Latest / Super Intelligence News and Report / Anthropic's $2 Trillion IPO Bet Meets $567 Billion in Compute Bills
Transcript
- Super Intelligence News and Report — Anthropic's $2 Trillion IPO Bet Meets $567 Billion in Compute Bills Host 1: Welcome to super intelligence news and report for Tuesday October 27th Host 2: Glad to be here Host 1: for you listeners who you know track the absolute bleeding edge of AI in business our mission on this deep dive is a rapid fire breakdown of of an industry that is well frankly eating itself alive today Host 2: Yeah building a mechanical god takes a lot of resources Host 1: exactly so we are skipping the basic definitions and getting right to the numbers the dates and what's actually shifting the ground under our feet Host 2: let's jump right into right into the research then Host 1: Right, so yesterday, October 26, 2026, OpenAI announced their models proved three major open problems. Host 2: Three of them? Host 1: Yeah. Host 2: Wow. Host 1: Yeah, all at once. Host 2: Okay. Host 1: They disproved the exponential time hypothesis. They found that on October 16th and checked it on the 17th. Plus, they proved Sage's 1991 local smoothing conjecture. Host 2: Which is just massive on its own. Host 1: Totally. And they generated an existence proof for 4D quantum field theories. Host 2: Oh, wow. Host 1: The math and ETH proofs are actually already lean checked on GitHub. Host 2: Right. And the physics one was held back by Dave Gross's team. I mean, so they could write it up themselves. Right. Host 1: Yeah, exactly. Host 2: And all of this is confirmed. So I mean, is this like AI suddenly finding a skeleton skeleton key for theoretical computer science? Host 1: I mean, it is definitely AI rewriting top tier math and physics, but wait, experts still have experts still have to verify that the formalized proofs actually match the original conjectures. Host 2: Right. Because Lean just proves the internal logic of the code is flawless. Host 1: Exactly. Host 2: It doesn't prove prove it's answering the right question you still need human shepherds to validate that translation Host 1: makes sense Host 2: but you know making these breakthroughs requires a truly massive amount of compute Host 1: Which brings up how how these companies actually pay for all that that infrastructure. Host 2: Yeah. And they are going straight to the markets. So also on October 26, Meta announced plans to raise up to $9.5 billion in stock. Host 1: Wait, 95 billion? Host 2: Yeah. Though only 12.5 billion of that arrives up front. But here is the wild part. NVIDIA, AMD, and Broadcom have expressed non-binding non-binding interest in buying up to $5 $5 billion of those shares. Host 1: Wait, wait, really? Host 2: Yeah. And that is confirmed. Host 1: Hold on. So NVIDIA is basically buying Meta stock to give Meta the cash to buy NVIDIA GPUs? Host 2: Yeah, it creates this bizarre circular financing loop. Host 1: That sounds dangerously close to a bubble. Host 2: Right. Right. A company like Meta that spent years aggressively buying back its its own stock is now well printing it to fund the AI arms race. Host 1: Just wild. Host 2: The chip suppliers are essentially funding their own their own customers to keep demand artificially high. Host 1: Just to keep the wheels turning. Host 2: Exactly. But, you know, printing stock isn't the only way to pay for this compute. Sometimes the money comes straight out of payroll payroll. Host 1: Oh, right. The Microsoft news. Host 2: Yeah. Yesterday, Microsoft offered voluntary voluntary buyouts to roughly 60,000 60,000 employees. This is at the VP level and below. The buyout offer itself is confirmed. Host 1: Right, right. Host 2: But Bloomberg reported, and this is reported, not confirmed, that Xbox and Copilot teams are included in the the cuts. Host 1: Wait, really? Host 2: Yeah. While data center and sales staff are excluded from it. Host 1: Wait, let me push back on that for a second. I mean, isn't it wildly counterintuitive that Microsoft might shed staff from its flagship Copilot team? Host 2: It sounds totally backwards, yeah. Host 1: Right, like you axe the people building the AI just to afford the servers that run it. Host 2: It feels counterintuitive until you look at the balance sheet. Tech giants are just aggressively rebalancing their budgets. Host 1: Oh, I see. Host 2: They are directly trading human payroll for AI infrastructure capital. if you protect the data center center staff but let the developers walk that is a very clear signal Host 1: so human capital is literally becoming secondary to silicon Host 2: exactly Host 1: wow and all that expensive infrastructure is well it's about to get weaponized in weaponized in a pricing war right now Host 2: yeah let's look at the products and tools Host 1: right so jumping back to October 26 OpenAI