approximately $80 billion with 350% year-over-year growth. He argues that a $1.3-1.5 trillion valuation is cheap relative to public market AI stocks, citing low price-to-sales multiples. Winton highlights that midterm election results may reduce data center capacity by 10-15%, potentially shifting profitability to infrastructure providers. He contrasts Anthropic's enterprise strength with ChatGPT's consumer dominance, noting AI agent penetration is only ~4% of knowledge workers. Finally, he suggests improved model efficiency will boost gross margins as compute costs remain flat.
Key takeaways
- 1 Data centers can reduce community electric power costs by 20-40% and increase local employment by ~20%, softening political blowback.
Unidentified speaker (speaker_2) · 4:15
Episode sections
- 0:00 – 5:29
ARK Invest Introduction and Midterm Election Impact on AI Infrastructure
Brett Winton introduces ARK Invest's focus on disruptive innovation and discusses how the Democratic midterm wins may reduce data center capacity by 10-15%, impacting Anthropic's revenue growth potential.
- 5:29 – 9:20
Data Center Economics and Consumer vs. Enterprise AI
Winton argues that existing data centers may become more valuable due to supply constraints, while noting that high compute costs could force companies like Meta to prioritize enterprise clients over free consumer products.
- 9:20 – 16:34
Anthropic Valuation and Revenue Growth Forecasts
Winton estimates Anthropic's revenue at ~$80B growing 350% YoY, suggesting a $1.3-1.5T IPO valuation is cheap relative to public market AI stocks, despite market skepticism about sustained hyper-growth.
- 16:34 – 20:06
Market Penetration and Historical Analogies
Winton contrasts ChatGPT's consumer dominance with Claude's enterprise strength, estimating AI agent penetration at ~4% of knowledge workers, comparable to the internet in 1990/1991.
- 20:06 – 22:44
Profitability and Future Margins
Winton describes Anthropic's profitability as accidental due to rapid growth, noting that improved model efficiency should further increase gross margins as compute costs remain flat.
Claims worth verifying
ARK Invest is invested in Anthropic through the ARK Venture Fund.
Unidentified speaker (speaker_0) ·0:20
How to verify: Check ARK Invest's public disclosures or the ARK Venture Fund's prospectus for portfolio holdings. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Primary sourceThe midterm election results will reduce active gigawatts entering the market by 10-15% on a forward basis.
Unidentified speaker (speaker_2) ·2:35
How to verify: Review ARK Invest's 'Big Ideas' report or related research notes published between 2026-08-13 and 2026-11-13 for the specific methodology and data sources used in this estimate. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
SpaceX offers communities 20-40% reductions in electric power costs and increases local employment by ~20%.
Unidentified speaker (speaker_2) ·4:35
How to verify: Check SpaceX's community agreements or press releases regarding data center projects for stated economic impact figures. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Anthropic's gross margin is in the 85% range.
Unidentified speaker (speaker_2) ·9:59
How to verify: Verify against Anthropic's financial statements or filings with the SEC (if available) or private market valuation reports during the IPO period. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic is generating approximately $80 billion in recurring revenue.
Unidentified speaker (speaker_2) ·10:40
How to verify: Check Anthropic's audited financial statements or investor presentations released prior to or during the IPO. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Anthropic's revenue is growing at 350% year-over-year (annualized from the last quarter).
Unidentified speaker (UNKNOWN) ·10:49
How to verify: Calculate growth rate using Anthropic's reported revenue for the most recent quarter and the same quarter in the prior year from financial filings. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Primary sourceOnly four companies (Micron, NVIDIA, OpenAI, Anthropic) have delivered >$30B incremental revenue growth while growing >3x YoY.
Unidentified speaker (UNKNOWN) ·12:24
How to verify: Review historical financial data for these companies and screen for companies meeting the >$30B incremental revenue and >3x YoY growth criteria using SEC EDGAR filings. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Primary sourceAnthropic's IPO valuation is expected to be between $1.3 and $1.5 trillion.
