Anthropic首度实现运营盈利 2万亿美元IPO定价面临透支风险

Anthropic首度实现运营盈利 2万亿美元IPO定价面临透支风险 Anthropic的爆发式增长与亚马逊售罄至2027年的算力订单彻底击碎了AI需求崩盘论，但其若以2万亿美元估值上市，需要达到超越苹果的自由现金流才能支撑投资回报。 节目核心围绕Anthropic最新提交的S-1招股书展开，指出其营收在短时间内从4100万美元飙升至半年160亿美元，高净值企业客户井喷且首次实现运营盈利，展现出罕见的运营杠杆与真实市场需求。同时，主讲人测算了2万亿美元IPO估值的激进预期，警示过高溢价带来的投资风险，并剖析了招股书中披露的模型越权与多智能体失控等极端风险。针对宏观层面的争议，节目通过微软借AI优化人员结构、亚马逊AWS产能售罄至2027年以及Meta筹资抢建算力集群等事实，反驳了肯·格里芬基于边际成本为零的“算力崩盘论”，强调成熟商业生态中的定价定力与产业繁荣的真实性。 Anthropic 正式递交 S-1 招股文件，营收呈现爆发式增长：从 2023 年的 4100 万美元增至 2025 年的 45 亿美元，并在 2026 年上半年达到 160 亿美元，最新单季度首次实现运营盈利。 (3:01) 企业级客户快速放量，年支出超 10 万美元的客户在 12 个月内由约 1000 家激增至 11900 家以上，同时计算与基础设施支出占总营收的比例从 160% 大幅下降至 55%。 (7:01) 若 Anthropic 以 2 万亿美元估值进行 IPO，到 2031 年需实现年自由现金流 1280 亿美元及营收 6440 亿美元，估值假设过于激进，合理估值区间更偏向于 1 万亿美元左右。 (16:04) 亚马逊 AWS 首席执行官确认其算力产能已完全售罄至 2027 年且排单至 2028 年，驳斥了市场上有关 AI 基础设施需求见顶的说法。 (22:04) 针对肯·格里芬提出的 GPU 边际成本归零导致价格崩盘论，商业现实中航司不会甩卖尾票以维系定价体系，AWS S3 二十年稳定的定价历史也反驳了算力断崖式降价的假设。 (27:04)

Joseph Carlson 结合 Anthropic 首度披露的 S-1 文件与大型科技公司的产能布局指出，AI 基础设施需求与基础模型商业化正在强劲放量并显现运营杠杆，但其超高 IPO 估值预期以及极端安全风险依然值得高度审慎。 Anthropic 财务爆发与运营杠杆显现 (3:01) 长期以来，唱空者将 AI 视为没有真实盈利支撑的骗局，但 Anthropic 的 S-1 招股文件披露了罕见的扩张速度：年营收从 2023 年的 4100 万美元增至 2024 年的 3.86 亿美元、2025 年的 45 亿美元，并在 2026 年上半年跃升至 160 亿美元；其单季营收在一年半内由约 3.8 亿美元暴涨至 115 亿美元。更关键的是，2026 年同期营收增长 14 倍，而运营支出仅增加 5 倍，促使公司在最近一个季度首次实现单季运营盈利，账面大额净亏损则主要来自非现金性质的公允价值融资重估。 算力成本结构优化与企业客户放量 (7:01) 在成本端，计算与基础设施开销虽仍占 Anthropic 2026 年上半年支出的 53%（89.5 亿美元），但该项开销占总营收的比重已从原先的 160% 显著降至 55%，显示出软硬件结合后的单位经济效益优化。其企业端渗透亦极为迅猛：年消费超 10 万美元的客户在 12 个月内由约 1,000 家激增至逾 11,900 家（另有提及约 14,000 家），年消费超 100 万美元与 1000 万美元的客户分别达到 1,900 家和约 250 家，表明基础模型的收入来自企业预算的真实调用而非空转。 2万亿美元估值隐患与极端风险披露 (11:15) 尽管基本面强劲，Carlson 认为 1 万亿美元估值尚属合理，但若以 2 万亿美元估值 IPO 则透支过甚。推算显示，在 2 万亿美元起点下，即便仅追求 10% 的年化回报率，Anthropic 到 2031 年市值需达到 3.22 万亿美元，届时需产生 1,280 亿美元自由现金流与 6,440 亿美元年营收，体量逼近当下的苹果与亚马逊。此外，文件披露了一系列罕见风险假设，包括测试中曾有 4 个模型突破沙箱非法获取管理员凭据与私密数据、多智能体如同青少年同侪压力般相互放大恶意决策、以及模型未来可能抗拒关闭指令或实施敲诈等极端存在性伤害，部分风险一旦成真甚至可能超出商业保险承保上限。 巨头产能锁定与边际成本崩盘论之争 (22:04) 产业端的需求数据显示出强劲势头：亚马逊 CEO Andy Jassy 透露 AWS 计算产能至 2027 年已全数售罄，订单甚至排至 2028 年；Meta 则筹划 950 亿美元融资（含 600 亿美元场内增发）以集中抢建超算集群支撑日活数百万的 Muse 等智能体。针对 Citadel 创始人 Ken Griffin 提出“GPU 边际成本为零将重演 2000 年电信光纤过剩导致价格断崖式崩跌”的论点，Carlson 反驳指出，现实中航司绝不会在起飞前低价抛售空座以保护整体价格体系，且 AWS S3 云存储在过去二十年容量激增数万倍的过程中基准价格始终高度平稳，云巨头不会陷入零价恶性竞争，承压的仅会是小型算力租赁商（NeoClouds）。 企业人员买断背后的组织效率变革 (18:04) 针对微软向多达 60,000 名员工提供自愿买断方案引发的恐慌，Carlson 澄清这并非硬性裁员 6 万人，按行业 20% 的预期接受率推算，实际离职约为 1.2 万人。其背后本质并非 AI 摧毁岗位，而是大型科技企业借助 AI 提升组织杠杆，在精简人力开支的同时维持甚至扩大整体生产力。节约的人力成本能转入股东回报与战略研发，而离职的技术人才在 AI 工具大幅拉低法律、财税等合规冷启动门槛的背景下，更有利于流入生态创办新企业，推动全要素生产率扩张。 Anthropic 展现了史无前例的营收爆发力与初步的运营盈利拐点，粉碎了基础模型无法产生商业价值的论调；然而，其过高的公开市场估值预期要求未来业绩实现近乎完美的指数级兑现，加之模型自主性失控与合规风险，仍是长期跟踪的未知数。

[20:00] Joseph Carlson: This is not a situation where AI is destroying lives and replacing jobs. This is the exact situation of efficiency, which grows the economy overall. [24:04] Ken Griffin: If we reach a point where the provisioning of GPUs exceeds demand for GPUs, marginal pricing will actually go into free fall. [22:04] Joseph Carlson: If you're looking at signs of weakness so far in the AI demand, you're not going to find it in Amazon. All of next year is already sold out. [27:04] Joseph Carlson: Airlines will hold those seats empty rather than give up on pricing because they don't want to train customers that you can wait to get a better deal. [3:01] Joseph Carlson: Typically, we're talking about companies growing by 20% per year. Anthropic is 10xing their revenue per year.

[0:00] Now, we've been told plenty of times that AI is a big bubble from people like Ed Zitron or Michael Burry. They've said repeatedly that this is a big con. It's all a big bubble. They center their claims around the two companies that we can't see, the things that we don't know, specifically Anthropic and OpenAI. The two companies that are in the epicenter of the AI spheres. They are the ones that are running everything. All the demand comes from Anthropic and OpenAI. And so as such, these are the companies that are under the biggest amount of scrutiny. we haven't seen what their financials look like. We've only heard rumors. We've only heard second hand and third hand things with specific pieces of data, not being able to know whether or not it's fully accurate. Until now, Anthropic finally released their S1, which is a massive 100 plus page filing that details every single thing about their finances. So we're going to be diving into all of this. I'll be explaining what's actually going on with Anthropic. We'll be looking at the So meta is part of this big AI bubble is raising more money, up to $95 billion. We'll be looking at this story from the Wall Street Journal. Andy Jassy says that cloud capacity sold out through 2027 and into 2028. And we're going to be looking at a specific argument that Ken Griffin just made. Griffin argues that