released released its mid-tier model GPT 6.1 Terra Host 2: which is is incredibly cheap right Host 1: very they priced it at one dollar per million input tokens and five dollars per million output tokens Host 2: Wow Host 1: yeah and and it matches Astra on the agent's last exam benchmark at like one tenth the price Host 2: that is a massive cost advantage Host 1: it really is the information actually reported again reported not confirmed that Rippling has already moved coding work coding work over to OpenAI because of it Host 2: The model release itself itself is confirmed though, right? Host 1: Yes, that is confirmed. Host 2: I mean, OpenAI just ignited a brutal middle tier price war on the exact same day that their biggest rival, Anthropic, is pitching its S1 to investors. Host 1: Oh, wow. So they are they are intentionally cratering the unit economics to like sabotage Anthropic's valuation. Host 2: Exactly. It It is a surgical strike on their IPO pitch. IPO pitch. Host 1: A total carpet bombing. So look at the through line of what we've covered for you today. I mean, AI is solving solving reality altering physics. Host 2: Right. Host 1: Tech giants are spinning up these circular financing loops with their chip suppliers. Host 2: and cannibalizing their own human workforces for compute. Host 1: Yeah exactly it really leaves you to ponder whether this entire news cycle proves the traditional economics of the tech industry are just collapsing under the weight of super intelligence Host 2: It is a profound question Host 1: now over to the main story Host 1: The numbers are officially out and they are breaking basically every conventional rule of Wall Street physics. Host 2: Oh, absolutely. Host 1: Right. Host 2: It's wild. Host 1: As of Monday, October 26, 2026, Anthropic has officially filed its public S-1. Host 2: Finally. Host 1: Yeah. For a NASDAQ listing under the ticker symbol ANT. A-N-T. We are looking at what is without a doubt the most anticipated and probably the most polarizing public market debut debut of the decade. Host 2: It It really is the moment the entire tech sector has been, you know, holding its breath for. Host 1: Yeah. Host 2: I mean, for months, we've had speculation. We've had all those leaked drafts. But now the actual audited numbers are out in the daylight. And the sheer scale of the financial mechanics at play here is just it's staggering in both directions. Host 1: It is. And, you know, that really brings us to the core mission for this deep dive. Because you have to ask yourself, as an investor or even just an observer of this industry, based on the hard data in this S-1, can Anthropic's growth actually justify a target valuation of, 1.5 to to 2 trillion dollars? Host 2: Which is a massive number. Host 1: Massive. And I need to state right up front, that valuation is reported, not confirmed. But like, how do you justify any price tag in that stratosphere when the vast vast majority of your revenue is pay as you go? Host 2: Right. Host 1: Well, you know, simultaneously you've locked in in over half a trillion dollars in compute costs for the years ahead. Host 2: Yeah, that is the defining tension of this tension of this entire document. You've got this incredibly flexible non-binding revenue on one side of the ledger Host 1: Yep. Host 2: and then these these massive inescapable just concrete liabilities on the other it's a fundamental mismatch in how the business operates versus how it actually pays its bills. Host 1: Exactly. And to to put this financial dynamic in perspective for you just imagine you signed a five year ironclad lease on a massive state of the art skyscraper. Host 2: Okay right. Host 1: You are absolutely on the hook for billions of dollars in rent over the next half half decade, no matter what happens. But every single one of your tenants filling that building. Host 2: They're They're on month to month contracts. Host 1: Exactly. Month to month. If they decide they don't need the office space next Tuesday, they just pack up and walk away. No penalty. Host 2: But you you still have to pay for the building. You Host 1: You still have to pay for the building. And that that is exactly how Anthropic has structured its infrastructure. Host 2: Yeah, that analogy perfectly captures the systemic risk here because you know the underlying hardware infrastructure required to train and run these massive AI models it forces you to secure secure capacity like 12 to 18 months in advance. Host 1: Right. You can't just buy it on demand. Host 2: No, you have to reserve the chips, the data centers, the energy grids but the customer on the other end, the enterprise actually using the AI, they just pay for the computational tokens they consume today. Host 1: Right. So you know to understand why this IPO is such a high wire act we have to look at the gauntlet