Free with a sign-in.
The best 8 min of 33
Steve Eisman: Here's What Anthropic Isn't Telling You
Steve Eisman says Anthropic’s S-1 leaves key growth and compute risks unclear; Dan Chiolan says Waymo leads the robotaxi race.
Prof G Markets · Ed Elson
- Steve Eisman · Investor
- Dan Chiolan · Research Associate, Prof G Media
The best minutes
- 2:15Q3 numbers reveal Anthropic’s growth slowed after token maxing3 min
- 5:42Anthropic’s commitments raise costs while price cuts test its moat5 min
Also worth watching
- 25:32Autonomous fleets create opportunities beyond vehicle makers3 min
- 20:59The U.S. leads AV commercialization while China leads manufacturing3 min
Skip the sponsor reads and opening market recap, the AV safety, rollout and regulation discussion, and the closing Anthropic recap.
Full analysis · 6 min read
Real Alpha · Nov 7
Steve Eisman says Anthropic’s Q3 growth fell to 45%
- Sequential revenue growth fell from 147% in Q2 to 45% in Q3, Eisman said.
- His estimate of $800 billion to $900 billion in commitments raises cost questions, but timing remains unclear.
- Investors should separate accounting losses from future spending and track growth after any price cuts.
Free with an account
Read the rest of the analysis
6 sections, the 7 other voices it tests Eisman against, and what it means for you. Every episode we cut gets one.
Check your inbox
We sent a link to your address. Click it to finish signing up. Not there in a minute? Look in spam.
or
Signed up before? The same buttons sign you in. The episode, its moments and the transcript stay free.
Eisman says Q3 growth slowed after token-maxing
Eisman’s most portable figure is sequential revenue growth, not first-half year-over-year growth. He said it fell from 147% in Q2 to 45% in Q3, after token-maxing eased.
His second measure weakened too. Revenue divided by compute and infrastructure expense fell from 2.08 times in Q2 to 1.7 times in Q3. The gap is 0.38 ratio points, our calculation of 2.08 minus 1.7, though revenue still exceeded that expense measure.
But this is a meaningful deceleration, not proof of shrinking sales or negative unit economics. Q3 still showed 45% sequential growth, with revenue above compute and infrastructure expense.
For example, Eisman’s Q4 baseline was 43% sequential growth and a 0.69 ratio. Q1 rose to 135% and 1.4 during token-maxing. That ratio was 0.71 higher, our calculation of 1.4 minus 0.69.
In 20VC in October 2026, LangChain CEO Harrison Chase said inference demand could support multiple providers. That broad market view does not show whether Anthropic’s own usage growth will endure.
Our read
Q3 is a pressure test after exceptional usage, not a verdict on long-term demand. Track sequential revenue beside compute expense, not growth alone.
The filing offers a stale operating snapshot
Eisman’s concern is disclosure timing. Many headline charts stopped at Q2. His Q3 comparison came from reports, not figures he could verify in the S-1.
That distinction matters because a filing can be accurate yet offer a stale snapshot. Elson raised whether the omission hid slower growth. Eisman answered emphatically that it was intentional.
Still, that exchange does not establish Anthropic’s reason for leaving Q3 out. Investors need later IPO materials with complete Q3 revenue, compute costs and any price changes.
Peter Diamandis’s Moonshots episode in October 2026 describes rapid revenue growth and improving compute economics, alongside heavy commitments. His $2.08 per compute dollar figure matches Eisman’s Q2 ratio, not a separate Q3 reading. That episode does not test Q3.
This comparison illustrates why dates matter. A strong Q2 ratio cannot establish that later growth or economics remain as strong.
Our read
A Q2 endpoint leaves investors with an incomplete view of the turn Eisman is describing. Later disclosure should be judged against the same operating measures.
Future commitments complicate today’s cost picture
Eisman estimated between $800 billion and $900 billion in future commitments, depending on what is included. He called the figure imprecise, not a current expense total or schedule.
Elson said the filing also points to substantially higher capital expenditure. Eisman would not assume Anthropic will own data centers, and said the accounting path remained unclear.
Eisman said he had read one report arguing Anthropic would spend $80 billion to $100 billion in CapEx by next year. He did not know whether that would be expensed or placed on the balance sheet. He gave no schedule.