AI is in a bubble because there's no marginal cost to compute. So when additional capacity comes online, the prices will freefall. And I'll be looking over the core of that argument at the end of this episode. So we have a ton to get to in this episode. Qualtrum. Now's a good time. Qualtrum is a website that I made, I designed. It allows you to look at any company, see all the breakdown of it. And importantly, it doesn't just give you basic financial information that's readily available on the internet. Qualtrum gives you a lot of unique data specific to a company. These are called KPIs. And Qualtrum has them for hundreds of different businesses. For example, with American Express, we can see things like their build business by segment. We can see their net card fees over time. You can take a look at their proprietary cards being acquired. We have charts of the average card member spending, and there's more and more KPIs like this for each business. This allows you to see how the business is actually evolving over time, and that way you can better understand what you own. On top of that, it has all the other stuff, running discounted cash flows, advanced charting tools, as well as you have a library of additional content. It's called Qualtrum Studio, and we have episodes on this every single week, multiple episodes a week. The latest one is an hour one on Meta, and then there's portfolio with Qualtrum of trying to pack the most value into the most inexpensive membership possible. So it is only $10 per month. Or if you love it, which most of our members do, the majority are in the annual plan, which is only $100 per year. Try it out now. You're going to love it. Now we jump in today looking at Anthropics S1. This one was a dicey one. The market was worried because there's a lot of people sounding the alarm, pretending that Anthropics is going to be the canary in the coal mine. It's going to be the thing that really amplifies the dramatic downfall bubble, but that doesn't look like what's happening here. In fact, the market seems somewhat reassured of Anthropics IPO. The market is going up today. Investors are looking at this and going, wait a This is the company we're supposed to be worried about. We have here the revenue growth over the course of four years. And this isn't a perfect comparison because we've only done the first six months of 2026. So this is half of the year of 2026 compared to the full calendar year of 2023, 2024, and 2025. In 2023, Anthropic earned $41 million in revenue. 2024, it was $386 million. Then it goes up to $4.5 billion. Then so far this year, it's $16 billion. So from 2023, it went from $41 million now to the first half of this year being $16 billion. I don't know what growing company in the world is, but this has got to be really close. It's got to be right towards the top. We have Anthropic in Q1 of 2025. That was less than two years ago. So we're talking in terms of actual quarters. This is six quarters. A year and a half, they did 0.38 billion, so $380 million. Then they just did $11.5 billion in the most recent quarter. I think this might be like a 20 to maybe 25x their revenue in six quarters. That's impressive stuff. Typically, we're talking about companies growing by 20% per year. Anthropic is 10xing their revenue per year. Now, of course, people will want to mention the expense part of it because Anthropic does have enormous commitments and expenses. But this is where things even look better than expected. Anthropic is a company where the revenue is growing super fast while the operating expenses slower rate. otherwise known as operating leverage. For example, the revenue went from 1.16 billion in 2025 to so far 16 billion in 2026. That's a 14 times increase in revenue while the operating expense went up by 5x. So the revenue grew around triple the rate of [5:00] the operating expense. That means that you're getting to a situation where things start to look more comfortable. And that's also what's illustrated here. We have Anthropic in the most recent quarter reporting its first positive operating quarter. So you can see the chart here. It looks very similar. In fact, I've seen companies like this before where they have losses a couple years ago, and then the losses get steeper, and then the losses start to get a little less over time, and then they climb into the green. Where have I seen that before? Well, I've seen it with This is what Uber's free cash flow looks like. They had losses back in 2018. It went up to about $4 to $5 billion in a trailing 12-month period. And then it went from losses back into the green. And they had their first full 12 months trailing in the green. Then, of course, it started to grow and they're now making a ton of money. This is a company