they're running outside of just the balance sheet. Host 2: For sure. Host 1: Because this filing officially starts the roadshow clock. They've got this massive syndicate of 22 banks led by Goldman Sachs and Morgan Stanley. Host 2: The heavy hitters. Host 1: Right. But this isn't this isn't just a straightforward financial story. The geopolitical and legal friction surrounding this company right now is just it's intense. Host 2: It really is. And, you know, we have know, we have to look closely at the facts of this regulatory landscape without taking any political sides here. Because the Because the market just has to figure out how to price this risk in. Host 1: Right, exactly. Host 2: So back in February of 2026, 26, the Trump administration issued a a directive banning federal agencies from using Anthropic's technology. Host 1: And that was explicitly due to the company refusing to allow its models to be used for mass domestic surveillance and fully autonomous weapons, weapons, right? Host 2: Exactly. They refused. And following that refusal and the subsequent ban, the Pentagon officially labeled Anthropic a supply chain risk. Host 1: Wow. And the legal ping pong since that designation has just been relentless because in August, a district court actually sided with Anthropic ruling that the government's actions were retaliatory. Host 2: Right. A temporary win. Host 1: Yeah. But then just a month later in September, a divided DC circuit panel flipped that decision and and upheld the Pentagon's designation. Host 2: So it's still active. Host 1: It's still active right now. Anthropic is asking the full court to rehear the case and I mean if you add to that the summer export controls from the Commerce Department that effectively effectively suspended customer access for weeks the regulatory headwinds here are basically hurricane force. Host 2: Oh absolutely not to mention mention the internal philosophy of the company itself I mean Dario Amodei Anthropic's CEO he publicly called for a coordinated effort to slow down down the pace of AI development just just six weeks ago. Host 1: which is just a wild thing for a CEO to say right before asking Wall Street for billions of dollars. Host 2: It is highly unorthodox. Host 1: Yeah. Host 2: But, you know, it it highlights the unique stance Anthropic takes takes on AI safety. They even list it prominently in the S-1 as a risk factor, basically noting they could be disproportionately targeted by regulators for their safety positions. Host 1: Or conversely, that the that the tech itself could pose catastrophic risks to humanity. Host 2: Right. They are actively telling investors our safety principles might limit our growth. Host 1: Which is what what makes this next part so hard to wrap your head around. Because despite all this all this massive regulatory friction, despite the government pushback and you know despite their own CEO urging an industry slowdown Host 2: the numbers Host 1: the company's financial growth is completely defying gravity. Let's get into the key numbers because the top line revenue explosion is breathtaking. First half 2026 revenue hit 16.3 billion dollars. Host 2: Yeah. Host 1: That is 14 times higher than the first half of 2025. Host 2: 14 times. Host 1: Yeah I mean you almost never see scale like that in the history of enterprise software. Host 2: No never and the momentum isn't just like a one-off spike second quarter revenue was 11.6 billion dollars compared to 4.7 billion in the first quarter. Host 1: Wow. Host 2: Yeah they nearly tripled their revenue revenue two quarters in a row and this is at an incredibly massive base scale. Host 1: And here's the metric that actually explains how they're doing it. They reported a 900% net revenue retention rate. Host 2: Unbelievable. Host 1: Right. Okay. Let's unpack this because this isn't like selling traditional software subscriptions. That 900% means existing customers are spending roughly nine times more than they were a year ago. Host 2: Yep. Host 1: For context, the median for a public cloud software company is around 108%. How does a company structurally increase a customer spend by 900% in in 12 months? Host 2: Well, it comes down to how API usage works. when an enterprise integrates Anthropic's AI models into their core products say I don't know routing their customer service generating code analyzing internal data they aren't buying a fixed number of user seats. Host 1: Right like old school software. Host 2: Exactly they are paying for compute cycles so as their own employees and customers use those AI features more frequently the meter just runs faster and and faster. Host 1: it just scales up automatically Host 2: right once it's it's embedded into a company's daily workflow the usage just explodes organically they have over 11,900 customers spending at least 100 