Is the current wave of AI investment a bubble?

Aravind Srinivas · Jun 5, 2024

AI is underhyped rather than overhyped — well-informed people mistake it for a bubble only because most of the world isn't using chatbots yet

Chat is an unfamiliar UI; once AI is delivered inside form factors people already live in — Word, Docs, Gmail, search — the impact will be tremendous, and that hasn't happened yet

46:17 20VC: Perplexity's Aravind Srinivas on Will Foundation Models Commoditise, Diminishing Returns in Model Performance, OpenAI vs Anthropic: Who Wins & Why the Next Breakthrough in Model Performance will be in Reasoning

Eric Vishria · Sep 25, 2024

The $600 billion AI CapEx-versus-revenue gap is not the right question to worry about; the revenue will materialize

Even setting AGI aside, the prize is so large that revenue follows; and the search analogy shows a powerful new technology can take five or six years to find its monetization model while users already love it

27:24 20VC: Benchmark's Eric Vishria on Where is the Value in AI: Chips, Models or Apps | Why Nvidia Will Not Be The Only Game in Town | The Commoditisation of Foundation Models | Which AI Apps Have Sustaining Value vs Hype and Short Term Revenue

George Sivulka · Jan 22, 2025

All AI companies and the entire S&P 500 are undervalued, including non-AI companies

AI is a real tangible technological shift on the order of what computing did for the economy over the last sixty to eighty years, and just as computers made legacy businesses better when used correctly, legacy firms that ride the AI agent wave get a massive unlock

Scope: self-labelled very hot take; conditional on ~$100T of value being created; requires companies to ride the wave of change

44:48 20VC: Why All AI Companies Are Under-Valued | The Future of Foundation Models: Scaling Laws, Generalised vs Specialised, Commoditised? | From Unable to Afford Rent to Raising $130M From Index and Peter Thiel with George Sivulka @ Hebbia

Byron Deeter · Aug 25, 2025

The current AI wave is a genuine generational technology transition, not just another market uptrend.

Mind-blowing demos are back and discussions involve things that were inconceivable a few years ago; it contrasts sharply with the post-2000s and post-pullback period of layoffs and survival.

5:04 20VC: Do Margins Matter in AI? | Is Defensibility Gone For Good? | Is Vertical SaaS Dead in a World of AI | What SaaS Rules Are BS and No Longer Apply in a World of AI | The Future of Venture: Why Chanel vs Walmart is BS with Byron Deeter

Brendan Foody · Sep 15, 2025

No AI segment stands out as overhyped — the most hyped areas (code, foundation models, finance use cases) are creating real value

The utility his own engineers get from Cursor, Claude Code and Cognition is incredible

38:40 20VC: Mercor: From $1M to $500M in 17 Months: The Fastest Growing Company in the World | How to Think About Margins and Revenue Sustainability in AI | Why Evaluation Benchmarks in AI are BS Today with Brendan Foody

Jonathan Ross · Sep 29, 2025

AI's valuation is driven by real value rather than a popularity contest, unlike assets such as crypto

PE firms are chasing cheap AI compute because it changes the bottom line of their portfolio businesses; when PE firms see value in something it's not hype

Scope: market has both weighing-machine and popularity-contest participants

51:46 20VC: OpenAI and Anthropic Will Build Their Own Chips | NVIDIA Will Be Worth $10TRN | How to Solve the Energy Required for AI... Nuclear | Why China is Behind the US in the Race for AGI with Jonathan Ross, Groq Founder

Zach Lloyd · Oct 17, 2025

The seemingly insane capital flows into AI are rational if you believe the technology is transformational — and AI will change every single business, producing both big losses and big winners

If the technology genuinely transforms every business, then the scale of investment is crazy but not crazy

Scope: describes himself as skeptical by nature; expects some big losses alongside big winners

31:46 20VC: The Startup Adding $1M ARR Every Week | Competing Against OpenAI's Codex and Claude Code: Who Wins | Why Gemini is Failing and GPT-5 Is Winning | Do Margins Matter in a World of AI | The Ugly Truth About AI Coding with Zach Lloyd, Warp