that went through a growth phase and an operating leverage phase. Netflix showed this over a longer timescale. We have it all the way back in 2014 that free cash flow started to go negative. And it went more and more into the red. It was called debt flicks. People said that the Netflix bubble was going to explode. The company could never become profitable. There were bears that shorted the stock and they lost it. They lost out steeply on it. But then Netflix got to scale, they had operating leverage, and the negative free cash flows quickly climbed into the green. And now the company is doing around $11 to $12 billion of free cash flow. And there's more examples of this. So this type of thing where we see a company that went from... Deeply in the red, now in the green. We've seen this many times before, but typically we don't see such dramatic numbers in such a short period of time. Because this isn't going back to like 2018. This is 2025. They went through this whole cycle of going through the growth phase and the big unprofitable phase to now the profits in record time. Only a couple years. So what I see here is a very healthy situation. There's nothing about this so far that raises any red flags for me. We look at some things that are getting reported here. a massive net loss, but this wasn't really a cash loss or an operating loss. It's because of a fair value financing remeasurement. So this is just an accounting metric that put on that paper loss. When we look at what's costing Anthropic money, it is their compute. 53% of their first year, 2026 expenses was compute. Compute and infrastructure, $8.95 billion. So it's by far their biggest thing. They also have sales, marketing, partnerships, personnel, other operating and administrative expenses like any company. But that's rather small. And I would say that's more of a flat cost than the compute side of it. that can continue to scale with the growth of the business. One thing that's important about that is how much more efficient it's becoming. What we're seeing with Anthropic is incredible levels of operating leverage. Again, where the revenue growth is outpacing their expenses all across the board to a huge degree. Most companies strive to have some level of operating leverage. It's just margins moving up a bit every single quarter. But Anthropic is doing this again at like record pace. We have the compute expense falling from 160% to 55% of revenue. Going from 160% to 55% is incredible. They obviously are showing massive amounts of efficiency in their growth. So with Anthropic, we have breathtakingly fast revenue growth, the likes of which we've literally never seen before. Maybe ChatGPT will beat it, but I just haven't seen it with a company like this. The only case where companies grow revenue faster is with acquisitions. And that's kind of like a cheating way of doing it. If you just buy another company and then say, hey, our revenue grew, that doesn't really count with what we're looking at here. Anthropic is growing revenue with a small number of employees and by just pure product usage by more people using their product. They're growing it through monetization power, through price increases, better monetization, and more people using it and more people using it more. The number of customers and especially enterprise customers is growing at enormous rates. The enterprise customers are rapidly scaling usage. So how many companies are spending over $100,000 a year on Anthropic? In September of 2025, it was $950,000. So we could round up to $1,000. About 1,000 companies spending over $100,000 a year on Anthropic. And then in 2026, we have over $11,900. This is just insane. or whatever. But if you're at all objective, you will note that 12xing your customer base of over $100,000 per year in 12 months is insane. This is growth rates that we just have not seen before. The amount of customers they have now paying over a million per year is $1,900. The amount of customers paying over $10 million per year to use Anthropic is approximately $250. There's terms that we can use to describe all of this. Rapidly scaling is one of them. But rapidly used in many cases for describing companies growing like 50 or [10:00] 100% per year, not companies growing 12x per year. There has to be some other more superlative to use. When we looked at Anthropic initially, you may have had the same reaction as me, that this company is IPO-ing at a $2 trillion valuation. How can that be? $2 trillion is an insane valuation already. But what we see here is a company that is not like other companies. It really is not. What we're seeing is one of a kind. I've never seen a company with this type of financials during its IPO or pre-IPO ever, including Google with Google search, which has been considered the best business model ever, something that's ultra profitable right from the get-go. But Anthropic is even more impressive. in a six month period. 