Eisman expects these commitments eventually to appear in the income statement, through depreciation or another route. The episode gives no confirmed timing, so investors should not treat the full amount as next year’s expense.
Paul Kedrosky on Prof G Markets in October 2026 said most commitments are future, massive and non-cancellable. He also said Anthropic’s move toward its own infrastructure could alter costs, reinforcing timing concerns rather than supplying a schedule.
Ed Zitron on The Weekly Show with Jon Stewart in October 2026 gave a $430 billion figure for non-cancellable commitments. Eisman’s $800 billion to $900 billion estimate depends on scope, so the totals are not directly comparable.
Paul Kedrosky · Prof G Markets · Oct 27Anthropic’s future compute costs dwarf its current losses
Ed Zitron · The Weekly Show with Jon Stewart · Oct 28Why Zitron says AI labs’ compute commitments require huge growth
Our read
The commitments make reported profitability an incomplete guide, but they do not establish when or how much expense will hit. Investors need the schedule and accounting treatment.
A reported price cut would test pricing power
Eisman said he had heard, but could not confirm, that Anthropic cut prices by about 50% in October to regain share. He did not present it as a verified filing fact.
The test is volume against yield. If price per use falls, Anthropic must show whether added customers or usage offset lower prices and compute costs.
Take the reported cut as a hypothesis, not a fact. A price reduction alone cannot establish whether customers are switching easily or Anthropic is trading margin for adoption.
Steve Eisman on The Real Eisman Playbook in November 2026 said Chinese open-weight models cost one-tenth as much. That later view strengthens the price-competition question, but does not verify the reported October cut.
Nikesh Arora on All-In in October 2026 said agents with users’ context, APIs and connections can create switching costs. That is a broader software-stack possibility, not evidence Anthropic already has those protections.
Steve Eisman · The Real Eisman Playbook · Nov 2AI firms may use safety rules to block cheaper Chinese models
Nikesh Arora · All-In · Oct 21AI application stacks can make user context a moat
Our read
The important evidence is what lower prices do to revenue, share and cost per dollar earned. A discount is a test, not a verdict.
Founder control limits public shareholders’ influence
Eisman said public shareholders have no rights. Elson described several share classes, including one Class F share controlled by seven co-founders. It carries a majority of the vote.
That structure limits ordinary shareholders’ influence, but it is separate from whether customers want the product or compute spending earns returns. Control tells investors who decides, not whether the business works.
Eisman compared the criticism with Meta, noting that founder control can coexist with strong stock performance. But that comparison does not settle valuation or shareholder protections. It only cautions against treating weak voting power as proof of a weak business.
For a board weighing exposure, control rights and operating economics belong in separate columns. The episode identifies the voting structure and Eisman’s judgment. It does not provide a full account of investor protections or a final IPO price.
Our read
Governance is a distinct risk to assess, not a substitute for analyzing growth and costs. Investors should know what influence their shares actually carry.
Elson calls the model risky, not the product
Elson’s closing view is narrower than a verdict that Anthropic is a bad company. He attributed the $80 billion first-half net loss to fair-value accounting on convertible notes.
That correction matters because a reported fair-value loss is not, by itself, a measure of cash burn. But it does not erase future compute costs or establish positive economics once commitments affect results.
David Sacks on All-In in October 2026 called Anthropic exceptionally capital-efficient, pointing to low cash burn and rising gross margins. His reading contests the bearish interpretation of current losses, not whether future commitments alter margins.
Eisman told listeners to watch CoreWeave debt. He said borrowing at 15% could be prohibitive for its data-center model, while acknowledging he could not prove it. This is a financing signal for infrastructure suppliers, not a direct measure of Anthropic’s solvency.
Next, track sequential revenue after discounts, compute expense per revenue dollar, and when commitments affect costs. Elson said $2 trillion had been an expectation for the IPO, but price cannot replace operating evidence.
Our read
The episode supports underwriting uncertainty, not certainty of failure. Until revenue quality and future cost recognition are clearer, Anthropic is not a proven safe investment.
What it means for you
- Rebuild quarterly ratios. Compare sequential revenue growth with revenue per compute dollar for each quarter. Keep Q2 and Q3 measures separate from first-half year-over-year growth.