Anjney Midha · Apr 14, 2026

Capital for the 1.3 gigawatt proof of concept is not a constraint, but scaling beyond it requires far more capital deployed across the US and allied countries, and there is no natural stopping point for raising as long as the capabilities frontier keeps moving

A healthy independent ecosystem needs continuous capital as long as machine learning keeps systematically delivering more capabilities to humanity

Scope: assumes the capabilities frontier keeps advancing

51:37 20VC: Anj Midha on Investing $300M into Anthropic | The Early Days of Anthropic & How 21 of 22 VCs Turned it Down | The Four Bottlenecks to Compute | What the China Has Smashed and Why We Should Be Worried

Josh Browder · May 18, 2026

We are not in an AI bubble — the value shift is real and if anything the current scale is not big enough

He has reversed his view from twelve months ago after seeing how real the shift is

Scope: explicit reversal of his own prior view

81:25 20VC: Turning Peter Thiel's $100K into $10M Angel Portfolio | The One Man Accelerator at The Four Seasons | Why VCs Can Be Sharks and What Founders Need to Know | Why Stocks and Cash are BS and You Should Invest in Land with Josh Browder

Matan Grinberg · Jun 13, 2026

We are not in an AI infrastructure bubble

There may be short-term corrections where consumption was allocated inappropriately, but on net demand justifies the build-out

Scope: allows for short-term blips and corrections

43:26 20VC: Who Wins the Model War: OpenAI, Anthropic or Open-Source | Token Maxing, AI Hangovers & The Coming ROI Reckoning | Labour Displacement Fears are BS & Overblown | From Physicist to Sequoia Founder with Matan Grinberg, Founder @ Factory

Harry Stebbings · Jun 15, 2026

The AI infrastructure bubble narrative is funny, stupid and moronic

30:26 20VC: Micron Will Be More Valuable Than Meta | How Export Controls Helped Not Hurt China | Power is the Bottleneck to AI | Why Dario Has Done a Disservice to AI with his Labour Replacement Messaging with Aravind Srinivas, Founder @ Perplexity

KR Sridhar · Jun 29, 2026

We are not in an AI infrastructure bubble in any fundamental sense; AI-driven demand is a secular revolution whose trajectory resumes after any corrections

People confuse stock prices with infrastructure and secular change; AI puts a hockey stick on the hockey stick of digitization, and for the first time in history we are manufacturing intelligence — no civilization has ever decided it has too much intelligence

Scope: expects bumps, pauses and course corrections along the way; distinguishes stock market behavior from infrastructure demand

14:34 20VC: Leo Aschenbrenner's Largest Holding: Inside the $90BN Bloom Energy | Why Electricity, Not AI Models, Will Decide the Winners of the AI Race | Why We Are Not in an AI Capex Bubble | Energy Sovereignty and The Future of Power with KR Sridhar

Christian Lanng · Sep 27, 2023 · hedged

AI is a paradigm shift comparable to on-prem to cloud, and investors who write checks into it — likely after a culling — will capture outsized returns

Reid Hoffman walked around San Francisco funding startups that survived 2008, and something similar will happen here

Scope: probably needs a good culling first

63:25 20VC: "How Being a Founder Almost Killed Me"; We Have Lied to a Generation of Founders | The Hardest Truths About Being a Founder Revealed | Why AI Co-Pilot is BS, Seat Pricing is Over & User Interfaces are Stupid with Christian Lanng

Woody Marshall · Oct 11, 2023

AI is a remarkably fundamental trend that will have significant impacts on consumers and businesses, even though many of today's frothy bets will lose money

That duality is the nature of the venture business — some of today's bets blow up and others we'll look back on wishing we'd made; the reason people get excited about technology is precisely its capacity for fundamental impact

Scope: TCV has made no direct AI investments; 100% of portfolio companies are leveraging AI operationally

41:38 20VC Roundtable: Are IPOs Back? Is Growth Dead? What Does it Take to Raise a Growth Round Today? How Do VCs Solve The Liquidity Challenge? Will We See a Massive Resetting of Valuations? AI Hype Growth Rounds?