grand a year and about 250 spending 10 million or more Host 1: but if you just read the mainstream headlines this morning you didn't see that growth Host 2: no you definitely didn't Host 1: you probably saw a truly terrifying number an 80.7 billion dollar first half net loss I literally had to read that line in the S-1 three times an 80 billion dollar loss in six months shouldn't even be mathematically possible without bankrupting the company entirely. Host 2: Right. And that headline number is highly misleading, which is exactly why we need to dig into the actual accounting mechanics here. Host 1: Please do. Host 2: So out of that 80.7 billion, 72.4 billion is a non-cash charge. It's related to convertible notes being marked to fair value. Host 1: So it's basically a penalty for being too successful. Host 2: Exactly. As Anthropic's private valuation skyrocketed, the accounting value of those convertible notes rose along with it. Host 1: Ah, I see. Host 2: So it's a paper loss on the loss on the balance sheet, basically caused by the company actually becoming vastly more more valuable. And then another $8.8 billion of that loss loss comes from specific note purchase agreements with Google and Amazon. It is not cash flying out the door to pay for daily operations. Host 1: Okay, so if we strip away the accounting quirks quirks and those paper losses, what is the actual reality of their margins? Like, what are they losing on the the actual day-to-day business? Host 2: This is where the narrative flips completely. Host 1: Okay. Host 2: The The gap operating loss was only $693 million. And that is actually down significantly from 2.14 billion a year earlier. Host 1: Wow. So they're narrowing their operating losses dramatically. Host 2: Dramatically. And even more surprisingly in the second quarter Anthropic reported its first ever quarterly operating profit. Host 1: No way. Host 2: Yeah. Their operating cash flow was positive at 3.93 billion though you know they do note that's partly due to the timing of customer payments. Host 1: Still that sounds like an incredible turnaround. But and this is a massive but Host 3: We have to go back to the skyscraper analogy. Host 4: The liabilities. Host 3: Yes. Because 80% of this incredible revenue is usage based. It's It's month to month. But to serve those massive 900% spikes in usage, Anthropic is locking in compute capacity years in advance. Host 3: And the expenses they have legally committed to, are terrifying. Host 4: They really are. The S1 details between 542 and 567 billion dollars in total committed future compute and infrastructure spending. Host 3: Half a trillion dollars. Host 4: Over Over half a trillion. And to really grasp the velocity of that spend, their non-cancelable purchase commitments jumped from 54.6 billion at the end of 2025 to 265.6 billion by June 30th of 2026. Host 3: Wait, they added over 200 billion in non-cancelable obligations in just six months? Host 4: In six months. Yeah. Host 3: And the specific line items are just as wild. I mean, they are paying about 3.1 billion dollars a dollars a month to SpaceX just for NVIDIA GPU compute. Host 4: That is just a staggering number. Host 3: That single run rate would sink most Fortune 500 companies in in a matter of weeks. And for anyone wondering why SpaceX is suddenly a cloud provider, it really highlights the sheer scale of the the processing power required here. Host 3: You aren't just buying rack space in a standard server room. Host 4: No, you're renting compute capacity by the exaflop. You're leasing massive massive clusters of GPUs through basically anyone with the infrastructure and the energy grids to support it. Host 3: Right. Host 4: And we should note that Anthropic expects to use more short term compute in the near term to keep up with demand, which costs even more per unit. Host 3: Of course. Host 4: Right now, they generate a bit over over $2 of revenue for every $1 of compute cost, which translates to roughly a 40% gross margin. But it raises an important question for for Wall Street. Host 3: Which is? Host 4: How do you price a historic unprecedented growth rate and a first time operating profitability against half a trillion dollars in locked in hardware liabilities? Host 3: Right. And that brings us to the intense debate happening on trading floors right now, I'll be honest, I look at this reported two trillion dollar valuation and I think it's a fantasy Host 4: Really? Host 3: Okay. Host 4: You're taking the bear case. Host 4: I Host 3: I am. I mean, how can you justify that when we are starting to see signs of fatigue in the AI space? Gary Marcus, who is a very prominent AI skeptic, is pointing to what he calls the death of token maxing. Host 4: Right. The data wall. Host 3: Exactly. And for anyone not living on AI Twitter, token maxing is essentially