- Test price elasticity. Ask what customer growth, usage and revenue followed any price reduction. Measure whether lower prices improved share without worsening revenue per compute dollar.
- Separate loss from cash. Have finance distinguish the convertible-note fair-value loss from cash burn and operating costs. Then model future commitments on their own timing and accounting assumptions.
- Track commitment recognition. Request a schedule showing when compute commitments and higher capital expenditure could affect reported costs. Do not treat the full estimate as a single-year expense.
What to watch
- Whether IPO materials include Q3 revenue, compute costs and any confirmed price changes.
- How Anthropic’s revenue-to-compute ratio moves after any price reductions.
- CoreWeave debt costs, which Eisman said he would watch as a signal for data-center financing.
Send this to your finance lead and the team reviewing AI infrastructure exposure.
Watch the minute where Eisman gives the 147% to 45% comparison.
Written by Real Alpha from the episode's transcript and the other moments it links, then checked against the transcript. Quotes are the speaker's exact words.
Minute by minute
- 1:25Anthropic’s Q3 growth slowed
Eisman says sequential revenue growth rose from 43% in Q4 to 135% in Q1 and 147% in Q2, then fell to 45% in Q3 as token-maxing eased. Revenue divided by compute and infrastructure costs also fell from 2.08 to 1.7. He says reports of an October price cut of about 50% would be another key data point, though he could not confirm them.
- 5:42The filing leaves future costs unclear
Eisman calls the S-1 backward-looking because its main charts stop at Q2, before the slowdown he sees in Q3. He says future compute commitments of roughly $800–$900 billion mean expenses will rise, while increased capital spending may add costs that are not yet reflected in the income statement.
- 9:25Eisman says shareholders have no rights
Eisman says a reported 50% price cut would reinforce his view that Anthropic has no moat if it must rely on price to gain share. Elson outlines the share classes, including a single Class F share controlled by the seven cofounders that carries 50.1% of the vote. Eisman says shareholders have no rights and compares the structure to Meta.
- 11:35Investors should watch CoreWeave’s debt
Elson says Anthropic had been expected to go public at $2 trillion, while Eisman says he has no expectation on its IPO price. Eisman says Q3 results could change investor sentiment and identifies CoreWeave’s debt yield, which he puts at about 15%, as a key signal for its ability to finance data centers.
- 15:39Waymo leads the robotaxi field
Chiolan says about 40,000 Americans and over a million people globally die in car crashes each year, and 94% of crashes are due to human error. He says Waymo operates in 15 U.S. cities and has 500,000 paid rides weekly, compared with Tesla’s service in six cities. Zoox has about 100 vehicles in two cities, while Uber facilitates autonomous rides through partnerships in five cities.
- 19:31Tesla’s cost advantage is substantial
Chiolan says Tesla has moved quickly from supervised rides to fully autonomous commercial service in six cities, but Waymo still does about seven times as many weekly miles. He puts Tesla’s Cybercab production cost at about $25,000, versus upward of $100,000 for a Waymo vehicle, whose sensor load makes it more expensive and slower to build.
- 21:07The international race favors China
Chiolan says China has a manufacturing cost advantage, while the U.S. leads in commercial paid rides, with no Chinese company matching Waymo’s volume. He expects competition to play out in Europe and the Middle East, where Chinese firms have gained ground through their existing vehicle presence; U.S. firms have focused on their domestic rollout.
- 23:17Regulation and public opinion remain hurdles
Chiolan says U.S. regulation is fragmented across states, forcing companies to seek approval and lobby state by state; he also notes rising federal lobbying by Waymo. He cites a Gallup poll in which agreement that autonomous vehicles improve road safety fell by 12% from the 52% recorded in 2018, despite safety improvements.
- 25:21Look to sensors and support services
Chiolan sees potential in LiDAR suppliers because 95% of passenger-car LiDAR sensors are made in China and bills could ban them from civilian use. He also points to vehicle maintenance and servicing, including Hertz’s new business line, and to trucking, where driver compensation and benefits make up about 44% of marginal costs.
- 28:16Anthropic’s reported loss needs context
Elson says Anthropic’s $80 billion first-half net loss reflects the accounting treatment of convertible notes whose fair value increased, rather than operating performance alone. He notes first-half revenue growth of 1,300% year over year, but says Q3 growth slowed to less than 50% quarter over quarter after reaching about 150% in Q2.