Unidentified speaker (speaker_2) ·15:54
How to verify: Check prediction markets (e.g., Polymarket) or investment banking estimates reported in financial news between 2026-08-13 and 2026-11-13. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Primary sourceChatGPT accounts for 70% of engagement hours across all chatbots.
Unidentified speaker (speaker_2) ·17:04
How to verify: Verify against third-party analytics firms (e.g., SimilarWeb, App Annie) or OpenAI's own usage statistics reported during the period. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Anthropic achieved two quarters of unintentional adjusted profitability.
Unidentified speaker (speaker_2) ·20:26
How to verify: Check Anthropic's income statements for the two most recent quarters, adjusting for convertible note fair value changes if applicable. For recent events, verify 2026-08-13 through 2026-11-13. Use another period only when the speaker explicitly names it.
Do your own research
- Verify Anthropic's S-1 filing for audited revenue figures and growth rates against the reported $80B and 350% YoY.
- Check EIA or utility commission data for actual data center gigawatt capacity additions and regulatory delays post-midterms.
- Review Anthropic's S-1 for gross margin trends and compute cost structures to assess margin expansion potential.
- Compare Anthropic's price-to-sales multiple against public AI peers like NVIDIA, Micron, and Microsoft in recent SEC filings.
- Analyze Meta's Q3/Q4 earnings calls for evidence of shifting focus from consumer agents to enterprise clients.
- Research historical internet penetration data from 1990-1991 to validate the ~4% knowledge worker adoption analogy.
- Identify conflicts of interest for ARK Invest regarding any holdings in Anthropic or related infrastructure providers.
- Check Anthropic's S-1 for details on customer concentration and contract terms with enterprise clients.
Full transcriptexpand
- 0:00
- Unidentified speaker (speaker_0): Anthropic, the company behind Claude, is set to make its debut on the stock market any day now. It's set to be one of the biggest listings in history, and investors are asking one question: is it worth the price? On this special episode of the Switzer show, I'll put that question to Brett Winton, chief futurist at
- 0:20
- Unidentified speaker (speaker_0): ARK Invest. ARC ARK is currently invested in Anthropic through the ARC ARK Venture Fund. Brett reckons investors could look back in two years and say, I can't believe you could have it for that price. Enjoy the episode.
- 0:37
- Unidentified speaker (speaker_1): Thanks for joining us, Brett.
- 0:39
- Unidentified speaker (speaker_2): My pleasure.
- 0:40
- Unidentified speaker (speaker_1): So, so
- 0:41
- Unidentified speaker (UNKNOWN): just
- 0:41
- Unidentified speaker (speaker_1): to
- 0:42
- Unidentified speaker (UNKNOWN): position you
- 0:43
- Unidentified speaker (speaker_1): with my audience,
- 0:45
- Unidentified speaker (UNKNOWN): how
- 0:45
- Unidentified speaker (speaker_1): would you be described
- 0:47
- Unidentified speaker (UNKNOWN): by the
- 0:48
- Unidentified speaker (speaker_0): business community
- 0:49
- Unidentified speaker (UNKNOWN): in the
- 0:49
- Unidentified speaker (speaker_1): US?
- 0:52
- Unidentified speaker (UNKNOWN): I
- 0:52
- Unidentified speaker (speaker_2): think it depends on who you ask. Um, but I've been director of research from Inception at ARK and Best, uh, and have been now the chief futurist. And at ARC, we focus purely on disruptive innovation, and we have since inception in 2014. Um, and so I think for a lot of people, they think that I
- 1:12
- Unidentified speaker (speaker_2): do a reasonable job of actually understanding where technology is moving and how it's going to structurally change the world. So just to give you an example, you know, we publish this big ideas report every year on kind of the big technologies and where we think they're headed. In 2019, the first page of the report said that deep
- 1:32
- Unidentified speaker (speaker_2): learning, which is neural nets AI, what we have now, is going to change the way in which software is written. And lo and behold, here we are, basically a business cycle later, you know, seven years later. Um, and indeed, all software is being written by deep learning. Uh, and so, you know, my job is to
- 1:52
- Unidentified speaker (speaker_2): figure out yeah, what structural shifts are going to happen and what they mean for business a and investment. investing.