this assumption that you can just keep force feeding bigger and bigger bigger mountains of data or tokens into a model and it will automatically just keep getting proportionally smarter. Host 3: Marcus is arguing that strategy is hitting a wall. a wall. Plus, we are seeing the rise of incredibly efficient, much cheaper cheaper Chinese models and signs of slowing overall growth in token volume. Host 4: Yeah. And tying into that macro skepticism, Apollo chief economist Torsten Slok pointed out over the weekend that AI hasn't really hasn't really delivered a measurable financial payoff for companies outside of the tech sector yet. Host 3: Exactly. Exactly. Host 4: The actual bottom line ROI is still largely theoretical for traditional businesses. Host 3: Yes. If a logistics company or a retail chain isn't seeing actual bottom line profit from AI, how long will they keep increasing their usage by 900%? Host 4: It's a fair question. Host 3: And Fast Company, along with Morningstar analyst Malik Ahmed Khan, raised a massive red flag in the data. The S1 gives us no cohort retention data. Host 4: Ah, right. We don't know who is staying. Host 3: Exactly. So we see that 900% net revenue retention overall, but we have absolutely no idea how widely that growth is spread out. What if that massive usage spike is just like five massive tech giants buying up all the compute while thousands of smaller enterprises are quietly churning out and canceling. Host 4: That's the fear. Host 3: Right. Without locked in customer contracts, that revenue could just evaporate the minute a cheaper alternative appears. Host 4: Well, those are valid structural vulnerabilities. I'll give you that. But there is a very strong counter argument here. Host 3: Okay. Let's hear the bull case. Host 4: The optimism is rooted in the sheer speed of this technological technological shift. Take him argues that obsessing over cohort churn right now just misses the forest for the trees. He compares this moment to the computing shift from horses to cars. Host 3: That's a big comparison. Host 4: It is but Anthropic isn't just selling a software tool. They are selling the new underlying infrastructure of the digital economy. And he points out that unlike the fiber optic overbuild of the dot com era, you know, where we laid millions of miles of dark fiber nobody used, there is a very tight supply of chips and memory today. Host 3: True, everyone is fighting for chips. Host 4: Exactly. This capacity is actually being heavily desperately utilized. Host 3: But how do the bulls ignore the half a trillion dollars in committed spend? I mean, even if you strip out the mark to market accounting, that is a staggering amount of liability for a company that just saw its first operating profit. Host 4: Well, bulls like Sonia Oja dismissed the loss headlines entirely as an accounting distraction that misunderstands hyper growth. And Malik Ahmed Khan Khan himself, despite his caution on the cohort data we just talked about, he calls this the fastest scale up of any technology business he has ever seen. Host 3: Fair point. Host 4: And Dan Ives at Yorkville Ives argues that institutional IPO buyers are going to completely look past the short term compute risks. They aren't buying this stock for 2026 profitability. They are focusing on the enterprise enterprise growth trajectory for 2027 and 2028. Host 3: So when you synthesize these two sides, it essentially boils down to just absolute faith in enterprise adoption. Host 4: Pretty much. Host 3: The bull case requires an unbroken exponential curve of enterprise adoption to absorb all all that compute capacity Anthropic is legally forced to buy. But the bear case exposes the fatal flaw of the business model. Host 3: If customer demand plateaus even slightly, those locked in hardware costs costs will just crush them. because they don't have the locked in customer revenue to offset the liability. Host 4: That is the multi trillion dollar gamble right there. And the minute this S1 dropped, we saw exactly how the how the broader market felt about it. Host 3: Oh Oh, yeah. So what does this all mean for the market? Host 4: The entire AI trade got a massive lift. The S&P 500 jumped 0.7% and it was almost entirely led by the chip stocks. AMD was up 2.8% Nvidia up 2.4 Broadcomcom up 2.3 Host 3: Wow. Host 4: Yeah Wall Street looked at Anthropic's massive compute bill and basically just saw a guaranteed revenue stream for the hardware manufacturers. Host 3: But if we map this out it highlights a a structural vulnerability that is genuinely wild. We are looking at a closed loop circular economy here almost half of Anthropic's first half revenue came came through cloud marketplaces run by Amazon, Google and Microsoft. Host 4: It is an unprecedented dynamic. Host 3: Just think about the about the conflict of interest