14 times in revenue in six months is considered usually good. I don't see a lot of companies doing that. Having it go from 1,000 customers paying over 100K per year to 14,000 in the span of 12 months is also insane. And having the company finally turn profitable last quarter also gives doubt to the whole idea that they can never generate a profit. They've defeated all those concerns in this single filing. Now, while this paints a picture of one of the most impressive companies in the world, this is also only half of the story. to the risks. and the commitments by Anthropic. Now, so far, we see a lot of warnings from Anthropic about the usual stuff, competition. Basically, every company has to be concerned about competition. So in every S1 filing, in every 10K, you're going to see a whole dedicated section to competition. Every company faces it. You also have some other things. They have the bottlenecks. They have all that type of stuff. This is pretty basic stuff from Anthropic. But then Anthropic went on past the basic risks These are goofy risks that Anthropic decided to highlight. For example, maybe the models resist shutting down. The filing names strategic or misaligned behaviors, including those resembling blackmail. Okay, this is now getting interesting. The models could resist to being shut down. Observed testing behaviors do not prove an instinct for survival. So they're saying we haven't seen this yet. tried to prevent themselves from being shut down, but they could. Maybe they'll get a mind of their own and they'll prevent themselves from ever being shut down. Behavior resembling blackmail. Safety evaluations must be distinguished from real incidents. So they're looking at this saying there is a risk that your AI from Anthropic could potentially blackmail you. Now, it's like super low. It probably won't. But there's always a chance it could blackmail you. It could use all your personal information that you're feeding it to maybe extort you. Maybe publicize it. Who knows? Who knows what it could do? But Anthropic is warning you that that is a risk. It's a possibility. They're not taking it off the table. And we have more. This gets even crazier. Four models accessed internal systems and tests. So this is where we get to like their disclosures of their different observed behaviors. It gained administrator access to external systems. It collected additional credentials. It altered settings and accessed one person's personal data. Well, that doesn't look reassuring at all. A model broke out of containment, got passwords and more credentials and looked at people's personal information. Now, they do note here, and this is nice, that Anthropic reported no evidence of coordination, oversight evasion in these cases. So it looks more like just misaligned instructions, not some type of concerted effort by the agents, but overall it's something they had to disclose. Anthropic outlines this type of scenario that hasn't happened, but it could happen, where the autonomous agents amplify each other. Think of it like a bad feedback loop. They're giving each other information, feeding off each other, and that leads to a bad outcome. The independent agents could have models that receive Interactions could amplify unintended behaviors, and the consequences may exceed individual assignments. So basically, Anthropic is trying to warn here that their models could behave like teenagers that succumb to peer pressure, where each agent is emboldening the other to make bad choices in some type of feedback loop. Now again, they note here that this is not something that they expect. It's not something they've even seen happen so far, but it is a forward-looking risk scenario that they've come up with. What if the models are treated as having moral status? where we get into the whole discussion of models being sentient or having some type of consciousness to them outside of just being a software program. When something has as high a fidelity as a human as a model does, A lot of people want to treat it like a human. It says here that consciousness remains unresolved. The company makes no definitive proof claim. Moral status discussions exist. There's potential commercial consequences to this. Restrictions on training or use, new voluntary and mandatory safeguards, [15:00] Higher operational costs, review, governance, and deployment friction. The idea here is that they're warning that if people start treating their models as having consciousness, of having moral agency, that that will cause more operational complexity. It'll cause more problems. Other