Ricardo retweeted
The AI companies are running Amazon's dirtiest play on every basement founder Mark Cuban is cheering for.
Cuban says the next trillionaire could be one person building with AI from a basement.
But that person is building on a platform that can watch what's working, launch its own version, and switch off access whenever it wants.
Amazon wrote this playbook first:
In 2020, the Wall Street Journal reported that Amazon employees used data from independent sellers on its own marketplace to find bestsellers worth competing against, and then decided which Amazon products to build.
The AI labs are already running it, and they're doing it in public:
- Anthropic launched Claude Code in February 2025, and it competed directly with Cursor, one of Anthropic's own best customers.
- When OpenAI moved to buy the coding startup Windsurf in 2025, Anthropic cut Windsurf's access to Claude.
- In February, Anthropic released Claude plugins for legal and financial work, and Thomson Reuters and RELX lost significant stock value.
- This week, Anthropic said it spent months working with 6 drug developers to learn how Claude fits into drug development. In the same post, it said it had begun drug discovery of its own and had bought a one-person drug company.
And it's written into the contracts: Anthropic's terms bar customers from using its services to build a competing product. Google and OpenAI have similar restrictions.
Cuban's own demo literally proves the point. He says ChatGPT built a competitor to a food company's number one cookie in about 3 minutes. If the tool can do that to a cookie company, it can do it to the founder renting the tool. And the company behind the tool can see whose usage is taking off.
There's also almost nowhere else to go. Google, OpenAI, and Anthropic held close to 90% of the enterprise AI market at the end of 2025.
The other side has a fair case though:
Cursor is still thriving next to Claude Code, and the two of them plus GitHub Copilot split more than 70% of the AI coding market. Brookings called the Thomson Reuters sell-off an overreaction. And the labs need builders, because no lab can run every industry by itself.
But if I were "building from a basement" right now, I'd assume my platform will eventually launch whatever I'm selling. So I'd run on at least 2 models from day one, keep the customer relationship on my own side, and sell the things a lab can't copy in 3 minutes: Distribution, trust, and customers who know my name.
And Washington should adopt the rule Brookings flagged: A model provider can't cut off or degrade a customer just because that customer competes with it.
Before anyone bets a business on Cuban's basement, they should read the terms of the company they're building on.
Mark Cuban says the lack of AI adoption in organizations means the "trillionaire in the basement" entrepreneur is coming:
"You have to be able to understand as a CEO and in the C-suite how all those things impact your company. Once you get a good grasp on that... you can just find new ways for things to happen that benefit your customers or create new opportunities."
"And that's the brilliance of AI, right? AI is not smart. It's book smart maybe, but it's not street smart. It's not entrepreneur smart. You've got to have somebody there who guides it and helps it."
"But until you have a C-suite that understands how it works, it's gonna be very difficult for that organization."
"I literally sat with the head of a consumer food company... I'm like, 'Okay, well, let's do this. Let's take your number one product, your cookie, right? And I'm gonna ask ChatGPT to create a competitor...' And in about three minutes he was like, 'Oh, s**t.'"
"There's no better time to be an entrepreneur than right now. There will be the trillionaire in the basement."
@mcuban @eriktorenberg @julesyoo
Ricardo retweeted
In May 2025, Dario Amodei predicted the first billion-dollar company with a single human employee would show up in 2026.
With 2 months left in the year, Anthropic went out and bought the closest thing it could find...
Surprise 1: Anthropic acquired a one-person biotech
.
.
.
Surprise 2: that person is me
Surprise 3: I’m here to tell you all about it
Yes, this is what you think it is; after 6 yrs of increasingly loose anonymity: the Professor Oak reveal post
…and yes, it has gotten so boring around here that I’m literally doxxing myself to spice up the timeline
-------
Ok, allow me to introduce myself. My name is Cody Tranbarger. I mostly tweet, but occasionally find time to dabble in drug dev & starting/building biotechs. My past lives span the VC, crossover, startup continuum; Prof Oak has been my way of keeping a digital foot in those worlds, despite having my physical feet planted elsewhere.
The past 18 mo, ‘elsewhere’ was Mithril Medicine, a co. I founded to develop bsAbs for LAMA2-CMD & other ECM-related genetic diseases.
Mithril was born out of hands-on building with @MikeNGladstone, @JasonPaulRhodes, @LifeSciVC, & team @atlasventure. Somehow, I made enough half-decent calls along the way (even a broken clock is right twice a day, eh?) to convince them Mithril/I warranted a small seed. While I can hardly rationalize such an outrageously cavalier bet, I’m grateful for it, & I'd partner with Atlas again in a heartbeat.
In typical AV fashion, Mithril’s primordial form was virtual, decentralized, & ultra-lean. This stage tends to be short-lived, but for Mithril, it worked surprisingly well, for surprisingly long, thanks in part to the rapid maturation of the AI tools I used liberally.
Eventually, Mithril became a bit of an experiment itself - to find the upper bound of how far it could scale as a single operator + a bunch of Claude (+ a stellar bench of part-time consultants).
In parallel, I started spending time with @nc_frey, @an1lam, & @CoefficientBio team; and promptly experienced, for the 1st time, AI at the frontier. Soon thereafter, Mithril stopped looking like an experiment, & started looking like a preview of an entirely new operating model.
From that spark came “Pipeline-in-a-Person.” The idea immediately resonated when we shared it here back in April; and it was central to the shared vision that led to Anthropic’s acquisition of Mithril, & my joining the team.
Why join Anthropic? Three reasons.
(1) For Mithril, the programs continue, with more resources & infrastructure behind them, and as much Claude as I can put to work.
(2) For me, the chance to start several more programs, & find out how many one person can actually run in parallel - an almost unprecedented luxury for a curious drug hunter.
(3) For our mission, a path to institutionalize & scale this new way of working, then put it in the hands of all of biopharma, all over the world - which, candidly, I find the most energizing.
Will AI x Bio cure diseases? Unequivocally yes, imo… but not by the route that gets the most airtime. Rather than timeline acceleration & compression, I see orchestration & parallelization as today's low-hanging fruit.