- 1:59
- Unidentified speaker (speaker_1): Okay. Before we start talking about anthropic, clearly. the the context of
- 2:06
- Unidentified speaker (speaker_1): what's going on right now and its implication for anthropic is obviously in the minds of lots of investors. Now the the Democrats' success at the midterms has meant
- 2:21
- Unidentified speaker (speaker_1): many people are arguing that the pace of the AI change that we are expecting before the midterms is likely to be slowed down. What's your view of that view?
- 2:35
- Unidentified speaker (speaker_2): Uh I think you should disaggregate the impact of the midterm elections into two categories. Um one is kind of the pace at which the underlying data centers and compute infrastructure is going to be built and its impact on that. And then the other is how kind of the regulatory apparatus is going to treat,
- 2:55
- Unidentified speaker (speaker_2): call it the frontier labs or these, you know, the cutting-edge AI systems. On the first case, as far as we can tell, um, it probably the mid impact of the midterm elections is to reduce by roughly 10 to 15 percent the amount of active gigawatts that you would expect to enter the market on a forward basis.
- 3:16
- Unidentified speaker (speaker_2): So a very simple way to interpret it is that means that OpenAI and Anthropic will be able to deliver only 10 to 15% less revenue than they otherwise would have at every kind of milestone going forward, because their ability to deliver, you know, these products in customers is really constrained by the
- 3:36
- Unidentified speaker (speaker_2): amount of computing power that they have. I do think that, you know, um the data centers are this problem for people that don't have them in their town, but once they get them in their town, they're like, oh, please give me more. And so that should actually cause some of the kind of political blowback
- 3:56
- Unidentified speaker (speaker_2): to soften. It's the same with robotaxis. robo taxis. People general nationwide aren't sure about robotaxis, robo taxis, but if they have autonomous vehicles in their town, they're like, oh, they are very politically popular because it means they can send their kids to the mall without having to drive their kid. And it means their kid is not going to get hit by a car while riding on a bike because they're safer than
- 4:15
- Unidentified speaker (speaker_2): human drivers. So with data centers, you know, if you look at what SpaceX is telling communities, they are going to knock 20% off or up to 40% off the cost of electric power for the people in the community or reduce property taxes by 20%, and fund all kinds of infrastructure, plus bring in, you know, literally increase the employed workforce in the
- 4:35
- Unidentified speaker (speaker_2): town by by 20 or so percent. Um, so these are actually amazing, both amazingly clean and amazingly economically beneficial businesses to have in your town. And so I do think that will help to unstick some of this, but um, it's likely in the immediate term there'll be some softening in the ability of data centers to get
- 4:55
- Unidentified speaker (speaker_2): built.
- 4:56
- Unidentified speaker (UNKNOWN): Um,
- 4:56
- Unidentified speaker (speaker_2): so that's like the one that's an area a lot of people focused. Um, I I actually there is so much skepticism in the market generally about AI. Um, that it it's not anthropic and an open OpenAI AI are aren't being um valued as if they're going to deliver the revenue growth that they well. could and should if they
- 5:16
- Unidentified speaker (speaker_2): get even like a small amount of the data centers that we expect are going to come online. So it's um kind of like a knee-jerk reaction to a revision in a future people already didn't believe.
- 5:29
- Unidentified speaker (speaker_1): Yeah. So there's two uh I think um implications of what you said. One is, and this is a question,
- 5:38
- Unidentified speaker (UNKNOWN): as a
- 5:39
- Unidentified speaker (speaker_1): consequence of the resistance for the development of new data centers, at least in the short to medium term, does that effectively increase the value of existing data centers? Because the the A, they're going to be threatened by by less data centres coming, and B, they arguably could charge more because
- 5:59
- Unidentified speaker (speaker_1): they're the the only game in town.
