The most rational take on AI you’ll hear this year

Key points

• **The whole thesis in one sentence — 'as big as the Internet or mobile, and *only* as big.'** *'My most controversial opinion is that I think that AI is as big a deal as the Internet or mobile and only as big a deal as the Internet or mobile.'* The deliberate ceiling is the point: it reframes the entire cohort's framing of AI as civilisational-singularity down to 'largest-platform-shift-of-a-generation', which carries very different valuation maths. **This is the explicit counter-altitude to [Laffont's '$4T AI IPO wave … we've never seen anything like it'](/issues/2026-06-07) in the same week's All-In.**
• **'We're in 1997' — radical uncertainty is the honest base case.** *'If you're going to make the Internet comparison, it's like we're in 1997. Like, it's very exciting. Most stuff kind of doesn't work yet. Most of the stuff that people are going to do hasn't been built yet. And it's not really clear how any of it's going to work when it does work.'* His own deck, by his account, is *'80 slides of saying we don't know.'* **For an investor this is the calibration note**: picking the winning model today is *'like being in 1997 and saying, well is it going to be Excite or Yahoo'* — the answer was generally neither.
• **The load-bearing trade: model labs commoditise, value moves up the stack.** *'Why would the model companies have pricing power and wouldn't all the value be further up the stack? Aren't you basically have you got like three to six companies selling a commodity at marginal cost.'* The mechanism is the absence of network effects: *'the models don't seem to have network effects, so there doesn't seem to be a winner takes all effect where one of these will run away ahead of the other. So you should have competition indefinitely.'* **This is the bear lens on the exact Anthropic/OpenAI ARR race the [20VC roundtable](/issues/2026-06-07) and [OpenAI CFO Sarah Friar](/issues/2026-06-07) episodes are pricing as a duopoly land-grab** — and it directly extends last issue's pricing-degradation forecast on Claude's 2x premium [/issues/2026-05-31].
• **'More like cloud than like Windows' — the structural analogy that decides who wins.** *'It should end up looking more like cloud than it looks like Windows.'* Windows had lock-in (developers standardised because customers did); AWS does not (*'an engineering company or a law firm buying a piece of software, you don't care which side it runs on'*). If the chatbot isn't the UX and the apps must be built by thousands of other companies, the labs are *'undifferentiated commodity infrastructure providers.'* **This is the same 'own the layer that captures volume regardless of which model wins' logic as last issue's [Gerstner token-flow / data-infra trade](/issues/2026-05-31)** — arrived at from first principles rather than a tape read.
• **The utility-margin reality check — telecoms as the cautionary comp.** On Altman's plan to *'be selling AI intelligence on a meter like water or electricity'*, Evans: *'my dear sweet child, you need me to explain the margin structure of the utility industry to you.'* The proof is mobile: mobile data consumption *'it's about, you know, 1500 to 2000 times what it was in 2010 globally'*, yet *'the stocks have gone nowhere in 25 years because it's an ex growth, low margin commodity utility.'* **The read-through to the AI-capex/ROI reckoning is brutal**: enormous volume growth and zero equity return can coexist when the layer is commodity infrastructure.
• **Today's economics are a temporary disequilibrium, not the steady state.** On the headline token bills — *'the open claw guy spent one and a half million dollars on tokens last month'* — his framing is *'that's like somebody getting like a 50 grand mobile data bill in 2010 that's temporary'* … *'what is the steady state equilibrium point where all of these lines, the lines on the chart kind of get lined up'* **This is the calibration knife for every '$X on tokens' data point in the cohort** (the Fortune-20 $200M, the Uber-COO burn-rate from [last issue](/issues/2026-05-31)): current spend over-states durable revenue because pricing is in radical disequilibrium.
• **Distribution — not the model — is the moat once the product is a commodity.** *'distribution of an adequate product, when the field is basically commodity distribution on brand become a big deal.'* Hence Google spraying Gemini everywhere and *'the model is just like the dumb thing underneath'* that powers the feature. And the durable incumbents' instinct: per Sinofsky, *'incumbents always try and make the new thing a feature. And sometimes they're right, sometimes it's a feature.'* **Direct read-through to the application-layer-defensibility debate in this week's [Mercor CEO](/issues/2026-06-07) and [No Priors full-stack-builder](/issues/2026-06-07) episodes**: if models are interchangeable, the contest is distribution and harness, not weights.
• **Why the labs keep hiring consultancies — the task is not the job.** The surprising trend (labs buying into McKinsey/Accenture-style services) resolves through one question: *'is the hard part of the job writing the code line by line'* … *'or is the hard part of the job something else?'* As he puts it, *'Claude co can write you the code, but what code do you want?'* And on the Jevons/price-elasticity history: *'before Excel, junior investment bankers worked really long hours. And now, thanks to Excel, Goldman's associates, all the work at lunchtime on Fridays'* — except, of course, they don't; cheaper inputs expanded the work. **The CRE-lawyer-relevant point: automating the deliverable doesn't automate the engagement.**
• **The jobs-apocalypse is 'horseshit' — and argument-from-authority gets no pass.** *'You can't kind of look at a senior partner at a law firm and say, well, 17% of their work could be automated. Like, this is horseshit.'* On Dario specifically: *'I'm interested in Dario's opinions on where models are going to go in the next six to 12 months. Not particularly interested in his opinions on theories of labor and market value and competitive comparative advantage.'* **The direct counter to the entry-level-jobs-wipeout narrative** — and a tidy inversion of the lab-CEO macro calls that anchor much of the cohort. The enterprise-reality brake: a typical *'Enterprise software sales cycle is like 18 months if you're lucky'* — and *'the enterprise sales cycle is shorter than the venture backed startup funding cycle. Longer, rather longer.'*
• **The floor under the bear case — even a brick wall today is transformational.** *'Even if the model stopped at getting better tomorrow, if this is it, and we hit a brick wall tomorrow, this is an incredibly useful technology that's going to change the world'* over the next 10 years; *'so you don't have to believe in any of that stuff to believe that this is a giant deal.'* **This is the steel-man that keeps the commoditisation thesis from reading as AI-skepticism**: the call is on where margin accrues, not on whether the technology is real. The deck's self-aware tagline: *'this is going to be completely different from everything else, just like everything else.'*
• **The actionable advice — and the one-line risk to social licence.** For careers: *'don't stick your head in the sand and say, I hate all of this stuff'* … *'What helps is you diving into this, completely submerging yourself in it and coming out, understanding what you can do with it.'* The underpriced macro risk he flags is the AI backlash — *'AI is like less popular than ice'* — though he debunks the water panic specifically (US data-centre water use *'came out at about 0.017% of US water consumption'*). **Same legitimacy-deficit thread as last issue's [Lonsdale ~30%-approval read](/issues/2026-05-31)** — Evans treats it as a real-but-fuzzy social-media-style backlash, not a financial catalyst yet.

Notable quotes

Themes

• AI hype de-escalation (internet/mobile, not singularity)
• Model-layer commoditisation
• Value capture up the application stack
• Token-economics disequilibrium vs steady state
• AI valuation / bubble debate