- 30:46Future commitments add significant risk
Elson says Anthropic’s future compute commitments exceed $800 billion and are not yet reflected in the income statement. He cites contracts of $110 billion with Amazon, $115 billion with Google and $161 billion with Broadcom, and says planned capital spending could push the total higher. He argues the business model remains unproven and the company is a risky venture investment, not a safe one.
Published Oct 28, 2026.
Transcript · 24 min read
Prof G Markets, hosted by Ed Elson, with Steve Eisman and Dan Chiolan.
Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there. Where the podcast’s audio and the video part ways, some times are a minute or so off.
If money is evil, then that building is hell. The show goes on! The folks in there are questions so soon! Welcome to Prof G Markets. I'm Ed Elson. It is October 28th. Let's check in on yesterday's market vitals. The major indices climbed, with tech rallying for a second day. All eyes will be on Microsoft and Google's earnings tonight. Brent crude was volatile again, but ended the day around $114 per barrel. the yield on 10-year treasuries was stable. Able, ahead of the Federal Reserve's interest rate decision out later today. And finally, Metashares declined nearly 2% following the completion of its $25 billion stock sale.
The sale was part of its plan to raise nearly $100 billion in equity for its AI ambitions. Reportedly, though, the offering was five times oversubscribed. Okay, what else is happening? As we discussed yesterday, Anthropik's highly anticipated S1 is out and investors are still digesting the filing. Today, we wanted to get another opinion about the company's upcoming IPO and some of the questions that have gone unanswered, including its financial feasibility, anthropics governance, and also the state of the AI bubble at large. So here to break it down is one of our favorite guests, Steve Eisman, the legendary Big Short Investor and host of The Real Eisman Playbook.
Steve, great to see you. I'm going to start. General, what did you make of this filing? What jumped out to you? Well, the thing that jumped out to me the most were two things. The first was, there's an awfully long list of future capital commitments that add up to something, I mean, it depends what you include, don't include, $800 to $900 billion. You know, when you're at $800 to $900 billion, what's $100 billion amongst France? It doesn't really matter. her so that's just one thing that that is i don't know if the word is concerning but just just the way these people operate is you know that's just what they do so that's the first thing what i thought was even more interesting was that um the numbers are only through the second quarter
Read the whole transcript
“Third quarter, token maxing is over. Sequential revenue growth goes from 147% to 45%. And the revenue divided by compute goes from 2.08 to 1.7.”
Right. That's why I bring it up. And there has been speculation that that was an intentional decision in order to hide what seems like a reality, which is that the business growth is slowing quite significantly, as you have just explained.
Do you think it was intentional? Only very intentional. That's all. Only very intentional. Very intentional. Yes. Not intentional. Very intentional.
“when you commit to $800 billion, your expenses are going up.”
i mean what's there to talk about you're you're a shareholder in this company you have no rights Okay. Let's move on. If you want to invest in that company, that's up to you. It's been disclosed. You have no rights. Zero rights. You want to play? Play. If you think that's very important, then you won't play.
Just to clarify for our listeners. There are multiple classes of stock here. You've got the regular voting shares. You've got the regular non-voting shares. Then you have this class T share, which is controlled by what they call the long-term benefit trust. And apparently, this is sort of this committee that gets to choose the board of directors. But most importantly, there is also a class F share, one single class F share, which is controlled by the seven co-founders of the company. And the share, most importantly, carries 50.1% of the vote. How much does that matter when evaluating an investment like this?
Like, you know, beyond if you would like to have a vote, does it tell you something larger about this company and whether it's investable? I don't think so. So, yeah, we could make the same criticism of Meta. And, you know, Meta as a stock generally has done very, very well. But this is what goes on in tech world. They don't like their own shareholders. That's basically what's happening. Founder worship, institutionalized. It's supposed to go public at $2 trillion. That has been the expectation. There has been some debate as to whether that price might be cut. People were talking about $1.5 trillion.
