The David Lin Report · · 37 min

Biggest Economic Disruption Ever: Why AI Revolution Has Barely Begun | Ethan Mollick

With Ethan Mollick

Summary

Ethan Mollick, a Wharton professor and author, discusses how AI is reshaping work, startups, markets and infrastructure. He argues AI already outperforms many junior analysts on tasks like stock advice and company research, though firm-specific judgement, internal politics and training pipelines still favour humans, and labour-market effects remain early and unclear. He tracks rapid gains in model task length, from about four minutes of independent work for GPT-4 in 2024 to roughly 17 weeks now, while noting models remain jagged. On investment, he finds Anthropic hard to price because its growth is high but its R&D costs are heavy and the market has not priced in an exponential, and he argues the AI compute buildout rests on a bet that token demand is nearly unlimited. He expects data center development to continue despite local opposition and environmental pledges, and sees AI lowering startup costs while a billion-dollar company still needs a billion-dollar idea.

Key takeaways

  • Ethan Mollick on Venture capital & startups: AI lets founders scale with fewer hires, but a one-person billion-dollar company remains unlikely.
  • Ethan Mollick on Frontier AI labs & model race: Frontier model capability is compounding exponentially, with recent models running ahead of trend.
  • Ethan Mollick on AI & jobs: AI already outperforms many first-year analysts on tasks, though junior staff still do work AI cannot.
  • Ethan Mollick (uncertain on Anthropic): Anthropic's $1.4T IPO valuation is very hard to price; it is an option on an imagined future.
  • Ethan Mollick on Enterprise AI adoption & AI software: Companies are seeing positive AI ROI but have only started reorganising around AI; productivity gains not yet visible.

What was said, by company and theme

10 views from 1 speaker. Every quote is checked against the episode's transcript (how).

Anthropic

Across the shows: 119 speakers in other episodes, 177 bullish and 90 bearish calls, past 30 days

Where everyone stands on Anthropic

  • Ethan Mollick · Wharton School, University of PennsylvaniaUncertain

    Anthropic's $1.4T IPO valuation is very hard to price; it is an option on an imagined future.

    Mollick says investors buy a company for today and also buy an option on the future world it imagines. Anthropic has very high growth but weak and strong spots in its numbers, and many of its costs go to R&D and building new models. It could cut costs and become profitable, but if it slows R&D, open-weight models might catch up and erode the value. Ordinary methods do not fit: it could be priced as a fast-growing software company, or as a company aiming to replace human labour in a few years. He says he has trouble pricing it because the market has not priced in an exponential, and we are not used to pricing industrial revolutions.

    “On one hand, you're buying a company for today. On the other hand, you're kind of buying an option on the future world they imagine.”at 20:57

AI infrastructure & data centers

Where everyone stands on AI infrastructure & data centers

  • Ethan Mollick · Wharton School, University of Pennsylvania

    Massive AI infrastructure spending is a bet on near-unlimited token demand, which is hard to price.

    Mollick says AI companies are betting that AI is the biggest technology ever, that it will reshape how industries work, and that demand for tokens is almost unlimited. The bet assumes they stay compute-constrained for years and can sell every token and FLOP they produce. Mollick says there are no good historical analogues for a product whose utility keeps rising. The bet is either an insane bet on a future that does not exist yet, or a correct call that AI changes everything, and the missing piece is whether enough tokens can be produced to meet demand. He notes that the same problem of assuming the past resembles the future applies to finance.

    “They bet we will be token constrained, compute constrained for the foreseeable future... It is either an absolutely insane bet on a future that doesn't exist yet or it's like, yeah, AI is going to change everything.”at 24:41
  • Ethan Mollick · Wharton School, University of Pennsylvania

    Local data center opposition is partly real but will swing, with more complex community-benefit deals ahead.

    Mollick says voter opposition has several causes. Some are real: data centers are light industrial complexes, they are noisy and use a lot of electricity, and water use is less of an issue than people think. Others come from a wider backlash against AI, which the AI companies have fuelled by saying AI will replace jobs or even kill people. Concentration of power also worries people. But he argues the backlash will swing back and forth because many communities will embrace data centers for large payments and lower property taxes. Loudoun County, one of the richest counties in America, is funded by data centers. He expects more complicated deals with community involvement, union labour and environmental pledges.

    “I think you'll see this backlash swing back and forth because a lot of communities are going to embrace this because you can get a $100,000,000 for your community and ... property taxes.”at 33:19
  • Ethan Mollick · Wharton School, University of Pennsylvania

    Environmental pledges may slow data center development somewhat, but will not stop it.

    Mollick argues that data center development will continue despite environmental pledges. He concedes that pledges may slow it somewhat, but says he doesn't think development will stop. He also says investors should separate individual AI-linked stocks from what is happening with AI overall, and that a neocloud funding decision could carry different implications. The case is brief and not backed by specific numbers or data.

    “And that may slow development somewhat, but I don't think you're gonna see data center development stop.”at 34:02

AI & jobs

Where everyone stands on AI & jobs

  • Ethan Mollick · Wharton School, University of Pennsylvania

    AI already outperforms many first-year analysts on tasks, though junior staff still do work AI cannot.