Michael Eisenberg · Jun 19, 2024

AI is simultaneously the most transformational technology of his lifetime and a massive financial gold rush that will become a bubble in which many people lose enormous amounts of money

Transformational technologies historically attract gold rushes; the internet produced both Pets.com and eBay/Amazon, so both truths can be held at once

Scope: both sides held together, not a retraction of either; 'AI' is a giant bucket that needs breaking apart

3:46 20VC: Foundation Models are the Fastest Depreciating Asset in History, Lina Kahn is a Threat to American Capitalism, PE is Not Coming to Save the M&A Market & How China Could Overtake the US in the AI Race with Michael Eisenberg

Mike Hudack · Sep 13, 2024 · hedged

The current AI/LLM experimentation is genuinely interesting but is in a hype cycle that will pass through a trough before reaching a productive plateau

You can see the minds working behind the scenes in products like Claude and ChatGPT, but these things go in cycles

Scope: doesn't know how it will end up

62:40 20Product: What Facebook, Monzo and Deliveroo Do and Do Not Do To Build Great Products | How to Structure Product Teams For Success | Is Simple Always Better in Product and The Art vs Science of Product Design with Mike Hudack

Fabien Pinckaers · Feb 12, 2025 · hedged

Even if an AI winter comes, genuinely disruptive AI use cases will emerge and Odoo will gain a lot from them

56:53 20VC: The $5BN Company Built from the Belgian Countryside | The Story of Odoo: The Company with No Plans to Sell, IPO & Their Billionaire Founder Who Does Not Care About Money with Fabien Pinckaers, Founder & CEO @ Odoo

Jonathan Ross · Sep 29, 2025

The stock market can overheat and crash independently of AI's underlying value, killing many good businesses on the way down

Control-system dynamics: people bid things up, realize the mistake, overshoot downward, spending retreats, and businesses lose the funding they need

Scope: independent of whether AI itself is valuable

53:29 20VC: OpenAI and Anthropic Will Build Their Own Chips | NVIDIA Will Be Worth $10TRN | How to Solve the Energy Required for AI... Nuclear | Why China is Behind the US in the Race for AGI with Jonathan Ross, Groq Founder

Harry Stebbings · Oct 17, 2025

There are good bubbles: this one will destroy a huge amount of capital while producing a huge amount of technological development

Bubbles are wrongly assumed to be purely bad; capital loss and technological progress can coexist

32:07 20VC: The Startup Adding $1M ARR Every Week | Competing Against OpenAI's Codex and Claude Code: Who Wins | Why Gemini is Failing and GPT-5 Is Winning | Do Margins Matter in a World of AI | The Ugly Truth About AI Coding with Zach Lloyd, Warp

David Cahn · Oct 27, 2025

AI is a genuine multi-decade generational transformation, but the market is pricing too high a probability that it all happens in a very short time horizon with a very specific chipset, so the short-term cycle will incinerate capital

In fifty years AI will have completely reshaped society, but the market implies that transformation lands almost immediately on today's specific hardware

13:52 20VC: Sequoia's David Cahn on The Winners and Losers in AI | The $0-$100M Revenue Club: Is Triple, Triple, Double, Double Dead? | The Future of Defence: Who Wins and Who Loses | How to Analyse Margins and Growth Rates in a World of AI

Chris Degnan · May 23, 2026

There is an AI bubble that will burst, but AI is still genuinely life-changing, so companies should build for speed now

AI is changing how people interact with technology, even if valuations and demand are inflated

Scope: holds both sides: bubble will burst AND the technology is transformative

60:01 20Sales: The $100M CRO Bubble: Why Anthropic Are Causing a Comp Crisis | Why You Should Never Hire From Salesforce or Service Now | How to Hire, Train and Forecase in a World of AI with Chad Peets and Chris Degnan

Vince Hankes · May 3, 2023 · hedged

AI may be a hype cycle akin to crypto or the dot-com bubble, so rather than chasing deals investors should focus on the underlying product and core customer value

In the dot-com peak, Microsoft, Intel and Cisco were the top three tech companies and half of $3T in market cap; an investor who bought Microsoft in 2000 would have waited fifteen or twenty years to break even, because speculation fuels more speculation

