Oct 24, 2025 · 54m · more-or-less
TPUs Are Winning? Gemini + Anthropic vs NVIDIA, and OpenAI’s Atlas Explained · More or Less Podcast
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this episode of More or Less, tech investors and media executives Jessica Lessin, Brit Morin, Dave Morin, and Sam Lessin analyze the escalating capital costs and chip rivalries driving the AI boom. They debate the workplace degradation caused by 'AI slop' and rogue meeting bots while examining the shifting cultural premium placed on authentic human effort.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 99.9% of the talking time here. How this is scored →
The hosts as informed peer 5.2 Guest teaching 3.4 Guest disagreement 4.0 The hosts pushing back 3.5
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 39:08 Sam dismisses the ongoing AI slop debate as completely boring
Sam outright rejects the premise and relevance of the discussion, bluntly declaring that everyone has access to the technology and that debating work slop is boring.
Hardest push from the hosts ▶ 18:23 Jessica rejects Sam's claim that compute expenditures are negligibleJessica forcefully refuses Sam's framing that OpenAI's product experiments are minor rounding errors, insisting that the company is spending vast amounts of money it does not have.
Biggest teaching moment ▶ 35:12 Dave explains scientific findings on cognitive overload from AI modelsDave educates the panel on cognitive overload papers, explaining how AI generation tricks users into believing they had breakthrough thoughts when they cannot actually explain them.
The host holds their own ▶ 21:35 Jessica breaks down TPU economics and OpenAI's 45 billion dollar hurdleJessica showcases deep industry domain knowledge, outlining the exact infrastructure cost per gigawatt and explaining why alternative silicon threatens NVIDIA's margin structure.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Opening Title Sequence and Theme Animation | 4 | 3 | 7 | 4 | The panel immediately jumps into a heated argument over AI work slop. Sam bluntly categorizes Brit's workflow as lowbrow imitation of machine technology, prompting Brit to strongly defend her creative output. | |
| The Yasso Bar Interlude: Sam's Frozen Treat Habit | 3 | 2 | 2 | 2 | The group pivots to lighthearted banter regarding Sam's compulsive habit of eating Yasso bars and an anecdote about an LP using AI to draft an iPhone-signed thank-you note. | |
| Rogue Zoom Bots and Meeting Etiquette | 3 | 3 | 2 | 2 | Dave describes discovering that his automated Zoom bot joined team meetings uninvited. The co-hosts discuss the emerging etiquette around leaving AI note-takers running in conference calls. | |
| Testing OpenAI's Atlas Browser and Platform Strategies | 7 | 4 | 3 | 3 | The group analyzes OpenAI's Atlas browser release. Dave examines Chromium extensions and lack of differentiated utility, while Jessica demonstrates deep industry background by comparing it to her scoop breaking the original Google Chrome launch. | |
| The Fading Sora Hype and OpenAI's Ballooning Compute Costs | 6 | 3 | 5 | 6 | Sam declares Sora's cultural moment over and brushes off its operational cost as a rounding error. Jessica pushes back directly, noting OpenAI is burning immense amounts of capital it does not possess. | |
| TPUs vs. NVIDIA, $45B Compute Bills, and the Macro Wager | 9 | 4 | 5 | 5 | Jessica leads an in-depth discussion on TPUs versus NVIDIA GPUs, detailing her reporting on OpenAI's TPU talks, Jensen Huang's reaction, and the staggering 45 billion dollar compute commitments. Sam contextualizes the spending through Pascal's Wager and macroeconomic pressures. | |
| The AWS Outage: Smart Home Chaos and Infrastructure Fragility | 4 | 3 | 2 | 2 | The co-hosts talk through the AWS outage that impacted Eight Sleep mattresses and smart-home routines. The conversation turns to historic architectural vulnerabilities like Twitter running off a single Mac mini. | |
| Market Liquidation: Crypto, Gold, and the Danger of Leverage | 4 | 3 | 3 | 3 | Dave asks about sudden market liquidations across crypto and gold. Sam explains how excessive trader leverage creates violent short-term swings that long-term fundamental investors ignore. | |
| Revisiting AI Slop: Cognitive Overload and Counter-Culture Cycles | 4 | 6 | 8 | 4 | Dave shares research on cognitive overload from rapid AI generation, while Brit defends her detailed editing workflow. Sam abruptly shuts down the conversation as boring and explains how cultural cycles always move toward rejecting easily automated output. | |
| The Death of Conference Swag and Bespoke Gifting | 8 | 3 | 3 | 4 | Jessica introduces her thesis that generic corporate swag is dead in favor of ultra-bespoke gifting. She shares her experience curating conference merchandise and picking a personalized book at an airport bookstore. |
Statements from this episode (24)
Sam Lessin: Lowbrow Culture Imitates Machines While Highbrow Rejects Them
“What lowbrow people do is they start talking like the machines. Like, they start trying, like, so there's a whole genre of people who are speaking like LLMs. They're imitating the fucking technology, and like, that is what lowbrow people do. What highbrow peop…”
Sam Lessin: No Thinking Person in 2025 Uses AI to Write Speeches
“Which is why, for instance, remember, there was a brief moment where everyone thought the AI's written speeches for dinners was like the cutest thing ever, and no, no thinking person would use AI to write a speech in twenty-twenty-five because it's just a lowb…”
Brit Morin: AI outlining followed by human editing is not AI slop
“Just because I use AI to outline my thoughts and like a five pronged approach, then I use real brain to go and edit all the copy that it had previously like template generated for me. Doesn't mean it's AI work slop”
Dave Morin: I routinely ask senders how much AI they used
“No, I will just respond to people now and ask how much AI is in this thing that you just sent me.”