- 6:02
- Unidentified speaker (speaker_2): Yes, yes. I I think that there is a uh if you can not just existing data centers, but if you're a company like SpaceX SI, which can successfully build this infrastructure and also can launch data centers into space where you know you don't have planning meetings you need to go to to go into space. And so uh
- 6:21
- Unidentified speaker (speaker_2): the election didn't really impact you know those prospects in a material way. Um then the compute that gets built is more valuable. And, you know, we think that it results in some share shift between um the gross profitability of OpenAI and Anthropic versus their data center vendors where
- 6:41
- Unidentified speaker (speaker_2): they have to pay a little more. Um, but you know, those companies' company's gross margins is it's very healthy, so they can pay,
- 6:48
- Unidentified speaker (UNKNOWN): they
- 6:48
- Unidentified speaker (speaker_2): can, they can afford to pay a little more. Um and ultimately that will just cause more,
- 6:53
- Unidentified speaker (UNKNOWN): you
- 6:54
- Unidentified speaker (speaker_2): know, more compute to get built because it'll look that much more profitable to build the physical infrastructure.
- 6:59
- Unidentified speaker (speaker_1): Yeah. But but effectively the cost to consumers is going to remain probably higher than it would if there's a a greater proliferation of data centers and AI businesses being given, I guess I don't want to use the the term in a uh derogatory way. the open slather, which you kind of expected under the
- 7:19
- Unidentified speaker (speaker_1): Trump era, where the Democrats are going to be probably a little bit more negative, a little bit more circumspect about what they let AI companies get away with.
- 7:29
- Unidentified speaker (speaker_2): Yeah, I mean, I I do think that the cost to consumers generally will um be slightly higher because you have less compute infrastructure. And the, you know, I the thing that I am most strategically interested in is um the likes of Meta and SpaceX's,
- 7:49
- Unidentified speaker (speaker_2): GrockBot, Meta's Muse, InSync, these companies that are trying to provide consumer-facing agents to end consumers
- 7:57
- Unidentified speaker (speaker_2): because they're there are extraordinarily valuable business clients whose needs can be met with the computers that are instead devoted to these consumers. And it's it's actually
- 8:10
- Unidentified speaker (speaker_2): if if you limit the compute supply too much, then it'll effectively price out that consumer business. Where if I am meta and I I have to decide whether or not I'm gonna let somebody use To Muse for free, or I have Anthropic knocking on my door and saying they'll pay three 3x X over the margin. mark for kind of my compute. Well, then maybe you make the choice to
- 8:30
- Unidentified speaker (speaker_2): pull the plug on Muse. Um, so I, you know, in some ways, we are there are so the if you look at Anthropic's gross margin, the enterprise business is extraordinarily um cash flow generative and healthy. The consumer business is companies, including ChatGPT and Meta and SpaceX,
- 8:49
- Unidentified speaker (speaker_2): investing in a prospective future where they think that'll be you know a really good strategic platform. But if you put too much um financial pressure on the ecosystem, it'll just cause everybody to flee into the enterprise market and some of the consumer products will, you know, have to be scaled back in terms of their ambition.
- 9:08
- Unidentified speaker (speaker_1): Okay. So Brett, uh I I know after talking to Kathy, it it makes total sense that you you guys have had had a pretty strong view on where anthropic was going to be going.
- 9:20
- Unidentified speaker (speaker_1): And um as a consequence of the changes since the midterms,
- 9:26
- Unidentified speaker (speaker_1): what kind of discounting are you factoring in on, say, your revenue forecasts, your earnings forecasts as a consequence?