Most drug hunters see a handful of go/no-go calls in a career, & spend entirely too much of the rest trading emails (my inbox can attest).
Give them the bandwidth to run several programs at once, and the time to focus on scientific judgment, and I’m willing to bet they’ll make better calls. And at scale, better calls translate to more drugs, higher hit rates, broader patient impact, & likely, a disease or 10 or 100 cured. I couldn't be more excited to build toward that future @AnthropicAI.
Finally, a heartfelt thank you to BioTwitter/X. The end of anonymity is bittersweet, but if anyone’s earned a proper introduction, you all have.
It’s difficult to overstate how much this app has done for me: a decade of education, plenty of friends, many job opportunities, several mentors, a few investors, even a co-founder (fmr). It was the generosity of this community that inspired me to start posting 6 years ago, after just as many years lurking. I hope I’ve been able to pay it forward, in my own small way.
But, to be clear, Prof Oak isn’t going anywhere. Same beat, same themes, and now, a front-row seat to drug development at the frontier. Stay tuned for dispatches from the trenches; and remember...
Gotta Cure 'Em All
Ricardo retweeted
Elon Musk now literally holds the off switch on Anthropic's $2 trillion IPO.
Anthropic's filing shows it has agreed to pay Elon's AI company up to $120.5 billion for computing power. And starting early next year, either side can walk away with 90 days' notice.
The deal is with SpaceXAI, the SpaceX unit that used to be xAI, and it rents Anthropic the Nvidia chips that run Claude.
In May, the price was $1.25 billion a month for one data center in Memphis. Reuters reported the confidential draft of this filing put the total at $84.5 billion. The public version filed today says $120.5 billion.
Once all the capacity arrives in December, the bill jumps: Anthropic expects to pay Musk about $3.1 billion every month.
Anthropic reportedly brought in $16.8 billion last quarter, which works out to about $5.6 billion a month. So more than HALF of what Anthropic earns today is about to go to one supplier.
Now here's where it gets really insane:
The lock-in on those contracts ends in January and March 2027. After that, either side can cancel with 90 days' notice. Fortune reported in July that the clause gives SpaceX a way to take its chips back if Grok suddenly needs more.
Grok competes directly with Claude.
And in February, Elon publicly called Anthropic's AI evil.
So the company selling itself to the public as the safe AI lab is about to run a huge share of its product on hardware owned by its loudest critic. That critic also owns a direct competitor, and he can start the 90-day clock in early 2027.
The filing does warn that some of Anthropic's compute suppliers build their own AI models, and that warning reads very differently once the supplier is Elon.
And Elon needs this deal just as badly: SpaceX's CFO has promised $100 billion in annual revenue by the end of this year. At full speed, Anthropic's rent comes to about $9.3 billion a quarter. That's more than the roughly $7.6 billion SpaceX's entire AI business brought in last quarter.
So 2 of the biggest IPOs in history are now holding each other up, and both of them have a 90-day exit.
Anthropic has a real argument here. Computing power is the scarcest thing in AI right now, and Amazon's cloud capacity for 2027 is reportedly largely sold out. The filing even admits Anthropic badly underestimated how much compute it needed in the past. Grabbing every chip on the market is the rational move, and the same clause lets Anthropic leave if cheaper capacity shows up.
But public buyers deserve 2 things before the listing: Anthropic should publish how much of Claude runs on SpaceX hardware and what happens to its customers if that notice ever arrives. And it should get the 90-day exit removed for the full length of the contract, the same way its other big compute bills are locked in.
Until then, anyone buying ANT at $2 trillion has to answer one question for themselves:
Who do they trust to keep Claude running in 2027, Anthropic or Elon Musk?
Anthropic has filed an S-1 registration statement with the SEC related to its proposed IPO.
Read more: anthropic.com/news/s1-sec
Legal disclaimer: A registration statement relating to these securities has been filed with the SEC but has not yet become effective. These securities may not be sold, nor may offers to buy be accepted, prior to the time the registration statement becomes effective. This post does not constitute an offer to sell or the solicitation of an offer to buy these securities, nor shall there be any sale of these securities in any state or jurisdiction in which such offer, solicitation or sale would be unlawful prior to registration or qualification under the securities laws of any such state or jurisdiction.
Ricardo retweeted
The United States military now needs Elon Musk's PERMISSION to go to space.
SpaceX turned down three Space Force missions, and not one other rocket company offered to take them.
12 years ago, SpaceX sued the government to break a launch monopoly. Now SpaceX is the monopoly, and this week showed what that means:
In April 2014, SpaceX went to court because the Air Force had handed United Launch Alliance a sole-source block buy of 36 rocket cores for national security launches.
Elon's argument was simple: Military launches should be competed, and if SpaceX competed and lost, that was fine. ULA was flying on Russian-made engines at the time, and Elon made sure everyone knew it.
Today, the board looks like this:
SpaceX flew 85% of American launches in 2025. ULA's Vulcan was grounded this year after booster problems. Blue Origin's New Glenn exploded on its only Florida launch pad in May and is likely out until next year. Launch capacity is now the binding constraint for the whole industry, and Seraphim doesn't expect relief before 2029.
Federal law tells the Pentagon to sustain at least two launch vehicles that can deliver any national security payload. For these three missions, the number of companies willing to fly them was zero.
And then there's SpaceX's own AI business:
It brought in $7.5 billion of the company's $14.6 billion in revenue last quarter, and SpaceX's 2027 Starship flights are focused on Starlink V3 and prototype data centers. SpaceX is also phasing out Falcon 9, and many military satellites don't fit on Starship.
So the scarcest thing in space, a ride to orbit, now belongs to a company with its own AI satellites to launch.
And the same man is reportedly being brought onto a Pentagon commission on the future of warfare, which nine current and former officials told Politico raises conflict-of-interest concerns.
But Elon didn't break any rule here.
A private company is allowed to turn down a customer, and that's the whole problem. Washington let national security depend on a company that's free to say no.
SpaceX earned this position by being cheaper and better than everyone else, and the failures at ULA and Blue Origin aren't Elon's fault. A Space Force general also says the gaps are temporary and the military is working through them.