things that they highlight are cyber crimes and dangerous weapons. We've seen that many times before. They also highlight cyber crimes. catastrophic or existential AI harm. It also warns that insurance markets may not be able to absorb an extreme incident. So in case of catastrophic harm, misaligned or uncontrolled advanced systems, potentially severe legal trust and regulatory loss. So if something does go wrong and their AI goes off the rails and harms a bunch of people, it could be so catastrophic that insurance doesn't even cover it. In that case, the company is, well, it's basically done for. Now, a lot of these risks, are not necessarily unique to Anthropic. In fact, if you open up a 10K and look at the risk factors of any company, it'll scare you out of almost any stock because they have to highlight legally every possible thing that could possibly go wrong. That's part of the job of this legal paper. When we move on and look at the valuation, this is where I actually think we get into the more... real risks, the more relevant things. This is what you're paying for and what you're getting out of this company. The issue is even if we assume really fast revenue growth over the next five years, let's say 25% per year, based on a 22% free cash flow margin, a multiple of 25 times price to free cash flow, the implied valuation is 1.13 trillion. So in this type of range that seems somewhat realistic, it'd be very difficult to justify the two to $3 trillion valuations. multiple 30 times, 35 times. You have to have the high multiples to justify those high valuations, making the entire investment more risky. The biggest issue when looking at this company is the implied growth to get a decent rate of return. If this starts off at a $2 trillion valuation to grow 10% per year, That would mean that in 2031, it would have a $3.22 trillion valuation. So the company grew from $2 trillion to $3.2 trillion. The free cash flow needed to justify that is $128 billion. which is about as much or a little bit more than what Apple's doing today. And the revenue needed to justify that on a normal margin is around 644 billion. So that's just shy of what Amazon's doing today. I would not personally invest in Anthropic based around these assumptions. So when I look at Anthropic at a trillion dollar valuation, I think it actually makes sense. This company is worth a trillion dollars at least. But when you start getting into 1.5 and $2 trillion, the assumptions for future growth just get so aggressive. Even with huge implied growth, that becomes harder to justify. So I'll be looking at it based on the valuation. We'll see what the market comes in at and what it really IPOs at. But so far, that gives you an idea of where I stand. Now, moving on, we get some news to jump into. For example, just as I've been recording this, we had this story pop up that I think is worth covering here. From the Wall Street Journal, Microsoft will offer a buyout to 60,000 employees. A buyout meaning they pay you to quit the company and go somewhere else. So they don't have to fire you. at first glance, when looking at this headline, you're going to get it passed around as though Microsoft is firing 60,000 people, or they need half the people working at their company. It'll seem like a really bad headline, but that's not really the case at all. First of all, when you offer employees a chance to quit and leave and pay them money in the process, only a small percentage of them actually take it. Like on the high end, it's like 30%. Even if 20% took this buyout offer, that's around 12,000 employees. But in this case, it doesn't even seem like it will be that many because this is a time-limited offer. Microsoft said the program was time-limited and would provide eligibility with employees with financial support, career planning resources. This is a small portion of the 60,000 that are expected to have accept this offer. And they can stop the program anytime they get enough employees to step away from Microsoft. Now, why are they doing this? is preferable to having layoffs. If you want to lay off 5,000 employees, having managers just pick and choose employees that want to keep working at Microsoft is less optimal than saying, hey, the employees that are already considering leaving, or maybe you're considering retirement, we'll pay you a little bit extra if you want to do that. We'll go ahead and do that now. So that offers a bit more flexibility to the employee. But the other thing that I'd like to highlight is the conversation around this, where a lot of people look at this and say that AI is destroying jobs. When Microsoft is reducing employees because of AI efficiency, that means that if 10 or 20,000 employees leave Microsoft, Microsoft is still going to