Oct 26, 2026 · 1h 3m
AI Can Be a Bubble and a Revolution at Once
E447: Anthropic at $2 Trillion? The AI Bubble, and Why the Best Startups Always Look Overvalued
The episode examines whether soaring AI prices are irrational excess or rational bets on a transformation that could remake software, enterprise work, and venture returns.
- 1AI valuations may be inflated even if adoption, productivity gains, and long-term demand prove real.
- 2Frontier labs face competing incentives around compute, model access, applications, services, and liquidity.
- 3Venture discipline increasingly means pacing investments, resisting consensus, and preserving agency amid technological abundance.
Don't miss
Matthew Harris explains why a $2 trillion Anthropic valuation could be defensible under continued hypergrowth yet vulnerable to shareholder liquidity dynamics.
The brief
Matthew Harris and David Whiteman test the AI bubble thesis against venture’s power-law logic, asking whether exceptional companies are overpriced or merely priced for extraordinary outcomes.
Consensus investing can make the presumed winner prohibitively expensive, while less obvious second movers still have room to emerge as strong businesses.
The frontier-lab debate turns on business models: should companies such as OpenAI and Anthropic remain model providers, or move into applications and services?
Harris treats a hypothetical $2 trillion Anthropic valuation as a test of hypergrowth, compute economics, and the liquidity pressures that can destabilize major AI issuers.
The conversation widens from today’s developer market to billions of software creators and potentially hundreds of billions of agents, while asking what capabilities trillions in compute will unlock.
The closing lesson is behavioral: pace investments, accept some upside sacrifice, and cultivate agency and perspective in children growing up with advanced AI.
What was said on this episode
32 statements · 19 positive · 8 negative · 2 mixed · 3 neutral
If AI resembles SaaS, its bubble may burst within months or quarters.
“If AI turns out to be like SaaS, which is to say yet another chapter in the book of technology, but not a game changer per se, that probably won't be November 2026, but we are definitely like months and quarters away.”
Listen at 0:37
Crusoe will return multiples of Bain Capital Ventures’ fund.
“Crusoe being one example, where we led the pre-seed and the seed that will surely return multiples of its fund.”
Listen at 1:45
Rejecting apparently overvalued startups systematically misses power-law companies.
“if you take a view that you're going to turn down things that are overvalued. At the very moment in time they're being valued, you will systematically miss the power law companies on average.”
Listen at 4:34
Investors should be highly price-insensitive when confident in a power-law winner.
“you should be extremely price insensitive if you have a greater degree of confidence that you're backing the power law winner.”
Listen at 7:41
AI investing is consensus-driven, but undiscovered winners still exist.
“I agree that we're in a consensus market. I disagree that there's no more diamonds in the rough.”
Listen at 8:47
AI market leadership is dynamic and continually challenged.
“leadership is actually not handed down from on high and protected forever, but is actually. constantly under siege.”
Listen at 11:24
Superintelligence is inevitable within one to three years.
“superintelligence is... perhaps not comprehensively imminent, but is certainly inevitable in the next one to three years.”
Listen at 14:16
AI-native services companies benefit from foundation-model advances.
“that they've benefited from advances made by the Foundation Model Labs. And we want to be levered to that, not fighting against that.”
Listen at 17:40
Anthropic’s legal-industry revenue will likely discourage it from starting a law firm.
“our view, probabilistically, is that that will inhibit them in starting their own law firm.”
Listen at 19:08
Frontier AI labs may never directly enter service industries.
“Our probabilistic view is that will be pretty long down the path and may never happen.”
Listen at 20:10
AWS’s relationship with JPMorgan Chase discourages Amazon from launching banking products.
“Amazon's never really done any of that, even though the plans are still on the shelf because JPMorgan Chase is just too valuable a customer if you're running AWS.”
Listen at 23:22
Differential compute access will be the main AI competitive battleground.
“the differential access to compute is going to be perhaps the prime battlefield in the next five years.”