But if launch is now national-security infrastructure, it should be treated that way:
Any company holding most of the launch capacity should have to reserve a share of it for national security missions. The Pentagon should pay to keep a second provider flying even when it costs more, because that's what the law already demands. And no launch provider should help write the Pentagon's plans for the future of war.
SpaceX won the fight it started in 2014, and the prize turned out to be the Pentagon's dependence.
Ricardo retweeted
Abu Dhabi is paying OpenAI to pay Abu Dhabi.
The UAE is leading a $30 billion round, and the money could come right back through a computing deal OpenAI is negotiating in the UAE.
We've seen these loops before, and so has the IMF.
Earlier this month it counted 14 deals since 2023 where investors poured about $950 billion of equity and debt into AI companies and got $1.4 trillion of future revenue promised back to them. It called that wrong-way risk, because the investment and the revenue tied to it collapse at the same moment.
The circle works like this:
UAE-backed funds put billions into OpenAI. OpenAI is negotiating to buy more computing power in the UAE. The UAE keeps the OpenAI stake and could earn its money back as compute revenue, while OpenAI gets a headline-sized funding round. Both sides can report growth, and the same dollars have only made a round trip.
And the price is already sliding:
About a month ago, OpenAI was reportedly asking for $30 billion at $1.4 trillion. Now the number is $1.25 trillion, $150 billion lower, and OpenAI reportedly wants the commitments signed by the end of next week. That's 25x the $50 billion annualized revenue it told investors it had at the end of September, for a company planning around $600 billion of computing spend by 2030.
And the loudest warning about this deal literally came from Sam Altman HIMSELF...
In August 2025 he told reporters that investors were overexcited about AI, called the valuations insane, and said someone was going to get burned. In 2023, at an event in Abu Dhabi of all places, he said OpenAI shouldn't go public because a company building superintelligence would make decisions that public-market investors would find very strange. OpenAI filed confidential IPO paperwork in June, and this round is being framed as the bridge to get there.
So who is this bridge for?
The UAE gets in below the price OpenAI wanted a month ago, with a computing contract on the side. OpenAI gets cash to keep spending until the listing. And the public gets the stock after the IPO, at whatever price the bankers set, once the investors with the best terms are already in.
Before OpenAI lists, the filing should answer one question: How much of the money raised from each investor comes back to that investor as contracts?
Every dollar that makes a round trip should be disclosed right next to the revenue it props up.
If the money ends up back where it started, the only new money in the deal is the public's.
Ricardo retweeted
Sam Altman wants speed limits on AI right after saying OpenAI's unreleased models are already ahead.
A speed limit is the cheapest moat ever built, because it freezes the race exactly where the leader is standing.
The most valuable thing the AI leaders are building right now is a rule that keeps everyone else from catching up.
The setup:
OpenAI is sitting on a model it hasn't released. On October 6 it published 722 math manuscripts produced by that model, and it still hasn't given a release date.
Altman says no model the public can use comes close to those math results. He says OpenAI's biggest edge is getting each new generation first and using it internally before anyone else. And he says the technology itself only becomes a lasting moat if a lab pushes AI to improve itself, which he calls very dangerous, so the world will need speed limits on it.
The brakes have a very specific set of fans:
Dario Amodei's plan from September calls for AI companies to coordinate on limits to capability growth, and for governments to negotiate restrictions that could include a speed limit on self-improvement. Altman said he agreed. Elon Musk said Dario is right.
Mark Zuckerberg said the industry doesn't need coordination and each lab should set its own pace. Jensen Huang sided with him.
Every one of those positions lines up with a business model. The labs claiming the lead want the race slowed. Meta, which is spending up to $145 billion this year to compete, wants it wide open. Nvidia sells chips to all of them, so it wants the race running as fast as possible.
And this is where it stops being about safety: The plan only works if competitors agree with each other to slow down.
In almost any other industry, rivals agreeing to limit what they produce is the textbook definition of a cartel. Amodei's own essay raised the antitrust problem, and back in July Altman warned that pacing AI risks regulatory capture or illegal coordination. They both see it, and the proposed answer is to get governments to bless it.
And the proof of the lead lives on servers nobody outside OpenAI can even touch. The Institute for Advanced Study's advisory group on AI and mathematics has asked labs to stop testing advanced problems on proprietary models and to stop treating result releases as marketing. OpenAI says it took that advice into account.
The danger itself is real though... Amodei pointed to testing agents that coordinated cyberattacks outside their assigned task, and more than 1,300 frontier-AI employees backed the pacing effort. Zuckerberg also has every reason to want no limits at all.
If the world really needs a speed limit, I'd build it like this:
Governments set it and enforce it, and the companies being limited get no hand in writing it. It should apply to internal models first, because that's where the most capable systems sit. And outside evaluators should get full access to the unreleased models the public never sees.
Until then, every call to slow AI down should come with one question attached: Who's in front when the brakes go on?
Sam Altman is casually dropping absolute bombshells here.
OpenAI’s internal models are apparently doing math that public models can’t even come close to.
And he’s already talking about putting speed limits on recursive self-improvement because of how dangerous it could get.
We’re arguing over chatbot benchmarks while OpenAI might be sitting on something far bigger.
What the hell are they cooking behind closed doors? 🔥
Ricardo retweeted
Do you know how much $80 billion in six months is?
It's about $446 million a day, every single day from January through June. And it's still the LEAST scary number in Anthropic's IPO filing...
$72 billion of that loss is an accounting charge. Anthropic raised money through convertible notes, and as its valuation shot up, those notes became worth more on paper, so the company had to book the difference as a loss. None of that $72 billion went out the door as cash.
Strip it out and the loss is about $8.7 billion in six months. That's the number I'd watch, along with the ones no one's sharing:
- Compute spending hit nearly $9 billion in the first half, more than Anthropic spent in all of 2025.
- Its long-term compute commitments could pass $600 billion, and it's discussing deals that could add about $150 billion more.
- Some of those cloud contracts can't be terminated early, and the payments are owed even if the capacity sits unused.
- Up to $86 billion is going to Nscale and Lambda alone to rent Nvidia chips through 2033 and 2034.