- 9:39
- Unidentified speaker (speaker_2): Yeah. I mean, like I said, you know, if they have access to 10 to 15% less gigawatts, then you would just take that off the top line. And then, you know, the most conservative case would be take it off the top line and say they still have to pay the same amount to the compute provider. But we actually, so their gross margin gets cut by that much. So
- 9:59
- Unidentified speaker (speaker_2): from the 85%
- 10:00
- Unidentified speaker (speaker_2): range, you know, down into the mid. 70s, which would still be very comfortable. Um, yeah, as it happens, we think there's somewhat of a split. But the reality is, you know, this is a company that um the market uh across OpenAI and Enthropic, they just doubt that these companies are
- 10:20
- Unidentified speaker (speaker_2): going to be able to grow at anything near the pace of growth they've been growing so far. Because if they if the market believed it, then you know the reputed valuation for anthropic, which you know people are rumoring around 1.4 trillion, uh, it would be coming in much higher because it's a company that's doing on
- 10:40
- Unidentified speaker (speaker_2): the order of 80 billion dollars in um in uh recurring revenue, growing on the order of 10x year over year. Now
- 10:49
- Unidentified speaker (UNKNOWN): we
- 10:49
- Unidentified speaker (speaker_2): don't think it's going to grow 10x year over year going forward, but if you take even the last quarter annualized, it's growing on the order of, you know, 350%, like four and a half X year over year. So, you know, even if you slow it down to twice year over year, like then you're going from 80 to 160 billion in revenue at very
- 11:09
- Unidentified speaker (speaker_2): high gross margin. If you're paying $1.4 trillion for a company that is doing 106
- 11:15
- Unidentified speaker (speaker_2): 160 billion dollars annualized is less than 10 times kind of annualized sales. You know, that's much cheaper than lots of public market exposures you can get, particularly those that are call it AI forward public market exposures. So there it you know as you can imagine we have um
- 11:35
- Unidentified speaker (speaker_2): you know uh more aggressive expectations than just uh doubling uh embedded in our um you know modeling uh for the company. And so I actually think that um Wall Street generally right now, at this moment, has um trouble
- 11:54
- Unidentified speaker (speaker_2): understanding um these companies, SpaceX, Anthropic, and probably OpenAI, when it comes, because of their incredible scale.
- 12:04
- Unidentified speaker (speaker_2): It means it's it's you can no longer interpolate and say, oh, well, I've seen a company like this before. And so then this is therefore what it should be. Nobody has seen a company this big growing this quickly. And so people are maybe rightfully really doubtful about the fact that it'll be able to continue to grow at all.
- 12:23
- Unidentified speaker (UNKNOWN): There
- 12:23
- Unidentified speaker (speaker_2): are there
- 12:24
- Unidentified speaker (UNKNOWN): are
- 12:24
- Unidentified speaker (speaker_2): 32 companies in the world are or at least in the US as far as we can tell, that have ever um delivered more than $30 billion of incremental revenue growth. Of those, only four have done so by growing more than three X year over year. And that's Micron, NVIDIA, OpenAI in this last
- 12:44
- Unidentified speaker (speaker_2): quarter, and Anthropic. And Anthropic did it growing 10x year over year. So everybody's like, I don't know. And when people do that, when they have to, when they can no longer interpolate, often they just say, Well, I'll just sit this one out and see. Um, and that's really, you know, both on the venture side and on the public market
- 13:04
- Unidentified speaker (speaker_2): side, where I think we do some of our best work because, you know, we have a good developed case for um how big at least the um knowledge work market is for these companies. which is on the order of $40 trillion.
- 13:18
- Unidentified speaker (speaker_2): And so we're fairly confident, I would say quite confident, that we are very early in the game between Enthropic and OpenAI and everybody else doing knowledge work LLMs, basically, there's 150 billion or so in revenue right now. So you could, you know, 100x a hundred X from
- 13:38
- Unidentified speaker (speaker_2): year. here and still not have um actually saturated the addressable market so far as we can tell.
- 13:45
- Unidentified speaker (speaker_1): Yeah. So at what price do you think anthropic will IPO
- 13:52
- Unidentified speaker (speaker_1): and at what and
- 13:54
- Unidentified speaker (UNKNOWN): uh
- 13:55
- Unidentified speaker (speaker_1): will that price be something where you you will be celebrating that is such a good price
- 14:01
- Unidentified speaker (UNKNOWN): that
- 14:03
- Unidentified speaker (speaker_1): the future will uh look i know it's a hard question but you're the futurist here Brett you can't oh hi for a minute so i
- 14:10
- Unidentified speaker (UNKNOWN): i
- 14:10
- Unidentified speaker (speaker_2): suspect give me
- 14:11
- Unidentified speaker (UNKNOWN): you i
- 14:11
- Unidentified speaker (speaker_2): suspect
- 14:12
- Unidentified speaker (speaker_1): yeah
- 14:12
- Unidentified speaker (speaker_2): yeah i suspect that two years from now people will look back and be like i can't believe you could have it for that price that's that's my strong suspicion and like i said i think it's because you don't the idea that their revenue is going to drastically slow down is like a hero call.