11:33 20VC: The OpenAI Memo: Why Invest? Is it too Late to Catch OpenAI? Are OpenAI's Models Truly Defensible? Does the Value in AI Accrue to Incumbemts or Startups - Application Layer/Infrastructure? What Happens with Regulation? with Vince Hankes @ Thrive

Vince Hankes · May 3, 2023

Crypto was the craziest excess of the low-rate era, and investors will look back on it as proof that hype cycles produce massive speculation in which even the assets people believe are most real turn out not to be

Crypto token peak market cap of roughly $3 trillion was comparable to the dot-com peak, and the most iconic company everyone talked about turned out to be a fraud

Scope: particularly from an investor lens

45:42 20VC: The OpenAI Memo: Why Invest? Is it too Late to Catch OpenAI? Are OpenAI's Models Truly Defensible? Does the Value in AI Accrue to Incumbemts or Startups - Application Layer/Infrastructure? What Happens with Regulation? with Vince Hankes @ Thrive

Mitchell Green · Mar 28, 2025

What is happening in AI looks very similar to the internet bubble, and investors learned nothing from 2021

People who have been in the industry far longer than him draw the same parallel, and the same behaviors are repeating

51:53 20VC: Why Traditional VC is Broken: How VCs Learned Nothing from 2021 | Why LPs are More Important than Founders & Advice to Emerging Managers | Bull Case for Bytedance & Why TikTok's Ban Doesn't Matter with Mitchell Green, Lead Edge Capital

Everett Randle · Nov 10, 2025

The current AI cycle mirrors the dot-com era: many companies will go to zero or fall 90%, but the few enduring winners will return many multiples of a normal cycle, so the job is to be positioned to survive the inevitable crash

Amazon's Series A was down 80% from IPO four years later but went from ~$40m post to multiple trillions if held; the same asymmetry should apply to the companies that define the next twenty to thirty years

Scope: holds both 'Gotham is burning' and 'we're very early' simultaneously

72:07 20VC: Benchmark's Newest General Partner Ev Randle on Why Margins Matter Less in AI | Why Mega Funds Will Not Produce Good Returns | OpenAI vs Anthropic: What Happens and Who Wins Coding | Investing Lessons from Peter Thiel and Mamoon Hamid

Chris Degnan · May 23, 2026

The current environment resembles the dotcom boom

38:55 20Sales: The $100M CRO Bubble: Why Anthropic Are Causing a Comp Crisis | Why You Should Never Hire From Salesforce or Service Now | How to Hire, Train and Forecase in a World of AI with Chad Peets and Chris Degnan

Mike Cannon-Brookes · Oct 13, 2025

AI companies do not yet have durable business models — revenue is circulating between AI companies, model providers, cloud providers and chipmakers while everyone loses money along the way

An AI company pays $100m to a model company, which pays $150m to a cloud provider, which pays $200m to NVIDIA, and each reports revenue while losing money

Scope: maybe they make it up at scale

16:55 20VC: Atlassian CEO on Why Everything is Overvalued & Are We in an AI Bubble | Do Margins Matter & Does Defensibility Exist in an AI World | Is Per Seat Pricing Dead & The Future of Vibe Coding with Mike Cannon-Brookes

Andrew Ng · Nov 17, 2025 · hedged

Heavy use of complex financial instruments to shift risk around, including circular deals, raises the risk of a bubble and is a mildly bubble-ish signal, though not yet alarming.

Overly complex risk-shifting increases bubble risk; bubbliness is a matter of degree rather than binary.

Scope: not alarmed; something to keep an eye on

49:45 20VC: Andrew NG on The Biggest Bottlenecks in AI | How LLMs Can Be Used as a Geopolitical Weapon | Do Margins Matter in a World of AI? | Is Defensibility Dead in a World of AI? | Will AI Deliver Masa Son's Predictions of 5% GDP Growth?