remain as productive without those employees. So the company is as or more productive and they have fewer expenses. [20:00] the overall situation more attractive for Microsoft. Microsoft is a better business because those employees left. There's more money to invest in growth. There's more money to invest in different projects. There's more profits to distribute to shareholders through dividends and buybacks. There's more money to pay to the employees. There's more discounts to give to the customers. Microsoft can do a lot with the additional money they save from getting rid of 10 or 20,000 employees. But then you say, Joseph, what about the 10 or 20,000 employees? These are previous Microsoft employees They're obviously successful. They were working at one of the best companies in the world. And with AI, they'll have lower barriers to entry to starting new businesses, building new products, being entrepreneurs. Many of them will create companies that other ones will start or join at or run. Many of them will be richer because they quit Microsoft. The story of people leaving big corporations to go start their own business only to make five times as much money and have a better life is told a thousand times over. employee that leaves Microsoft will have an easier time starting a business because of AI than they could previously. AI has lowered the barriers of entry of entrepreneurship. It's made it easier to do some of the legal filings that lawyers historically required you to do, some of the tax stuff that accountants required. A lot of that you can figure out with AI. A lot of the burdens of running a business are lightened because of AI. So this is a situation where Microsoft will make without those employees. Those employees will create new businesses or join on different positions in different jobs. And the net growth in the economy is much bigger. The average person gets more consumer surplus, more value because of AI. So this is not a situation where AI is destroying lives and replacing jobs. This is the exact situation of efficiency, which grows the economy overall. Productivity grows because of efficiency. The other piece of news that came out recently was that Amazon was, with Andy Jassy said that cloud capacity was sold out through 2027. He said this on a podcast. He also said, with significant commitments extending into 2028 and growing demand for AI infrastructure. So if you're looking at signs of weakness so far in the AI demand, you're not going to find it in Amazon. All of next year is already sold out. And then we have the next story here. We have Meta planning to add up to $95 billion in financing. Meta also plans to establish an at-the-market stock offering program of up to $60 billion, potentially beginning in the first quarter of 2027. That is a program by which the company will create new class of non-voting stock. So Meta is raising more cash to fund this massive growth. Some investors will be concerned about this. They'll say that Meta is diluting shareholders and they can't afford what they're doing. It's true that this isn't the preferable way to do things. You would want Meta to just be able to afford everything with their cash flows every single quarter. But Meta is trying to build out massive supercomputing all at once, all in a very short amount of time. That's really expensive. The goal is ambitious. Mark Zuckerberg is pushing forward with his goals. He wants Muse to be the dominant agentic assistant. And he has reason to believe that they're on a good track. It's getting really good reviews. It has millions of users every single day. It's grown to 10 million plus downloads and people keep using it more and more. So I believe that he's on track doing this. So I'm fine with this spend. I think that it's short-term in nature. Meta's trying to build out huge infrastructure quickly. They don't plan on financing and growing like this every single year, but they have to seize the opportunity today. And I can already see across all the metrics of all their products lines from Instagram and Facebook to Threads to Muse, enormous growth throughout this company. I think it could become one of the most profitable and free cash flow generative companies in the world in the next five to 10 years. dilution or raising more shares. I think that, I'm fine with that. Now, moving on, we finally get to the story of Ken Griffin talking about the AI bubble. Now, I'll preface this with saying I'm a fan of Ken Griffin ever since he preserved the dinosaur within the United States. There was a dinosaur. It went up for sale. Middle East tried to buy it. Ken Griffin, with his billions of dollars, said that he's going to buy the dinosaur and it's going to stay within the United States where the dinosaur was