    Mollick says the answer is clearly yes in many tasks, though not most. He cites data showing AI gives better stock advice than most first-year people doing that work, and says it is good at company and industry analysis, writing and website design. He adds that a first-year associate at a bank or PE fund still does things AI handles poorly: meetings to build consensus across the organisation, understanding office politics, and the implicit firm-specific judgement that sits outside the analyst report. He also notes companies face a tension because junior analysts are a training pipeline, so they must separate training goals from output goals.

    “I wouldn't say most tasks, but I would definitely say in many tasks, the answer is clearly yes... AI is probably a better analyst than most first year analysts.”at 14:25
  • Ethan Mollick · Wharton School, University of Pennsylvania

    AI will bring a disruptive transition where some jobs vanish and new ones emerge, with policy critical.

    Mollick says the labour-market models are confusing and the evidence is early. He notes a pattern of fewer junior hires, but possibly more coders hired across the wider economy because everyone can now code, which could raise software demand. Economists expect transformational growth, and the debate is whether the productivity boost is around 2% or closer to 0.5% of GDP. He says AGI models are unclear, so the best guess is a period of transformation and disruption to everyday life in which government and policy will matter a lot.

    “I think the general approach is we think there'll be a period of transformation where some jobs disappear, other jobs are created.”at 29:37

Venture capital & startups

Where everyone stands on Venture capital & startups

  • Ethan Mollick · Wharton School, University of Pennsylvania

    AI lets founders scale with fewer hires, but a one-person billion-dollar company remains unlikely.

    Mollick ran single-person startup experiments and thinks a reasonably sized business is achievable with AI. He argues a one-person shop is a bad idea because AI is jagged and running many agents still needs human attention. Businesses fully run by the same models also converge: answers, marketing and code look like everyone else's, so a moat needs human variation. He says a billion-dollar company needs a billion-dollar idea, not just AI. AI shortens the startup cycle: cheaper product-market fit, less capital and fewer hires (one marketer or coder doing the work of 10), though scaling to billions faces supercharged AI competition.

    “I think you can get a reasonably sized one. I don't... I think a billion dollar company kind of requires you to already have a billion dollar idea someone wants to acquire from you”at 2:36

Frontier AI labs & model race

Where everyone stands on Frontier AI labs & model race

  • Ethan Mollick · Wharton School, University of Pennsylvania

    Frontier model capability is compounding exponentially, with recent models running ahead of trend.

    Mollick tracks task length. When his first book came out in 2024, GPT-4 could do about four minutes of independent work at a 50% success rate. By April 2026 the figure was about 17 hours of human-equivalent work, and now it is about 17 weeks, though the benchmarks had failed at one point. He says the Fable 5.2 model is a little above the trend line, that models are improving faster than people expect, and that he sees no reason the curve is ending. It is very strong at design, web development, launching a startup and academic research, but still jagged.

    “The models are getting better in exponential curve... GPT four... could do about four minutes of independent work without human intervention with 50% success rate... we're up to now about seventeen weeks of work”at 10:53

Consumer AI products

Where everyone stands on Consumer AI products

  • Ethan Mollick · Wharton School, University of Pennsylvania

    Big game studios are pivoting to AI, while indie developers hide AI use, flooding games with slop.

    Mollick argues AI is a multiplier across software development, including games, and that output is booming. People can produce far more code, but companies do not yet know how to ship far more product because of other bottlenecks. Games have extra pressure: the industry is rough, with frequent layoffs and strong job protections. The big AAA studios are pivoting to AI for nearly everything, and he thinks GTA will be the last handcrafted game of that scale. Independent single-person developers are creating more, but a small, vocal audience punishes visible AI use, so indie studios use AI secretly. This creates an explosion of creativity alongside a lot of slop, and separating the two is hard.

    “The triple a studios, the big studios are all pivoting towards using AI a lot for everything. I think GTA is going to be the last handcrafted game of that scale we ever see.”at 17:26

Enterprise AI adoption & AI software

Where everyone stands on Enterprise AI adoption & AI software

  • Ethan Mollick · Wharton School, University of Pennsylvania

    Companies are seeing positive AI ROI but have only started reorganising around AI; productivity gains not yet visible.

    Mollick speaks to many large organisations and says none are backing away from AI. They are finding ROI, not uselessness or difficulty. But real transformation requires changing how the organisation operates at a fundamental level, and firms are only beginning to ask how to reorganise around AI work. Some have built software factories that are autonomously built. Some workers now produce far more code, but the gains are absorbed by old processes such as two-week sprint meetings. Productivity data is only starting to show the effect, and the real curve has not yet appeared.

    “Companies are just in the early stages of trying to implement how AI transformation works. They are seeing positive ROI, by the way.”at 26:40

Across the shows

What other episodes Chorus covered said about the same companies in the 30 days up to this one.

  • Anthropic119 speakers · 177 bullish, 90 bearish calls

Topics

  • AI in video games
  • AI-enabled startup scaling
  • IPO valuation pricing
  • compute capex and token demand
  • data center buildout
  • data center local opposition and community deals
  • enterprise AI ROI and transformation
  • junior analyst replacement
  • labor market transformation
  • model capability progress

More from The David Lin Report

All The David Lin Report episodes

Ask your AI about this episode

Chorus gives Claude, ChatGPT and other assistants every view from this episode and every other show Chorus follows. Ask who agrees, who changed their mind, and what else was said about the same companies.

Setup guide

A weekly email with the calls that mattered from the most important voices in investing. Unsubscribe any time. Privacy Policy

Chorus reports what speakers said. It isn't investment advice. How Chorus works · About Chorus