Oren Zeev · Feb 2, 2026

The notion that only growth matters is dangerous and is currently pushing companies into unsustainable behaviour such as circular revenue deals

In a circular deal each side books a million of revenue and a million of cost, so perceived value is created but no real value; he has seen this movie many times and some of these companies will implode

Scope: describes it as a grey area short of outright fraud; some such companies may still succeed

15:44 20VC: 50% of Funds Will Go Out of Business | Why Growth Expectations Today are BS and Will Not Last | Why Oren Zeev Takes $0 Management Fees But 30% Carry | Why GPs Should Not Tell LPs Their Strategy

Jerry Murdock · Aug 22, 2026

NVIDIA's flat stock reflects a normal plateau plus genuine uncertainty about underlying growth, because circular transactions between hyperscalers are obfuscating what real demand looks like

Markets don't go up like a rocket ship forever, and hyperscalers making circular deals to get ahead of the game means nobody can read true growth; a 10-15% market dislocation would make people feel poor and hit credit and demand

Scope: acknowledges 'we don't know'

54:57 20VC: The AI Bubble Will Burst: Half the Neoclouds Will Die | China: Should We Ban Chip Exports & Be Fearful of Chinese Open-Source | Mag7: Who Dies and Who Thrives: Why Meta is Meh and Microsoft is Mega

Harry Stebbings · Aug 5, 2024

An executive at one of the largest companies with access to the budgets says AI spend is like the Manhattan Project — once you're in, you cannot pull out

Scope: source unnamed

9:41 20VC: Sequoia's David Cahn on AI's $600BN Question | Why the Data Centre is the Most Important Asset | Servers, Steel and Power: The Core Pillars Powering the Future of AI

David Cahn · Aug 5, 2024

The cloud providers are playing defense on AI while Meta is playing offense, because the cloud players are locked in a prisoner's dilemma where not investing risks losing share in one of the greatest businesses of all time.

Cloud incumbents must protect an existing business; Meta has no cloud franchise to defend, so it can be creative and can stop investing if it decides AI isn't worthwhile.

26:14 20VC: Sequoia's David Cahn on AI's $600BN Question | Why the Data Centre is the Most Important Asset | Servers, Steel and Power: The Core Pillars Powering the Future of AI

Harry Stebbings · Aug 26, 2024

AI capex is an arms race in which incumbents have no choice but to spend

40:02 20VC: Why the IPO Market is not Closed | Why Revenue Multiples are BS and Founders Need to Change | Advice From Jack Ma, Jamie Dimon and Evan Spiegel | Lessons from Taking Snap & Alibaba Public with Imran Khan

Jonathan Ross · Sep 29, 2025

Hyperscaler AI capex is not driven by a purely economic framework but by existential fear of being locked out of their own business, though the financial returns will also be positive

At a Goldman Sachs event, not one of ~50 investors managing $10B+ was 100% convinced AI couldn't do their job in ten years; hyperscalers feel the same, so the alternative to spending is losing their business

7:28 20VC: OpenAI and Anthropic Will Build Their Own Chips | NVIDIA Will Be Worth $10TRN | How to Solve the Energy Required for AI... Nuclear | Why China is Behind the US in the Race for AGI with Jonathan Ross, Groq Founder

Harry Stebbings · Jan 10, 2024

AI is still an irrational segment where founders are being offered term sheets at 2021-style prices

Scope: confined to AI; the rest of the market has reset

20:43 20VC: Did Figma Kill M&A Markets in 2024, The Three Biggest Mistakes Made in Growth Investing, The Three Requirements Companies Need to Go Public in 2024 with Ed Sim and Jamin Ball

Harry Stebbings · Oct 25, 2024

AI investing is more crazy than ever, not cooling off

He has met three companies that raised at north of $750M pre-product

40:13 20VC: The Truth About Multi-Stage Firms; Why Portfolio Services are for VCs not Founders | Why Politics is Rife & Decision-Making is Broken in Large VCs | Why Reserves are Bad for Founders & How Boutique Firms Will Win with Mark Goldberg @ Chemistry

Mitchell Green · Mar 7, 2026

Funding Anthropic/OpenAI spinouts at $1-2B on nothing more than an idea is lunacy, and there is no evidence such enormous idea-stage raises have ever worked

He can't identify a single case where a napkin-idea company raised at hundreds of millions or billions and actually worked

Scope: frames the historical question as genuinely open — 'I don't know'

40:23 20VC: Why the SaaS Apocalypse is BS | Why China Will Win the AI War | Why 50% of VCs Should Not Exist and are Tourists | Why Stock-Based Comp is the Hidden Sin of the Valley with Mitchell Green, Lead Edge Capital