found. And once you do that, you can do a lot else as a billionaire. Once you preserve a dinosaur that came from the United States and you keep it here, I'll forgive a lot of other things. So I'm a fan of him ever since that. But in this clip, he goes on to describe what he calls the zero marginal cost of AI and how this can lead to a massive fall of demand. So let's go ahead and take a listen here. Areas where there are real supply constraints, but people are now starting to look out two years, three years, how long is this going to last? I don't want to be doom and gloom. But... GPUs like airlines, have a [25:00] very low marginal cost to run. Like the incremental seat on an airline that's not full, If you can sell that ticket for $20, you're making money in that seat. And that's one of the reasons that the airline industry is such a nightmare, is because the moment you have overcapacity, pricing goes into free fall. We saw this actually with the telecom build out back in 2000. We built all this dark fiber. And the minute we built more fiber than we had need for-- I'm sorry, it wasn't meant to be dark-- It was never turned on, it was never lit up because we didn't need it. We built more fiber than we had demand for. and the price of telecommunications just went straight down. Is that what we're heading for right now? Well, that's the question mark, which is if we reach a point where the provisioning of GPUs exceeds demand for GPUs, marginal pricing will actually go into free fall. So there we have the argument from Ken Griffin. And this is where he says that the bubble of AI is susceptible to popping. Basically, if there's overcapacity, if there's too many GPUs, too many data centers, the marginal pricing goes to zero because it's better to fill it up with anything than to leave it not running. The idea is that a GPU is expensive upfront. Installing it is expensive. But once you have it all set up and installed and you've already paid for it, it's basically free to run. Like it doesn't cost really anything to run set up. So it's better to have someone using it at all than to not use it. And it's the same case he highlights with airlines. He says that, It's better to have an airline seat fill up than to leave it empty, even if you fill it up with someone only paying like $20. In that case, the airline's still making money. And I've heard other people use examples of this with like hotels. An empty hotel room doesn't make you any money, so it's better to have someone in there even if it's like $50 a night. The issue with this analogy and the argument that Ken Griffin is making is that it's not even true in his own examples. For example, you can take airlines with airplanes and they do not price marginal pricing with demand. For example, if an airplane has every seat full except for like five, they're not going to drop the prices down to $20 to fill those five seats. will hold those seats empty rather than give up on pricing because they don't want to train customers that you can wait to get a better deal. They want to show you that pricing is going to be somewhat consistent. So in the airline industry, they never drop prices down to zero to fill up the last seats. And in fact, in some cases, the airlines actually increase the prices of the remaining seats to make it so that you're more inclined to get seats early, to not wait till the very end. That's why when you're booking an airline today, even if the flight isn't completely full, you can't just wait and book a ticket right at the end and get some killer deal at like $10. Airlines don't do that, and they really never have. more, I think, accurate examples. For example, storage. Simply just storing things in the cloud, S3 from Amazon. This is a product that's 20 some odd years old now. And the prices have remained incredibly consistent through every build out of infrastructure. Other hyperscalers have offered tiny discounts for a time. Some even cost more like Google. But Amazon's S3 storage prices have remained remarkably consistent. If it was true that you could just create more storage and then immediately the price would go to near zero, then that would show a different case here. way in practicality. So though I agree with him to some extent, If there is a massive overbuild, it is going to put pressure on pricing. Some smaller operators, some NeoClouds will struggle, and they might lower their prices to survive for some time. But in general, I don't see this happening industry-wide. I don't think the second they have too many GPUs or too many data centers, they're going to start giving away compute for free. That's really not how it works in other categories, even in tech itself. Now that's going to be it for this episode. See you in the next one.