Sam Lessin says he frequently walks out of meetings
“I'm a big fan of just leaving meetings. I do that frequently.”
Brit Morin: OpenAI's Atlas browser delivers a far worse ChatGPT experience
“I signed in with my OpenAI account, obviously, and then I started using it like I would have used ChatGPT in my Google Chrome browser, and it was hallucinating way more. It didn't have like a lot of memory stored about all the things I had talked to it about. …”
Dave Morin: OpenAI's Atlas browser adds little value over a separate window
“I can't describe to you what the use case is of how the browser works. Other than that, it shows you ChatGPT in the window to the right, and then you can click like a single button to summarize an article. But all that does is pull the current URL into it. It …”
Lessin: OpenAI will not match Grok on erotica-friendly content
“And I have a hard time imagining that like open AI will be willing to go where like Grok will be willing to go with it effectively. And the middle ground will just be weird.”
Dave Morin: OpenAI's Atlas browser is primarily about building memory
“Yeah, this is about building up memory. It's about building memory. It's like very clear. I mean, that is the one thing that I read.”
Sam Lessin: OpenAI's Sora is not going to work
“That, like, that was, like I mean, I understand it's still probably number one in the free app store or something, but, like, it's not gonna work.”
Brit Morin: OpenAI will improve Sora and regain user engagement
“I bet they add improvements, though, and it gets like another, they get another hook on the line.”
Sam Lessin: Surplus product managers drive companies to launch excess products
“They need, they have so many product managers, they're hiring all the product people, and then you give them things to do, or they get sad, and so you end up doing more things because you've hired all the people, and you can't have them do nothing, and they al…”
Lessin: Gemini and Anthropic Are Trained on TPUs Rather Than GPUs
“If that's true, then the top two coding models Gemini and Anthropic are trained on TPUs, not GPUs. And I think that's a really big deal.”
Lessin: Custom AI Chips Do Not Need to Match NVIDIA's Performance
“They make a mistake thinking that a TPU or a Tranium or from Amazon or something a company develops itself has to be as good as NVIDIA. It doesn't because NVIDIA's margin is so big. It can actually be a fraction of as good as NVIDIA and still cost effective.”
Sam Lessin: AI Customer Service Savings Will Be Competed Away
“If everyone has the same improved customer service, it just ratchets the competition up. Right. And those numbers get passed back to you. Don't get to just like take all the money that we spend on customer service and be like, well, now we get that money. That…”
Sam Lessin: Meta Can Absorb AI Capex Failure; OpenAI and NVIDIA Cannot
“Like, I'd argue the cost for meta of belief isn't that high. No one really cares. They can fork over a few years of cash flow, right? Like, investors will look the other way. They'll kind of be like an Oculus. They'll be mad for a minute, but get over it, righ…”
Lessin: Macro Incentives Will Sustain Aggressive AI Capex for Another Year
“I also think there have been all these sort of incentives and macroeconomic things that have encouraged the spending, and I think they'll probably go for another year, if you think of, like, the midterms.”
Lessin: Markets ignore one-time outages, caring only about systemic risks
“The whole thing about one-time hits, they don't really matter. Like anyone, it's like, I think this is the same thing with like, these outages, a lot of things is like, systemic things people care about. One-time hits, it doesn't, whether you're spending a hun…”
Lessin: X's algorithm no longer heavily promotes photos
“The feed ranking has changed from when they used to promote the shit out of photos.”
Sam Lessin: MrBeast's content is 'manual slop'
“Well, I, I'd have to pull it up, but I was calling Mr. Beast's content manual slop.”
Dave Morin: Scientific papers show AI models cause cognitive overload
“In fact, there's a bunch of science people are, I think there's, I'll have to go find it, but there's been a few papers produced about the sort of cognitive overload that these models are creating because they create so much content so quickly and sort of lead…”
Sam Lessin: AI text expansion and summarization have net-zero impact
“You basically have this, like, war of attrition Where you have all these people making more, and then everyone's like, on the other side, like, making less, and the net impact of it is zero, right?”