- 14:32
- Unidentified speaker (speaker_2): It is not at all suggested by their present or retrospective financials, but it's priced as if that's the call that that everybody is making. I think it's it's not the fact that being skeptical of superintelligent super intelligence is
- 14:52
- Unidentified speaker (speaker_2): also like media catnip that's a way to like be put on media does not help the kind of information sphere. But um there are there are other times where companies have come to market and people have been like, well, we've never really seen this before. And so they've been structurally underpriced. Google, when it came to market, they did a weird IPO
- 15:12
- Unidentified speaker (speaker_2): structure, and you know, you people just got it for cheap. cheaper than they should have. Um, here you have a company coming through a pretty politically choppy environment for its sector, um, with also, you know, uh, you know, NVIDIA CEO having just resigned. Um, so it's not the most market receptive
- 15:32
- Unidentified speaker (speaker_2): environment for the stuff they're doing. But if you look at the results they're delivering to enterprises that are using their products, including ours.
- 15:39
- Unidentified speaker (speaker_2): and the consequent financials that they're delivering, which are extraordinary, um, you know, I think people will look back and be like, it was weird that capital markets could not understand this company so
- 15:54
- Unidentified speaker (speaker_2): as much as we apparently aren't able to. And if I had to guess, it'll come, you know, you know, somewhere between 1.3 and 1.4 1.5 trillion in value uh when it comes to market. But the bankers, you know, they kind of figure out where the market is. And that's what if you look at prediction markets, that's roughly the range where it's it's
- 16:14
- Unidentified speaker (speaker_2): expected to come.
- 16:16
- Unidentified speaker (speaker_1): I I I've got to say I'm I'm staggered at the the pace at which anthropic um delivers new products. But the one thing I think I'm personally stagnant about is the popularity of Claude.
- 16:34
- Unidentified speaker (speaker_1): Has anyone done an objective assessment of Claude's acceptance and its market reach compared to its rivals?
- 16:45
- Unidentified speaker (speaker_2): Well, I mean, what we've done is we've looked at like Claude Claud is uh certainly among enterprise. users extraordinarily valuable and popular on the consumer-facing side. Like if you look at people using chatbots, actually um chat GPT is far and away the leader. So 70% of engagement hours across all of the
- 17:04
- Unidentified speaker (speaker_2): chatbots goes to ChatGPT. And um chat GPT has is on our estimates as more than a billion weekly active users who only use that AI system. They don't use any of the others. Their only exposure is to chat GPT, which is kind of really fascinating. And it speaks to kind of like
- 17:24
- Unidentified speaker (speaker_2): the OpenAI's positioning, particularly against that consumer market. Most of those people aren't paying for it. And so I don't know, you know, they're they're treating it as a, you know, a a literally a chat chatbot bot rather than something doing a lot of productive work for them is my suspicion. Um and the pace of um
- 17:44
- Unidentified speaker (speaker_2): improvements in AI means that very soon that that same chatbot will become much more capable for them, even if they're just relying on that advertising supported tier. Um for Claude and Codex, which is you know, Claude Cloud co-work and codex, Clodex Codex is um open OpenAI's AI is more. more agentic kind of
- 18:04
- Unidentified speaker (speaker_2): enterprise-facing and developer-facing system. Um, we think collectively across the two, there are, you know, less than 100 million. We know Codex has 40 million, we don't know precisely for anthropic, but on the same order of users. And the way to think about that is if say it's 80 million across the two using it today, um, there are roughly a billion
- 18:24
- Unidentified speaker (speaker_2): knowledge workers in the world. So it suggests. that despite the fact that, like you say, it's very popular and these companies are delivering all-time record, you know, revenues in a very short time, they are still, you know, I we think actually quite a bit less than 10% penetrated in terms of knowledge work users. Um, and in terms of