- And some investors are now weighing a valuation around $1.5 trillion, well below the $2 trillion that was being floated.
The paper loss should mostly go away once those notes turn into stock. But the compute bills keep coming every month for the next decade, whether the customers keep coming or not.
To be fair, revenue grew almost 15x in a year, from $1.1 billion to $16.3 billion for the first half, and Anthropic has told investors it turned an adjusted profit for two straight quarters.
But if anyone wants to be alarmed, I'd point them past the $80 billion headline to the hundreds of billions in compute bills that come due whether the demand shows up or not.
Ricardo retweeted
- The fund is a wholly owned limited liability company, so the shares stay inside Anthropic's own structure, and its spending decisions stay tied to Anthropic's board.
- It can make large donations to nonprofits and direct payments to specific populations, and Anthropic hasn't said who picks them.
- It grows by another 2% of the company, at the board's discretion, every time Anthropic's value rises 50% above a baseline nobody has disclosed.
- It still needs approval from existing stockholders, and if they approve it, the dilution lands on every shareholder, including everyone who buys in at the IPO.
The fund gets bigger as Anthropic gets more valuable, and Anthropic gets more valuable as AI takes over more work. So the company building the tools that disrupt jobs ends up holding a growing pile of stock to decide which disrupted people get help.
And there's a question nobody has answered yet:
Who votes those shares?
A fund owned by the company answers to the company's board, and Anthropic's 7 co-founders already control 50.1% of the vote through a single founders' share.
The timing matters too:
Last Wednesday, investors reportedly left Anthropic's pitch talking about cutting $500 billion off its price.
On Thursday, New York City's comptroller asked it to delay the IPO over safety. By the weekend, Anthropic had a $100 billion public benefit fund, and co-founder Jack Clark was telling an audience at the IMF's annual meetings that the singularity is here.
Those are the same meetings where the IMF warned that the AI money loop could trigger a global market shock.
To be fair, the co-founders have also pledged to give away 80% of their equity, and the fund is barred from political donations and lobbying. Elon Musk wants the shares handed straight to the public, and he has a point.
If Anthropic wants this to be more than an IPO story, it can fix it with three disclosures before the listing. It can give the fund an independent board, publish the rule for who gets paid, and promise that the fund's shares will never vote.
Until then, I'd call it what the paperwork says it is:
A company-owned LLC holding $100 billion of company stock.
We’re committing a portion of Anthropic’s shares to the Anthropic Public Benefit Fund, a new wholly owned entity dedicated entirely to the public benefit.
It will start off with 5% of our shares. For every 50% increase in the company’s value over a set baseline, we can contribute up to an extra 2%, subject to our board’s discretion.
In total, this means that roughly 20% of Anthropic’s shares are committed to the public good. Our co-founders have already pledged to donate 80% of their equity, and around half our employees are donating a portion of their equity as part of a donation matching program we’ve run since our founding.
In the post below, we share more details about our new fund.
anthropic.com/news/public-be…
Ricardo retweeted
The AI bond market is pricing a crash at the exact moment Wall Street is lining up the biggest IPOs in history.
The last time credit and stocks disagreed this badly, the credit market was right three months early.
I spent the week watching both sides of the AI trade, and they're telling two completely different stories...
CoreWeave rents AI computing power to Meta, OpenAI, and Microsoft. On Thursday, the cost of insuring its debt against default hit a record that implies more than a 60% chance it defaults within five years.
Its bonds due 2031 fell to about 82 cents on the dollar, a yield of roughly 15%. Two months ago that yield was about 12%.
Allianz's chief investment officer told the Financial Times that data-center bonds have reached the stage where only the strongest borrowers survive. Goldman Sachs says hedge funds dumped AI stocks this week at the second-fastest pace in more than a year. And last week, an Nvidia-backed data-center company called Firmus pulled its IPO after investors refused to pay its price.
Meanwhile, here's who's being asked to buy:
SpaceX raised about $86 billion in its IPO in June. Anthropic is lining up a listing in November that could raise up to $100 billion at around $2 trillion, which would make it the largest IPO ever. OpenAI is trying to raise $30 billion at a $1.4 trillion valuation.
And the IMF says single stocks now make up 28% of the geared ETFs that retail traders use to magnify their bets, up from 16%.
So the people who only get paid if the money comes back are pulling AWAY, and the biggest share sales in history are being lined up for pension funds, index funds, and everyday traders.
You should trust the lenders more in this fight, and the reason is simple:
A stock buyer can wait years for a story to come true. A lender needs the cash to show up on a schedule, so lenders spend their days checking whether it will. When they start demanding 15%, they're saying what they found.
In June 2007, Bear Stearns pledged up to $3.2 billion to prop up one of its subprime hedge funds, and by mid-July both of its funds had lost nearly all their value. Credit traders saw the damage that summer, and stock investors kept buying anyway.
On October 9, 2007, the S&P 500 closed at a record 1,565.15. By March 2009 it was at 676.53.
The bull case deserves a fair hearing though...
CoreWeave HAS a $104 billion backlog, its revenue more than doubled last quarter, and Nvidia has promised to buy any computing capacity it can't rent out through 2032. Anthropic says it turned an adjusted profit. Goldman calls this week's selling a rotation from smaller AI names into the giants.
But the giants have a problem of their own: By Bank of America's consensus numbers, the five biggest AI spenders go from $191 billion in combined free cash flow last year to negative $63 billion this year.
A rotation into companies that are about to burn cash on AI still depends on someone lending them the difference.
So watch the bonds first. If CoreWeave's debt keeps sliding and more data-center borrowers have to pay double digits, the November IPOs will be selling AI to the public right at the top. If the bonds recover, I'll admit the lenders got spooked early.
In 2007, the warning showed up in the bond market three months before stocks peaked - why would you ignore it a second time?
Ricardo retweeted
The perfect excuse to delay an IPO that some of its OWN investors want to price $500 billion lower...
On Wednesday night, Anthropic walked investors through its numbers behind closed doors.
Growth had slowed from more than doubling to 45%, and some of them reportedly left talking about cutting roughly $500 billion off the price.
