Overview of Big Technology Podcast with Alex Kantrowitz
This Friday edition covers three main AI-world flashpoints: Anthropic’s still-explosive but slowing growth ahead of a possible IPO, a reported espionage case inside OpenAI, and a lively debate over AI-generated “food slop” images on restaurant signs and menus. The discussion also touches on the broader market pressure hitting AI infrastructure companies, especially NeoClouds like CoreWeave, and the rising cost of capital that may be exposing weaker parts of the AI boom.
Anthropic IPO Panic: Huge Growth, but the Rate Is Slowing
Anthropic reportedly told potential investors that its Q3 revenue reached $16.8 billion, up about 45% from Q2 and more than 3.6x its 2025 full-year revenue. That is still extraordinary growth, but the market reaction was nervous because Q2 had grown more than 140% quarter over quarter, driven heavily by the Claude Code surge.
Why investors are spooked
- Growth is still massive, but the acceleration has clearly cooled.
- Some market watchers expected Anthropic to justify a $2 trillion IPO valuation, but the new numbers suggest that could be too rich.
- The hosts argue that a more realistic valuation might be closer to $1.5 trillion if the slowdown continues.
Important wrinkle: distillation revenue
Anthropic also reportedly said that 5%–7% of revenue came from “distillation” activity, where outside companies query Claude heavily to train competing models. Anthropic is now cracking down on this behavior, which likely means some revenue will disappear, adding to the deceleration.
Broader takeaway
The conversation centers on whether this is:
- a temporary cooling after an unusually hot Claude Code moment, or
- evidence that Anthropic’s growth was more of a spike than a sustained exponential curve.
The hosts repeatedly stress that the episode reflects a Rorschach test: bears see a slowdown; bulls see unprecedented growth from an already huge base.
The IPO Process Feels Messy and Leak-Driven
A major theme in the Anthropic discussion is frustration with how information is being released.
Key concerns raised
- Anthropic is reportedly sharing financial details with large investors before a formal S-1 filing.
- Leaks to the press and conflicting reporting are creating confusion about the company’s true trajectory.
- The hosts argue that this is an unusually chaotic way for a major public offering to take shape.
Why the S-1 matters
Ranjan Roy insists the market needs the actual filing because it will clarify:
- revenue quality,
- profitability,
- partner costs,
- training expenses,
- and the real economics of the business.
The broader point: everyone is trying to price Anthropic as if it’s already public, but the company’s final business model is still unclear.
NeoCloud Selloff: Debt, Rates, and Bubble Fears
The episode then turns to the NeoCloud trade — companies building AI data center and compute infrastructure — and why the market has turned bearish on them.
What’s happening
- CoreWeave, Nebius, and IREN all sold off sharply.
- Options traders piled into bearish bets on CoreWeave.
- CoreWeave’s bonds, issued at 9.625%, were reportedly trading at yields around 15%–16%, implying a steep drop in price and rising distress concerns.
Why the market is worried
The hosts connect the selloff to:
- higher interest rates
- rising borrowing costs
- massive debt loads
- doubts about whether the infrastructure buildout can generate enough returns
Ranjan’s point is that capital is no longer being treated like it’s infinite; the market is starting to price in the cost of financing all this AI buildout.
Firmus as an example
Ranjan also highlights a striking story about Firmus, an Australian NeoCloud with:
- major Nvidia involvement,
- a proposed IPO at a huge valuation,
- origins as a Bitcoin mining company,
- and a messy founder history.
The story is used to illustrate how speculative and fragile parts of the AI infrastructure boom can be.
OpenAI Espionage Allegation: A Hidden Recorder in a Conference Room
The next major segment covers a reported case in which an OpenAI employee allegedly hid a small recording device in an executive conference room.
What the report says
- The employee allegedly wrapped a microphone/circuit board in toilet paper
- and inserted it into the USB port behind a wall-mounted TV.
- The room was used for leadership meetings, government briefings, and sensitive product discussions.
- Prosecutors say he used or attempted to use a device to intercept oral communications.
What is and isn’t known
- There is speculation about possible foreign-state involvement, but no definitive proof is provided in the transcript.
- The employee’s lawyer argues this is a recording-statute case, not necessarily a foreign-agent or trade-secret case.
Why it matters
The larger takeaway is that Silicon Valley AI labs now assume:
- espionage attempts are plausible,
- insiders may be targeted,
- and sensitive information can be stolen with very small, cheap devices.
The hosts note that similar warnings are reportedly being given at Anthropic, Google, and other AI companies.
How AI Makes Spycraft Easier
One of the funniest and most unsettling threads in the conversation is how cheap and accessible this kind of spying has become.
What changed
In the past, covert recording devices required serious technical expertise or spy-agency resources. Now:
- an off-the-shelf chipboard can be bought online,
- and a chatbot can help someone figure out how to assemble it.
That makes amateur espionage much more feasible — and potentially more common.
Related “honeypot” warnings
The hosts also talk about classic honeypot tactics:
- strangers trying to befriend employees,
- people asking about work in casual settings,
- romantic or sexual setups used to extract information.
The point is that AI companies are now training staff to assume that even innocuous interactions may be intelligence-gathering attempts.
Quick Take on Apple’s New Home Hub Device
A shorter segment discusses Apple’s rumored tabletop smart display / Home Hub.
Reaction
Ranjan is unimpressed, arguing that:
- Siri still isn’t good enough,
- the device feels limited,
- and the UI seems awkward for things like cooking if you still need to touch the screen repeatedly.
The hosts agree that the hardware makes more sense if Apple can dramatically improve Siri and make the product genuinely useful.
AI Food Slop Debate: Ugly, Or Beautiful Democratization?
The episode ends with a fun argument about AI-generated food images on restaurant signs and menus.
Alex’s view
AI food art often looks:
- uncanny,
- repetitive,
- and sometimes just plain gross.
Ranjan’s view
He strongly defends it as a form of creative empowerment:
- small business owners can make their own signage cheaply,
- no need to hire a designer,
- and the visuals are part of the messy, democratic energy of the AI age.
Their compromise
Alex eventually comes around to the idea that the problem is less AI itself and more that:
- the outputs look too similar,
- and the models should include more variation or randomness.
They close with examples that show how AI signage can be absurd but effective — including a sign advertising a spa where the AI apparently generated a sign showing a sign in front of a spa.
Main Takeaways
- Anthropic is still growing extremely fast, but the slowdown from Q2 to Q3 is enough to unsettle IPO watchers.
- The biggest question is not just revenue growth, but how durable that growth is and what the business model ultimately becomes.
- NeoClouds are under pressure as higher rates and rising borrowing costs make their debt-fueled buildout look riskier.
- The OpenAI espionage case shows that AI labs are now serious intelligence targets, and the tools needed for spying are becoming cheaper and easier to use.
- The food-slop debate captures a broader truth about AI: ugly outputs can still be economically useful, especially for small businesses.
Notable Line
- Ranjan’s repeated refrain on Anthropic: “Release the damn S-1.”