- 18:44
- Unidentified speaker (speaker_2): the utilization of their Of those service services amongst those knowledge work users, even less than that. If you look looked at the median employee. paying for AI in the US, uh, you know, a business is paying on the order of you know $10 to $15 per month per employee, but the top one percent of firms are paying, you know, seven to eight thousand dollars per month. And so it's like
- 19:04
- Unidentified speaker (speaker_2): as you can
- 19:06
- Unidentified speaker (speaker_2): learn how to use these things productively, actually you can spend a lot more efficiently, so far as we can tell. And so um kind of the you know, people often ask, like, what inning are we in, or how how far through this game are we? And on on a chat bot to internet to smartphone user penetration
- 19:25
- Unidentified speaker (speaker_2): basis, we're roughly where we were in January of 1997 in internet penetration of PCs. So that was, you know, Yahoo was uh just venture funded, you know, Google hadn't been founded yet. uh pretty early, you know. Um, and but on a knowledge worker to um our our uh
- 19:46
- Unidentified speaker (speaker_2): agentic use user versus total knowledge workers basis, we believe we're on the order of four percent penetrated. And that's more like where we were in nineteen ninety, nineteen ninety one. The browser was just being invented. And that that actually feels more like where we are in the cycle. We don't even know exactly what the business dynamics are going to look like.
- 20:06
- Unidentified speaker (speaker_2): And I think that there's, you know, there's it's it's it's quite rare, but it happens occasionally that a business just it becomes profitable by accident. And it grows so quickly that it grows into profitability before really kind of like, you know, modern call it venture theory would suggest it should. So that happened to Google, you know, Facebook was
- 20:26
- Unidentified speaker (speaker_2): like that. Uh and anthropic uh sprinted two quarters of unintentional, you know, adjusted profitability if you exclude their converts, which um if their value rises a lot, it appears as a negative on their income statement, which you know is an accounting trick. Um so I I think that you know
- 20:45
- Unidentified speaker (speaker_2): that's that the that kind of uptake of a service that then leads to kind of this cash flow generation is rare and it's a sign of how um large a market they're penetrating relative to their size.
- 21:00
- Unidentified speaker (speaker_0): Okay, Brett, last question. What do we need to know about anthropic, which you have not discussed with made me so
- 21:09
- Unidentified speaker (UNKNOWN): far?
- 21:13
- Unidentified speaker (UNKNOWN): I
- 21:13
- Unidentified speaker (speaker_2): mean, the way we think about the market is as Enthropic and OpenAI invest in training, it allows them to serve more users or um with less gigawatts of compute. And so you know one of the remarkable things about these companies is, you know, not only are
- 21:33
- Unidentified speaker (speaker_2): they already in market with these products which are screaming off the shelves that they're delivering very profitably. but so long as they continue to improve the products, it suggests that their gross margins are not going to decline, but going to increase, depending on the underlying cost of compute. But they in terms of like their ability to
- 21:53
- Unidentified speaker (speaker_2): deliver kind of revenue per gigawatt, we think is going to continue to increase, which, if they can hold compute costs flat, would lead to more growth. gross profit expansion. And Anthropic's already at 85% gross profit. So um I I do think it's um this is a unique set of companies coming to
- 22:13
- Unidentified speaker (speaker_2): market in business history between SpaceX, Anthropic, and probably OpenAI next year. I, you know, and I think there'll be a raft of companies that are built on top of and alongside them that will be very valuable. I think we'll point back to these three as being, well, this.
- 22:27
- Unidentified speaker (speaker_2): really showed us we were entering the singularity. Right.
- 22:31
- Unidentified speaker (UNKNOWN): Great.
- 22:31
- Unidentified speaker (speaker_1): Thanks for joining
- 22:32
- Unidentified speaker (UNKNOWN): us. My
- 22:34
- Unidentified speaker (speaker_2): pleasure.