Listen at 26:48
Rapid access to sufficient compute creates compounding advantage in AI models.
“If you have the compute to train the most impressive model in the shortest period of time, I think that's going to create compounding advantage.”
Listen at 27:14
Elon is currently best positioned among three leaders to build data centers and attract capital.
“I think Elon is the best of the three at that particular game, both of building data centers specifically, but also about sort of just, you know, world building in people's minds and attracting vast capital to vast projects.”
Listen at 29:25
OpenAI’s multi-model Codex strategy is strategically excellent.
“I really like that move. I think it's genius, actually.”
Listen at 30:52
Frontier labs restricting API access to vertical applications would be a bad decision.
“I think that would be a bad business decision.”
Listen at 32:33
A reported $2 trillion Anthropic valuation is reasonable given revenue growth.
“I think the valuation. based on the reported revenue numbers and revenue growth, is actually reasonable.”
Listen at 33:16
“If that growth rate can continue or anything like it can continue, they will grow into and through that valuation in a very reasonable period of time.”
Listen at 33:49
Anthropic could quickly appear undervalued if hypergrowth continues.
“if hyper growth continues, it'll quickly seem undervalued.”
Listen at 34:15
AI agents could expand the addressable user base to 80–800 billion entities.
“it's not 8 billion that you're going, you know, that 8 billion humans getting more productive. It's actually. 80 or 800 billion new entities being created.”
Listen at 40:17
Expanding AI demand will benefit pricing.
“I think there's no doubt that demand is going to expand and that will have a benefit to pricing as demand expands.”
Listen at 41:32
If model progress continues, $2 trillion will be only a starting valuation.
“If the 2027 frontier models are as transformational relative to the 26 models as the 26 models were to 2025, then I think $2 trillion is surely just a starting point.”
Listen at 41:50
Consensus AI data-center CapEx may reach $7 trillion over five years.
“it could be seven. How much of that will be spent on expanding access to more and more people using more and more AI versus expanding actual functionality?”
Listen at 43:50
AI’s projected $5 trillion CapEx reflects intensive training for unprecedented capabilities.
“the reason the number is $5 trillion and not $2 trillion or some other lower number is actually because of the intensity of training models to do unprecedented things”
Listen at 44:33
Reliable two-month reasoning would make Anthropic’s 2027 models necessary for many tasks.
“if you told me that Anthropics 2027 family of models could do two months long reasoning tasks reliably, then I don't think you have a choice.”
Listen at 46:29
Continued 2025–2026 AI progress would substantially benefit Hertz and other rental companies.
“If the pace doesn't even have to steepen, if it just continues on its trajectory, the 25 to 26 trajectory, if that pace continues, then absolutely both hurts. And even the car rental companies with their relatively modest reasoning tat.”
Listen at 48:01
Hertz is well positioned to launch its own insurance business using AI.
“aren't you perfectly situated to launch your own insurance business?”
Listen at 49:44
More capable AI models will prompt traditional companies to pursue new businesses.
“as the models get more performant and more capable, if they have talented management, they'll start quickly thinking, how do we use these models to do things we never even contemplated?”
Listen at 50:52
AI-generated software will include both substantial useless and useful output.
“we're going to get a lot of useless software. But we're going to get a lot of useful software too.”
Listen at 52:32
If AI follows prior technology waves, its bubble is in the eighth inning.
“if this is like other technology waves, we're in the eighth inning.”
Listen at 53:27
This venture vintage has more than a 10% chance of depressed returns.
“I do think, is there a chance that venture returns are depressed in this vintage? Yes. That's greater than 10% chance.”
Listen at 55:21
Venture investors should pace their investments consistently.
“the conclusion we've come to is you just got to. pace”
Listen at 56:36
Statements are attributed to the speaker as said on the episode and reflect their view at the time, not PodLume's. They are not advice.
Weekly email
State of the Pod, every Monday
The week's most-discussed names and stories from podcasts, with the moments behind them. Free, five minutes, unsubscribe in one click.