Jerry Murdock · Aug 22, 2026

Today's price insensitivity in venture is evidence that we are still in the hype cycle

Expectations have run beyond everyone's imagination, so round sizes and valuations are being set by what the market is doing rather than by thought

32:03 20VC: The AI Bubble Will Burst: Half the Neoclouds Will Die | China: Should We Ban Chip Exports & Be Fearful of Chinese Open-Source | Mag7: Who Dies and Who Thrives: Why Meta is Meh and Microsoft is Mega

Jason Lemkin · Aug 9, 2023

The real problem with the AI wave is not sameness but that entry prices are too high to wait for a breakout — you can't lean in at $100-200m valuations, so the wait-until-$1-2m-revenue discipline is no longer available

He'd love to wait until a company breaks out at one to two million in revenue and then lean in, as he could with TalkDesk, but the prices at that point now preclude it

Scope: he frames himself as discouraged rather than jaded

25:10 20VC: The Memo: The State of the VC Market: Why Seed Funds Can't Invest in "Hot Startups" Anymore, Why Series A & B is Terrible, Why the IPO Market Will Explode in 2024 & Why VC DD is BS & Every VC Has More Fraud in their Portfolio with Jason Lemkin

Immad Akhund · May 12, 2025

AI is overhyped and overvalued at the seed stage, so while he still does some AI deals, he has become far more selective and looks to less crowded areas like fintech, space tech and hard tech.

He hears the same idea pitched a fourth time, with founders raising at ~$40M valuations on a little traction with good investors already on board, so the math doesn't work.

Scope: about seed-stage AI pricing specifically; he still does AI deals — four of his last eight looked at were AI

35:48 20VC Exclusive: Mercury Founder Launches First $26M Fund | Why Founders Should Take the Highest Price | Why Serial Entrepreneurs are Better | Why AI Is So Overhyped | The Future of Venture Capital with Immad Akhund

Gili Raanan · Mar 28, 2026

The current venture market will not work out and will end in serious catastrophe for many players

Entry prices at seed have risen dramatically (e.g. $15M post in 2012 to $100-150M post today) while the probability of hitting a successful company stays around 1%, so much of the capital flowing in will be wasted

Scope: for many players, not all; argued primarily from cybersecurity data

0:00 20VC: The Venture Model is Broken | You Need to be Greedy and Selfish to Win Early Stage Investing | Why Margins Do Not Matter for Early-Stage Startups | The Growth Rate that is Required in a World of AI with Gili Raanan, Founder @ Cyberstarts

Jonathan Ross · Feb 17, 2025

A huge amount of money will certainly be incinerated in AI, but in aggregate more money will be made than was put in

In any real technology wave, early heavy bets succeed and then everyone piles in; not every individual company can work, so you must judge aggregate returns separately from individual bets

Scope: aggregate vs individual bets distinguished; 'I also bet' on the aggregate upside

43:36 20VC: NVIDIA vs Groq: The Future of Training vs Inference | Meta, Google, and Microsoft's Data Center Investments: Who Wins | Data, Compute, Models: The Core Bottlenecks in AI & Where Value Will Distribute with Jonathan Ross, Founder @ Groq

Philipp Freise · Jun 30, 2025

The underlying principles of rational capital allocation have not changed despite AI; some AI models justify enormous capital and others do not

Things look different every five years, but separating noise from reality shows the fundamentals hold — as Buffett has argued and Freise agrees

Scope: differentiates between AI businesses case by case

17:44 20VC: Inside KKR's Monster $8BN European Fund | The $500M Turkey Gamble That Went Wrong | Do Andreessen & General Catalyst Scare KKR? | Will AI Kill the PE Model? | Can The PE Model Survive without IPOs and Where is the Liquidity with Philip Freise

Brendan Foody · Sep 15, 2025

He is not concerned about aggregate AI CapEx, though specific investments show exuberance and need scrutiny on ten-year ROI

On a ten-year investment horizon the market generally looks like it's at a discount