here. Those three tech giants, Amazon, Google and Google and Microsoft are Anthropic's biggest landlords supplying the actual compute power Host 4: Right. Host 3: They are Anthropic's primary primary sales channel providing access to their enterprise cloud customers. Host 4: Right. Host 3: And because they are all developing their own internal foundational AI models, they are Anthropic's fiercest competitors. Host 4: It's crazy. Host 3: Anthropic is entirely dependent on on the very platforms that are actively trying to to render them obsolete. Host 4: Yep. Host 3: And it's not just a bottleneck in the supply chain. We are seeing are seeing it in the capital markets too. The race for funding is heating up at exactly the wrong time for an IPO. Host 4: It's crowded out there. Host 3: Very. Anthropic isn't the only one trying to drain the the markets available capital to pay for all this compute. On the exact same day Anthropic filed Meta announced plans for a 9.5 billion dollar stock offering specifically to fund their AI expansion. expansion Host 4: and in a perfect example of that circular economy striking again Nvidia AMD and Broadcom are reportedly looking at buying 5 billion dollars of that Meta stock themselves Host 3: Wait really Host 4: yeah the hardware companies are funding the software companies so the the software companies can turn around and buy more hardware from them Host 3: That is incredible and we can't forget OpenAI in all this. They are currently currently out there seeking another 30 billion dollars in private funding at a 1.25 trillion dollar valuation which again is reported not confirmed Host 4: Right Host 3: but the point is everybody is courting everybody is courting the exact same pool of public equity investors at the exact same same time Wall Street is about to have about to have to make some incredibly ruthless choices on who actually gets the capital to survive and they'll be be making those choices based on variables we can't even fully see yet Host 4: which brings us to the great unknowns of this S1 because despite dumping hundreds of pages of financial data onto the market there are still massive blanks we need filled in before this stock trades Host 3: right like we don't have the final price range we don't have the share count we don't know the actual size of the capital raise they are targeting Host 4: exactly Host 3: we also don't know if that surprise operating profit from Q2 Host 5: is going to hold. Especially since the filing explicitly explicitly warns they're going to rely on more expensive short term compute in the near future just to meet demand. And crucially, we haven't seen the Q3 revenue breakdown, to know if that that tripling growth rate actually carried into the fall. into the fall. Host 6: Then there are the the governance and legal risks you have to watch closely. Host 5: Oh, the founder LLC? LLC? Host 6: Right. Through a structure called a founder LLC the seven co-founders of Anthropic will retain 50.1% 50.1% of the voting power Host 5: Wow. Host 6: Yeah. So how are public institutional investors going to price a stock where they are handing over billions of dollars but have virtually zero say in how the company is run? Host 5: Especially for a company that has pledged 5% of its shares to a public benefit fund and where the founders are pledging 80% of their personal equity to charity. Host 6: Right. It is a very unorthodox mission driven structure for Wall Street to digest. And of course, everyone is watching to see how the full DC circuit rules on that Pentagon supply chain designation, which could, you know, lock them out of massive government contracts. Host 5: Right. And as for the timeline you need to watch, we are looking at at a reported road show beginning as soon as November 9th and a target listing date for the week of November 16th. Again, these dates are reported, not confirmed. Host 6: Just want to to make that clear. Host 5: Exactly. Though, if you look at Polymarket, traders are currently giving an 83.5% probability that shares will will be actively trading by November 30th. Host 6: Well, I want to leave to leave you with one final slightly mind bending fact from the filing to mull over as we wait for this to unfold. to unfold. We talked about those massive Fortune 500 companies spending over $100,000 a year with Anthropic. Host 5: Right. Host 6: Well, in the third quarter, Anthropic accounted for on average less than 0.1% of 0.1% of those companies total expenses. So you really have to ask yourself does that tiny fraction mean there is a massive untapped runway for future growth as as AI eats the enterprise? Host 6: Or is it proof that enterprise AI adoption is currently a mile wide but still only an inch deep? Host 5: We'll see you back here tomorrow at the same time and please check the show notes for all sources.