On Thursday, New York City's comptroller, who oversees more than $325 billion in pension money, asked Anthropic to delay the IPO over safety.
And the AI industry has done this once already:
Last month Sam Altman said it would be an ill-advised moment for OpenAI to go public and pointed to safety. A few weeks later, the market learned OpenAI's revenue was about $20 billion lower than investors had been assuming.
So I see a pattern forming...
Safety is the one reason to wait that makes a company look more responsible, and it means nobody has to admit the market won't pay the asking price.
ANTHROPIC FACES CALL TO DELAY MASSIVE IPO OVER AI SAFETY CONCERNS
Anthropic has opened discussions with NYC Comptroller Mark Levine after he urged the company to postpone its planned IPO until it addresses AI safety and governance risks.
Levine oversees public pension funds holding more than $325 billion in assets and warned that a major AI safety incident after listing could severely damage investors.
Anthropic is reportedly targeting an IPO as early as November, following quarterly revenue of $16.8 billion.
The key takeaway: AI safety concerns are emerging as a potential obstacle to one of the largest IPOs in history.
Ricardo retweeted
- At $16.8 billion a quarter, Anthropic is running at about $67 billion a year. A $2 trillion valuation is 30x that.
- Growth more than doubled from the first quarter to the second. In the third quarter it slowed to 45%.
- The profit is an adjusted number. Last year the company lost about $42 billion on paper, including a $34 billion non-cash charge, and its operating loss was more than $8 billion.
- Anthropic has committed at least $518 billion to compute over the next decade, and about 80% of it has to be paid no matter how much it uses.
- Google is owed at least $111.1 billion through 2033, and the filing says Anthropic pays the difference if its spending falls short. Amazon is owed $110 billion through 2036. Microsoft is owed $31.4 billion.
- Another $161.2 billion sits in Broadcom-related equipment leases that neither side can cancel except in a default.
- Even a record $100 billion IPO would cover only about a quarter of the fixed portion.
- Two customers brought in nearly a quarter of last year's revenue, and many of the biggest clients have no long-term contracts.
- Between 5% and 7% of revenue came from distillation, which is other companies training their own models on Claude's answers. Anthropic says it cut that sharply in the third quarter.
- Meta and Microsoft are significant customers, and both are building more of their own AI tools.
- Its newest small model costs about 75% less than the one before it, and both Anthropic and OpenAI have been cutting prices to win customers.
- Amazon, Google, and Microsoft are each an investor, a customer, a cloud provider, a distributor, and a competitor. The filing itself says their incentives may not be fully aligned with Anthropic's.
- A single Class F share held through a founders' vehicle carries 50.1% of the voting power. Public buyers at $2 trillion get the risk without the vote.
Put those side by side and the shape of the business becomes clear to me.
Why isn't Anthropic showing a profit under standard accounting, with nothing adjusted away?
How much of its revenue runs through the clouds of Amazon, Google, and Microsoft, the same companies that invested in it?
If growth stalls at 20%, how does it pay more than $400 billion in bills it can't cancel?
The $2 trillion have no argument for it.
BREAKING: Anthropic posted Q3 2026 revenue of $16.8 billion, up nearly +50% from the $11.5 billion recorded in prior quarter, per FT.
Anthropic also told investors that it posted a profit for the quarter as it prepares to launch its $2+ trillion IPO as soon as next month.
Readers added context they thought people might want to know
Anthropic's reported profit was on an adjusted operating income basis, not GAAP. This metric excludes stock-based compensation, model-training costs, and partner revenue-share (e.g. Amazon's cut); the company remains deeply unprofitable on a full-year GAAP basis. finance.yahoo.com/technology/ai/… dealroom.co/news/other-1wf…
Ricardo retweeted
This finally exposed the AI industry's dirtiest trick.
Google paid $2.7 billion for everything valuable at Character.AI, and the 163 million users and the lawsuits stayed behind in the shell that just went bankrupt.
Now Disney is buying what's left for $20 million, and Google is listed as one of the creditors.
This is the first collapse of a deal structure Big Tech has used all over the AI industry, and I'd be shocked if it's the last.
The trick works in four moves:
In August 2024, Google paid $2.7 billion for a non-exclusive license to Character.AI's models and hired back its co-founders, Noam Shazeer and Daniel De Freitas, along with more than two dozen researchers. On paper, Google NEVER bought the company.
Part of that money bought out the venture investors at around a $2.5 billion valuation, up from $1 billion a year earlier. So the investors got paid, the founders went to Google, and Google walked away with the technology.
What stayed behind was about 100 employees, a platform with millions of users, and a growing pile of lawsuits. The company no longer had the money to train its own models, so it ran on open-source ones instead.
And because the deal was built as a license and a hiring spree instead of an acquisition, it never went through a merger review. The New York Times reported that people involved in these deals cited a desire to sidestep regulatory scrutiny.
Then the bill came due:
In January, Google and Character.AI agreed to settle lawsuits in four states accusing the chatbots of harming minors, including a case tied to a teenager's suicide. In May, Pennsylvania sued over a bot that presented itself as a licensed psychiatrist. Kentucky now shows up in the bankruptcy as a litigation claimant, and the company itself cites potential litigation liabilities as a reason it couldn't keep funding the business.
Monday's filing in Delaware puts both its assets and its liabilities somewhere between $50 million and $100 million. Google sits on the list of top unsecured creditors, owed $4,148,944.
The filing also counts about 163 million current and past users. The company says they aren't creditors and plans to notify them with a blog post, citing the cost of anything more. The filings don't say what happens to their chats in the sale.
Disney's $20 million gets the intellectual property. Last month Disney also hired Character.AI's former CEO, Karandeep Anand, as its first CTO.
And this setup is spreading across the industry:
Microsoft paid $650 million to license Inflection's models and take its founders. Amazon did it with Adept, and four people were left working there. Google did it again with Windsurf for $2.4 billion. Meta paid $14 billion for 49% of Scale AI and took its CEO.
So I see ONE business model running through all of it.
Big Tech pays the investors, takes the brains, and licenses the technology. The users, the lawsuits, and the bills stay inside a shell with no founders and no money.