What is your expectation for when this thing goes out? I have no expectations. I have no information on what I read in the papers. Do you have a view on investor sentiment? I mean, one thing that I can't quite figure out myself is, did this filing make investors more or less confident in the business? I think it actually made it more confident because people weren't pointing out that it's only going through the second quarter and what that meant so it's gonna be very interesting to see what sentiment looks like if if and i don't know if this happens they have to drop third quarter numbers in before they go bubble yeah the third quarter numbers could potentially change the game if that happens would you expect that the sentiment will turn quite precipitously i mean it should but you know we're not conducting a group therapy session with the investing world so i don't know how people that's all i want me too i don't know how people are going to react i never know i absolutely don't i love when people you know some people they'll always say it's in the price it's in the price and i always want to say how do you know like who did you ask Did you take a poll?
Like, on basis, do you say that? But people say this all the time. And I always say, I don't know. Just as we start to wrap up here, any other things that you're looking at in AI world, things that you think are especially important that we should keep track of? Well, the other big thing to keep track of, obviously, is the price of CoreWeave's debt. You know, at 15%, I think, although I can't prove it, I think it's prohibitive for them to raise debt at 15% and build AI data centers. but you know if it gets worse then they're really out of business you know they their business model is basically we raise debt we build ai data centers that's it that's the entire business model and if they can't raise debt they can't build ai data centers that's really the now maybe you know NVIDIA will come in and backstop the whole company you know it's certainly possible But barring that, you know, I would be watching the debt markets.
Steve Eisman is the legendary Big Short Investor and host of The Real Eisman Playbook, which is a great podcast. I highly recommend it. Steve, always appreciate your time. Thank you. After the break, the state of play in the autonomous vehicle race. And for even more markets insights, you can subscribe to my weekly newsletter. Simply put, go to edwardelson.substack.com. We'll be right back. badly. And you'll also get to participate in audience Q&A, so bring your best questions. Scott will probably still answer the ones you didn't ask, but that's okay. Get your tickets at PivotTour.com. That's PivotTour.com. We will sell out, so get your tickets. We'll see you in Denver.
Support for the show comes from Gusto. Nobody dreams about payroll tax deadlines, and yet they show up anyway like a walking nightmare you're forced to deal with on a quarterly basis. That is, unless you have Gusto. Gusto is online payroll and benefits software built for small businesses. It's all in one. remote-friendly, and incredibly easy to use. So you can pay, hire, onboard, and support your team from anywhere. Automatic payroll tax filings, simple direct deposits, health benefits, commuter benefits, workers' comp, 401k, you name it. Gusto makes it simple. They have options for nearly every budget. You get unlimited payroll runs for one monthly price, no hidden fees, no surprises.
They've got plenty of built-in automated tools designed to help you save time, like offer letters, onboarding docs, direct deposit, and more. It's quick and simple to switch to Gusto. just transfer your existing data to get up and running fast. Plus, you don't pay a cent until you run your first payroll. Try Gusto today at gusto.com slash markets and get three months free when you run your first payroll. That's three months of free payroll at gusto.com slash markets. One more time, gusto.com slash markets. Are you tired of hearing about business, tech, and finance from the bros? We are the Tiger Sisters, a top business podcast with the Vox media podcast network tiger sisters releases every monday and we decode money power and love and who doesn't love all those things we are the internet's wall street and silicon valley big sisters tune in every monday for late night sister talk meets boardroom strategy check out tiger sisters on youtube spotify or wherever you get your podcasts welcome to the tiger fam we're back with prof g markets earlier this week we discussed the future of tesla and its foray into the lucrative autonomous robotaxi business but while tesla rolls out their robotaxis across
the country they face one major obstacle and that is competition companies like alphabet amazon and Uber are also looking to capitalize on the future of transportation. So we would like to take a moment to attempt to answer the following question. Who is actually winning in autonomous vehicles? And what does the future of autonomous actually look like? Today, we're speaking with our very own research team associate, Dan Shallon, who recently presented on this very topic on our ProfG Plus substack. It was a great presentation. I learned a lot. So that's why I wanted to get into it today dan i see you every day but thank you for joining me on the show i just want to start with a general question which you started with in your presentation which i thought was very helpful which is how big of a problem um our vehicle deaths today? In other words, make the case for us why autonomous is actually a good thing. It's sort of hard to conceptualize just how dangerous driving is because we as Americans do it so often. We spend on average about an hour a day behind the wheel. But every year, 40,000 Americans die in car crashes. Globally, that number is over a million. It's the leading killer of people ages 5 to 29. And I think why those numbers are relevant to autonomous vehicles
is because 94% of car crashes are due to human error. So really, driving isn't dangerous because cars are dangerous or roads are dangerous. It's because humans are dangerous. So I think that's why autonomous vehicles are a very exciting technology, because it can allow us as Americans, as people around the world, to take something that we need to do every day and make it significantly safer. Looking at the players in this industry, obviously Tesla is getting into RoboTaxi's. The rollout has been, I don't even know what to say about it anymore. It's been, it's taken a while. Who are the actual players in this industry though? And how do they all stack up against each other? Yeah. So I think we can start with Tesla. And to understand Tesla's autonomous business, you really have to split it into two. So the first is the consumer side.