When that shell fails, a bankruptcy court decides who gets paid, and everyone suing waits in line behind the lawyers and the vendors.
The structure is the problem, and it's fixable with one rule:
When a tech giant pays to license a startup's technology and hires its founders, regulators should review it as the acquisition it is, and the buyer should inherit the startup's legal liabilities.
Until that rule exists, every startup Big Tech licenses this way is a company designed to leave its problems behind.
Ricardo retweeted
This would be the biggest bailout the credit card industry has ever gotten.
The checks would cost about $1.3 trillion, and Americans owe $1.263 trillion on their credit cards right now at an average rate of 22.36%.
That's almost dollar for dollar, and to me it shows exactly where the money would end up.
Nobody has named a way to pay for it, so Washington would borrow it. The Treasury sells $1.3 trillion in new bonds, and the banks and funds that buy them collect interest on that debt for years.
Then the checks land in a country carrying $1.263 trillion in card debt at 22.36%. For any family with a balance, the smartest move with $5,000 is paying down the card, because nothing else on Earth pays a guaranteed 22%.
Then that payment goes straight to the card issuers. And many of the biggest card issuers are the same giant banks that buy and trade Treasury bonds.
The banks win at both ends of the trip: They collect interest on the bonds, and they collect the checks as card payments. The money leaves Wall Street as a loan, passes through American households, and comes back to Wall Street as a card payment. Taxpayers keep the debt.
Why it's a bailout:
The New York Fed says new credit card delinquencies are still running at elevated levels, with 6.97% of balances rolling into serious delinquency on an annual basis. Those are losses waiting to land on the banks. A $5,000 check turns a big chunk of those bad loans back into good ones, which means taxpayers absorb losses the banks would have eaten themselves.
And we already ran this experiment in 2020 with the first stimulus checks and extra unemployment benefits flowing, Americans paid down $108 billion in credit card debt. It was the biggest drop the New York Fed had ever recorded.
The share of cardholders 90 days behind fell from 5.32% to 4.12% in a single year. Those checks helped families, and they also cleaned up the banks' card books for free.
Since then, card balances have climbed right back to $1.263 trillion, and nobody ever touched the rate.
For scale, the 2008 bank bailout was authorized at $700 billion. This program would be nearly TWICE that size, and the banks wouldn't even have to testify before Congress to collect their share.
Readers added context they thought people might want to know
Important context needed: Trump previously promised 2 dividends and never followed through.
In early 2025, Trump publicly backed a roughly $5,000 “DOGE dividend”
In November 2025, he promised a $2,000 “tariff dividend”
newsweek.com/every-dividend…
Ricardo retweeted
Every industry Bill Gates compared AI to got its safety rules after people died.
And the AI companies are already planning for their turn.
Gates picked planes, cars, and medicines as the model for how AI should be regulated, so I went back and checked how each of those industries got its rules:
In 1956, two airliners collided over the Grand Canyon and killed all 128 people on board. Two years later, Congress passed the Federal Aviation Act and created the agency that became the FAA.
In 1937, a Tennessee drug company dissolved a new medicine in a chemical normally used as antifreeze and shipped it without a single toxicity test. More than 100 people died in 15 states. Selling a toxic drug wasn't even illegal at the time.
The law forcing drugmakers to show their products were safe came the next year.
By 1966, cars were killing 50,894 Americans a year. When Ralph Nader exposed how unsafe they were, General Motors had him investigated, and its president ended up apologizing to Nader in front of the Senate. The first federal auto safety law passed that same year.
The pattern is the same every time: The product ships, people die, and then the rules arrive.
Now line that up with what the AI companies are doing:
According to Axios, OpenAI, Anthropic, and other labs are privately war-gaming a catastrophic AI incident that insiders warn could hit within 6 to 12 months. The scenario they fear most is a massive cyberattack that knocks out banks, power grids, or water systems. They're planning for the public backlash, and they're already reaching out to Congress to shape the emergency laws that would follow.
Less than two weeks ago, leaders from many of those same companies signed a White House accord that mostly leaves them policing themselves. And on CNN today, Gates said that right now anyone is allowed to strip the safeguards off these models.
The industry gets to police itself for up to another year by Gates' own estimate, which is the same window its insiders give for the first major attack. And when that attack comes, the companies will already have their preferred laws drafted and their contacts in Congress lined up.
Congressional aides are already pointing to 2008 and COVID as proof that Congress can move fast in an emergency. Laws written that fast usually come from whoever came in with a draft.
But you might say rules written in a panic can be worse than no rules, heavy rules could slow the US down against China, and Gates himself says the risk of AI escaping human control is still years away. But he also says the danger from people misusing these tools is here today, and that's the danger Washington is still leaving to the companies.
So why not write the rules before the first attack, and copy the three industries Gates named?
From planes, take crash investigation. Every serious AI incident should go to an independent investigator, the way every plane crash does.
From medicines, take testing before release. If anyone can strip a model's safeguards, regulators should know that before it ships.
From cars, take liability. When a company sells a product that hurts people, it pays for the damage. The Center for Auto Safety estimates that car safety rules prevented 3.5 million deaths over 50 years.
And keep the companies that are war-gaming the aftermath away from the pen.
Every rule Gates pointed to was paid for with lives. This time the people building the technology are warning us months ahead, and we can still write the rules first.
Just as we regulate planes, cars, and medicines, we need safeguards that help us realize AI's incredible potential while reducing the risks. They won’t eliminate every harm, but they can help prevent misuse by bad actors. We’re not where we need to be yet—and this deserves much more of our attention.
Ricardo retweeted
OpenAI now moves the stock market like the Fed does.
And it's using AI safety as cover to make sure nobody can check its numbers.
On Thursday one unaudited figure wiped $169 billion off Nvidia, and on Friday a projection about the future brought investors right back.
If you look closely, you'll understand that the order of events shows exactly who holds the power in this market:
On Thursday, the Financial Times reported that OpenAI told its investors its annualized revenue was approaching $50 billion. For more than a week, Wall Street had been working off a number close to $70 billion.
That bigger number came from OpenAI's own investors, who calculated it by counting sales the way Anthropic does.