And that's FSD. That's the software that Tesla sells directly to Tesla owners, which allows Tesla drivers to have the car drive itself under supervision. Currently, about 1.7 million people globally pay for it. But then there's the other side of their autonomous business, which is their robo-taxi business, which is their autonomous ride share business that's in direct competition with Waymo and other players here in America. So here they're doing well. They're scaling very quickly. They have a fully commercial service in six cities. And just last month, they started putting cybercabs on the road, which are their purpose-built autonomous vehicles with no steering wheel.
So despite the fact that they are moving very quickly, they are very clearly in second place to Waymo. Currently, Waymo does about seven times the amount of miles that Tesla does weekly. They're in 15 American cities. They're doing 500,000 paid rides weekly. So they're very far in the lead. I think there are two others worth mentioning. The first is Zoox, which is owned by Amazon. It's their self-driving startup. They're currently in two American cities, and they have about 100 vehicles. on the road. So they're pretty early stage. And then the last one is Uber, which actually doesn't have any vehicles or technology on its own. But through partnerships with other autonomous vehicle companies like Zoox, like Waymo, it's able to facilitate autonomous rides in five American cities currently. I'm surprised to hear you say that the Tesla rollout has been quick because you say those numbers those waymo numbers which are phenomenal compared to tesla i guess what has been promising about the tesla rollout of their robotaxi specifically in contrast to waymo which on the outside it seems like waymo has been a lot more successful yeah so i think the thing that that tesla has really done well in these past few months is scale up very quickly um for a while
sort of in typical Elon Musk fashion, we were getting promises about an autonomous service rolling out in Austin and California, when in reality, you'd call a Tesla robo-taxi and there'd actually be a human safety monitor behind the wheel. In the past couple months, they've ditched the safety monitors in six cities, like I said, they're fully commercial, fully autonomous. But the biggest advantage that Tesla has over Waymo is cost. So those Tesla cybercabs cost about $25,000 to produce, whereas Waymo's costs upward of $100,000. They have a sort of much more thorough sensor load in the vehicles that they produce, which makes them more expensive. It makes them harder to build and take longer to build. So Tesla's really ramped up. Now, still, I don't mean that to say that Tesla is in first place in any sense. Tesla is still significantly behind. But just in the past couple of months, they've gone from kind of not existing in terms of their
“America is definitely in the lead because there's no company in China that's doing as many paid rides as Waymo is.”
One of the other obstacles for autonomous has been regulation. What is the regulatory picture in America look like right now? So it's currently very fragmented, and that's because it's all happening at the state level. There's really been no federal framework or action sort of passed by the federal government to guide these companies on how they should be regulated. And that is a very messy process because it means each of these companies needs to go through the painstaking process of lobbying each state individually to be able to test and then to, you know, be able to operate non-commercially and then commercially and then expand from there. And so what that has produced is a pretty large lobbying battle between, you know, drivers unions and unions at the state level and then these robo-taxi companies.
We've also seen Waymo dramatically ramp up their federal lobbying spend in the past couple of years. years. So I think the political battle is going to continue to play out in America. I do think it's also interesting to note that autonomous vehicles are not sort of insulated from the political backlash that the rest of, you know, AI and autonomous robotics in general have received here in America. We talk about it on the show often. It's the same force that has sort of prevented data center construction in many local communities. So public opinion on AI is souring, and I think we're seeing that reflected in what people think about autonomous vehicles.
There's a really interesting poll here. Gallup polled Americans in 2018 about whether or not they think that some amount of autonomous vehicles improve road safety overall. And at that time, about 52% of Americans agreed. They were in the same poll last year, and the share of Americans who agreed actually went down by 12%. And that's despite the fact that, you know, in that time, autonomous vehicles went from basically not existing to getting exponentially safer every single year. So it's really, you know, it's impossible to divorce people's sentiment about AI and this technology in general from the technology itself.
Despite the fact that it is statistically safer, is my understanding. um as we wrap up if you wanted to make bets on autonomous say someone's listening to this they're
“currently 95% of passenger car LiDAR sensors, which are in these autonomous vehicles, are made in China.”
Really fascinating stuff. Thank you. If you want to watch a replay of the live stream that Dan did on this industry, it's like an hour long, I want to say. It's super in-depth, but it's super informative. You can go to profgmedia.com, click on the Prof G Plus tab. We do a lot of these more in-depth deep dives over on Substack. Dan, thank you. I'll see you tomorrow. Appreciate your time. Thanks for having me, Ed.
Before we end, let's return to Anthropik's S1 filing, which has polarized a lot of investors. Some look at the revenue growth and they say that the company is incredible. Others look at the net losses and they say it's terrible. The largest IPO in history is bringing out a lot of emotions from investors, which I get. But it also makes me think that maybe this IPO is fundamentally misunderstood. So to end this episode, I would like to do a little bit of myth busting on Anthropic and more specifically on Anthropic's IPO. So. Number one. Yes, the company did rack up an $80 billion net loss in the first half of the year, which sounds bad, but no, that doesn't actually matter. Why? Because those losses were a product of the increase in the fair value of Anthropik's convertible notes. In other words, investors lent Anthropik money, and in return, those investors later received equity. And because of how accounting laws work, Anthropic technically had to write that up as a widening expense. But in reality, it was simply a reflection of how valuable Anthropic had become in a remarkably short period of time. So the $80 billion net loss, that's a red herring. Okay, number two. Yes, Anthropic's
first half revenue grew 1300% year over year, which is incredible. But no, as Steve points out, it has not kept up that extraordinary growth, not even close. The part that the S1 in many places left out is that revenue growth slowed down significantly in Q3. As Steve said, it was 150% quarter over quarter in Q2, and then less than 50% in Q3. Now, that's not necessarily bad, but it does show that maybe the AI frenzy is beginning to come to an end. And finally, number three, yes, Anthropic did narrow their losses quite significantly. It was more than $2 billion in the first half of last year versus less than $700 million this year. That is good, but no, that trajectory is not going to continue. The most important numbers in the filing came at the end of the document, where we learned that Anthropic has signed more than $800 billion in future compute commitments, none of which are reflected in the current income statement, but all of which eventually will be. That includes an $110 billion contract with Amazon, a $115 billion contract with Google, a $161 billion contract with Broadcom. What it doesn't include, however, are the billions of dollars in capital expenditures, which they are about to
register probably to build out their own compute. That's not something I believe because I had a hunch about it. I believe that because they literally told us that in the S1. In other words, that $800 billion number is actually probably going to be a lot higher. And as a result, this company's losses are about to explode. Now, did they time this IPO so that we wouldn't have to look at those losses on the income statement? possibly, but that's a theory. It's not a fact. And this segment is specifically devoted to things that we know for sure. The truth about the Anthropic S1 is that the income statement doesn't tell you the full story. It doesn't mean it's a bad company. It doesn't mean they don't have an incredible product. They do. But it does mean that there is still a lot of risk. The business model hasn't been figured out yet. And until it is, we must treat Anthropic with the appropriate level of caution. It's still a highly risky venture bet. Things could still go spectacularly wrong. What it certainly isn't is a proven safe corporate investment, something that it's potentially $2 trillion valuation might suggest. In sum, not much has changed. But whatever picture the profitability is painting right now, all I can tell you is that next year, that picture
is going to look drastically different. Okay, that's it for today. This episode was produced by Claire Miller and Alison Weiss and engineered by Benjamin Spencer. Our video editor is Brad Williams. Our research team is Dan Chalon, Kristen O'Donoghue and Mia Silverio. And our social producer is Jake McPherson. Thank you for listening to Profit G Markets