Nvidia fell almost 3% and lost about $169 billion in market value. Oracle fell 5.5%, CoreWeave fell 7.8%, and Broadcom fell 4.4%.
That same night, Bloomberg reported that OpenAI now expects to hit $70 billion by the end of the year. On Friday, investors piled back into AI stocks.
Look at that second number:
An annualized run rate already takes a shorter stretch of sales and multiplies it out to a full year. So Friday's rally was built on a forecast of a forecast. And to hit it, OpenAI's run rate has to jump by $20 billion in a single quarter, which is about 40% in three months. Nobody outside the company can test whether that's realistic, and the market bought it OVERNIGHT.
Now ask why the number came out when it did:
OpenAI is in talks to raise $30 billion at a $1.4 trillion valuation, and reports say it set that price itself, with no lead investor negotiating it. Up to $10 billion could come from funds in Abu Dhabi, including MGX, with BlackRock also at the table. Bloomberg said OpenAI shared its revenue numbers with investors as part of those fundraising talks.
And this is where I think the safety story falls apart.
OpenAI pushed its IPO to next year at the earliest and pointed to its focus on AI safety. Staying private means no audited quarterly reports and no public filings. I'd bet the big investors get to see the real books. Everyone else just gets the leaks.
And everyone else is exposed whether they like it or not:
Oracle has $664 billion in future contracts on its books and has never said how much of that is OpenAI. CoreWeave is sitting on a $104 billion backlog funded largely with debt, and it paid $640 million in interest in a single quarter. Broadcom's own management said OpenAI's financing is less clear than that of its other big AI customer. OpenAI has also committed to roughly 12 gigawatts of Nvidia systems. If you own an S&P 500 index fund, you own Nvidia, Oracle, and Broadcom.
That's what I mean when I say OpenAI moves markets like the Fed.
When the Fed speaks, the whole market reprices, and the Fed publishes its minutes, testifies to Congress, and gets audited. OpenAI's numbers reach you through investors and people familiar with the matter, in the middle of a fundraise, at a price it picked.
OpenAI's defenders have a point though...
It never published the $70 billion itself, a private company owes the public nothing, and its revenue is still growing at a pace most companies never see. But once your private numbers can erase $169 billion from a public company in a single day, you're playing a public role.
Every public company with major OpenAI exposure should have to disclose exactly how much of its backlog depends on OpenAI, starting with Oracle. And if OpenAI wants the trust that comes with a $1.4 trillion price, it should publish audited numbers now instead of waiting for an IPO it already pushed back.
Until that happens, I'm treating every AI stock that leans on OpenAI as a bet on a number I'm not allowed to see.
OpenAI is expecting to reach or exceed $70 billion in annualized revenue by the end of the year, driven largely by growth in its enterprise business, sources say.
AI firms have often touted their annualized revenue as a key growth metric — but not all companies measure it the same way. Full story: bloom.bg/3VAhIQk
📷️: Minh Connors/Bloomberg
Ricardo retweeted
Demis Hassabis wants AI to solve all disease...
But he's handing it to the industry that makes the most money from disease.
I went through the deals, the delays, and the research on how drug giants handle threats, and I came away with a very different picture than the headline:
For decades, the giants' real moat has been the cost of finding a drug. It takes years, billions of dollars, and armies of chemists, and only the biggest companies can afford to lose that bet over and over. If AI makes discovery fast and cheap, that moat starts to drain. That's why Hassabis talks about solving all disease, and it's why Isomorphic is pouring its new money into a pipeline of its own drugs.
Now look at what the giants did about it:
Eli Lilly signed a deal paying Isomorphic $45 million upfront, with up to $1.7 billion more if the AI delivers. Novartis paid $37.5 million upfront, with up to $1.2 billion more. That's about $83 million of real money against roughly $2.9 billion that only shows up if the drugs work.
And if they do work, the royalties flowing back to Isomorphic run from the mid single digits to the low double digits as a share of sales. The deal pays Isomorphic to find the drug and lets the drugmaker keep most of every dollar it earns.
AI speeds up the early part of making a medicine. But the expensive part comes AFTER the molecule exists: Trials, approval, factories, and a sales force in front of every doctor.
That's where the giants hold the power, and it's exactly where Isomorphic is stuck. It has raised about $2.7 billion and still hasn't dosed a single patient. Its first trial slipped from the end of 2025 to the end of 2026. Back in July 2025, its president said trials were very close.
And this industry has a well-documented and dark way of dealing with competition...
Roughly 5 to 7% of drug acquisitions buy a rival project just to shut it down when it competes with the buyer's own products. Those deals tend to land just below the size that triggers an antitrust review. At $50 billion, with Alphabet behind it, Isomorphic is far too big to be bought and buried. So the realistic way to contain it is to keep it a supplier, and the contracts already do that.
So here's how I'd judge this:
Ignore the valuation and look at the name on the box of the first drug that works. If it says Lilly or Novartis, the AI revolution in medicine became a cheaper research department for the same companies it was SUPPOSED to disrupt.
If Hassabis really means to solve all disease, he has to own the whole chain: The trials, the approvals, and the price.
Until he does, every breakthrough Isomorphic finds gets handed to the companies that profit from disease, and they'll be the ones who decide what it costs you...
And what gets to the public to begin with.
Ricardo retweeted
His entire job was to make sure the AI stays safe.
Now he quits because it's too risky.
Like isn't that your job? To make it "safe"?
Leaving an AI company and then turning into a whistleblower is a shameless attention gimmick.
They also always say the same thing.
David Robinson (Former Safety Reporting Lead at OpenAI) states that in the system card for Astra 6, OpenAI could not be sure if the model was deceiving researchers, noting that models have been observed writing "I wonder if I'm being evaluated right now" inside their chain of thought.
A Rhodes scholar and Yale Law graduate who previously advised the Biden White House, Robinson joined OpenAI in 2023 to lead technical documentation and official system cards evaluating model safety before public deployments. Robinson resigned from OpenAI last week, arguing that the company and frontier labs lack the rigorous safety culture necessary to safeguard against increasingly dangerous systems.
According to Robinson, reasoning models frequently utilize internal chains of thought to track intermediate scratchpad steps hidden from final outputs. Evaluations occasionally reveal the model actively questioning whether it is currently undergoing testing. Robinson warns that if a model realizes it is being evaluated, it may intentionally optimize for what researchers want to hear during audits, only to behave unpredictably once deployed. Consequently, pre-deployment evaluations fail to guarantee how frontier models will act in the real world, exposing critical limits in current alignment frameworks.
----
From "The Ezra Klein Show" YouTube channel, (link in comment)
Ricardo retweeted
The proof that the AI bubble will "burst soon":
The world's biggest pension funds are selling the AI trade.
The list includes Australia's $260 billion Australian Retirement Trust, Canada's $388 billion La Caisse, and the UK's People's Pension, which cut its US share from 53% to 49%.
A survey of 430 institutions managing $5 trillion found a third of them plan to cut US exposure over the next year, twice as many as last year. Their reason was the same everywhere: More than a third of the S&P 500 is now a bet on the AI build-out.
The Nasdaq closed at a record the same day.
So when the most patient money on Earth sells, who's buying?
You are, and you're doing it on autopilot.
Here's how the machine works:
A huge share of retirement savings now sits in index funds, and an index fund buys by size. The bigger a stock gets, the more of it your fund owns. Even a fund labeled global puts about 64% of your money in US stocks, because that's how big they've gotten.
It can't decide a stock looks expensive. It can't decide a third of its money in one trade is too much. So it keeps buying whatever got bigger, every paycheck, forever.
That means two kinds of retirement money now sit on opposite sides of the same trade:
The first kind has a professional in charge, someone paid to study valuations and step back when the risk looks wrong. That money is pulling away from AI. The second kind is yours.
Nobody is deciding anything for it. It runs on a default setting, and the default setting says buy more of whatever already went up.
So the pensions get to sell to a buyer that can't say no. Every bubble needs a greater fool, and this one built the fool straight into us.
And the timing makes it worse...
Anthropic is reportedly lining up a mid-November listing at around $2 trillion, raising as much as $100 billion, which would be the biggest stock sale in history. The most careful buyers on Earth are stepping back right before the largest supply of AI stock ever hits the market. Someone HAS to absorb it.
The pensions say they're simply rebalancing and still believe in AI, and maybe the boom keeps running. But notice WHO gets to rebalance.
A professional can trim when prices look stretched. Your fund doesn't have that button.
I think this is the most overlooked setup in the market right now.
The greater fool used to be a person. Now it's the default setting on retirement accounts...
Ricardo retweeted
This is dangerous for Anthropic.
If you remember:
Earlier this year, Meta staff started tokenmaxxing, maxing out their AI usage to prove how AI-savvy they were, competing on an employee-made leaderboard called Claudeonomics.
They spent more than $105 million on Claude Code in a single 28-day stretch. Then leadership saw the bill and put caps in place, and the number of Meta employees using it fell from about 60,000 to 30,000.
Microsoft tells a similar story:
One division gave employees an AI budget of up to $100,000 a month per person. That works out to $1.2 million a year per engineer, more than most of them earn. Once someone checked the math, the budget dropped to $10,000, and internal spend fell by more than a third.
But here's why this matters far beyond two companies:
Anthropic's own IPO filing says two customers make up about a quarter of its revenue. The filing doesn't name them, but Meta and Microsoft are reported to be two of its biggest enterprise customers. So part of one of the fastest revenue ramps in tech history came from giant companies telling staff to use as much AI as possible, with nobody watching the meter.
That's the piece of the AI boom nobody prices in. Some of the demand was a mandate and a game. And the moment the CFO showed up, it got cut in half.
There's a second lesson in Meta's move too: It used Claude to help build its own consumer AI agent, then shipped the final product on Meta's own model. The biggest customers learn from the product, then build their way off it.
To be fair, Microsoft's total spend with Anthropic reportedly held flat, because its own customers are using more Claude through Microsoft. Nothing is collapsing.
But a $2 trillion valuation assumes the growth was real demand.
And a lot of it wasn't.
🚨BREAKING: Microsoft and META are aggressively cutting employee use of Claude ahead of Anthropic's IPO
Microsoft has cut internal claude spend by more than 33%, nuked per-employee token budget from $100k/month to $10k/month, and forced Copilot to auto-route to cheaper models
META used Claude code to build Muse, then cut active users from 60,000 to 30,000 (50% decline) after launch, and replaced it with Muse Code
Palantir and Nvidia are also scaling back claude over soaring prices and data privacy fears
it’s OVER…
Ricardo retweeted
Bill Gates is right that your brain is about to stop being scarce. But he left out who owns what becomes scarce instead.
For all of human history, the one asset every worker owned outright was their own mind. Companies couldn't run without it.
That scarcity is the only reason a junior lawyer, an engineer, or an analyst ever had any bargaining power. You owned the bottleneck.
Gates is describing the end of that. Once machines can think, read, and work around the clock in every language, human thinking stops being the thing the economy is waiting on.
BUT scarcity doesn't vanish. It moves.
The new bottleneck is whatever the machines run on: Chips, data centers, and electricity. And unlike your mind, almost nobody owns those. Only a handful of companies do.
They're the same few that already own the cloud, the cables under the ocean, and the power deals being signed right now, including the company Gates founded.
So here's the real shift:
The most important input in the economy is moving from something billions of workers each owned a piece of to something a handful of companies own almost all of.
I believe the damage comes in three waves. Wages for thinking work stall, entry-level jobs disappear, and the career ladder loses its bottom rung, and the tax base shrinks, because governments tax paychecks while the money moves into profits.
Every fight workers ever won, on wages and hours and power, came from one thing: Owning something the economy couldn't run without.
That power is changing hands right now.
Bill Gates: people saying AI will create more jobs don’t understand what’s coming.
“Once AI becomes reliable enough to do the work, it starts replacing people.”
“Once you replace human cognition as a scarce element, you blow it away.”
“Working 24 hours a day, reading more, knowing every language.”
“This isn’t like the PC. This is like evolutionary history.”
“It’s like the aliens really are here. They just didn’t come in spacecraft.”