Scope: there are definitely individual cases of exuberance

38:10 20VC: Mercor: From $1M to $500M in 17 Months: The Fastest Growing Company in the World | How to Think About Margins and Revenue Sustainability in AI | Why Evaluation Benchmarks in AI are BS Today with Brendan Foody

Miles Dieffenbach · Aug 4, 2025

OpenAI could still be a zero because it does not control its own destiny — it burns $5-10B a year and sits behind a $70-80B preference stack

It raised two of the largest venture rounds ever in twelve months not for interest income but to fund burn; if capital markets get smoked and nobody will write a $40B equity check, the pref stack becomes a problem

Scope: unit economics are improving rapidly; conditional on a bubble popping

69:16 20VC: Inside Carnegie Mellon's $4BN Endowment | Why 90% of LPs Shouldn't Invest in VC | The $140BN Problem with Multi-Stage Funds | The Hidden Math Behind DPI, TVPI, and Illiquidity with Miles Dieffenbach

David Cahn · Oct 27, 2025

The thing people currently get wrong is that anything multiplied by zero is zero: market volatility doesn't matter long-run for a great business, but if you overextend and go bankrupt in a crash there is no recovery

Time horizons have compressed so much that people ignore survival risk

47:15 20VC: Sequoia's David Cahn on The Winners and Losers in AI | The $0-$100M Revenue Club: Is Triple, Triple, Double, Double Dead? | The Future of Defence: Who Wins and Who Loses | How to Analyse Margins and Growth Rates in a World of AI

Jerry Murdock · Aug 22, 2026

Even when the underlying asset retains long-term value, heavy debt dependence means a dislocation sharply depresses asset values and triggers margin calls that bankrupt the builders

Dot-com era fiber was always valuable but the companies that laid it went bankrupt

Scope: decline may not be permanent

9:41 20VC: The AI Bubble Will Burst: Half the Neoclouds Will Die | China: Should We Ban Chip Exports & Be Fearful of Chinese Open-Source | Mag7: Who Dies and Who Thrives: Why Meta is Meh and Microsoft is Mega

Jonathan Siddharth · Dec 1, 2025

Whoever wins the superintelligence race will likely also win search, consumer devices, operating systems, cloud, business productivity software and social networking — which is why enormous forward bets by the big tech companies are rational rather than bubble behaviour

The prize is roughly $30 trillion of knowledge work automation plus every adjacent market, so the cost of not winning is too high

Scope: conditional on believing AGI is achievable

44:20 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Harry Stebbings · Dec 1, 2025

For a company like Meta the AI spend is a rational asymmetric bet: losing $100B is only twelve to eighteen months of free cash flow and everyone else likely fails too, whereas not spending while a rival wins costs two to three trillion in market cap

The downside of failure is small relative to cash flow; the downside of being left behind is existential

45:52 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Jonathan Siddharth · Dec 1, 2025

Incumbent social networks have to make the big AI bet because attention is their single unit of value and users only have four to five hours a day — a more engaging AI-built app could take it all

Attention is fixed and zero-sum, so a more engaging competitor is existential

46:12 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Anish Acharya · Feb 9, 2026 · hedged

We are not in an AI bubble, and the subsidization that does exist is a good thing

OpenAI 3x'd capacity and 3x'd top line to $20B with all inference supply spoken for — unlike prior bubbles where supply was built far ahead of demand; customer prices are rising rather than compressing; and the subsidy is intelligent, funded by big tech and the labs, benefiting consumers and startups

Scope: says this is not his area of focus or expertise; concedes there is always some distortion in subsidization

31:22 20VC: Is SaaS Dead in a World of AI | Do Margins Matter Anymore | Is Triple, Triple, Double, Double Dead Today? | Who Wins the Dev Market: Cursor or Claude Code | Why We Are Not in an AI Bubble with Anish Acharya @ a16z

Lin Qiao · Jul 20, 2026

We are not in a CapEx bubble; the binding constraint is the lower layers of the AI stack — energy, chips and physical manufacturing — which were never designed for 100x scaling

Supply chains for servers and data center hardware bottleneck on the smallest components like transistors, and no part of that industry was designed for this scale

37:42 20VC: Are OpenAI and Anthropic Overvalued? The Open-Source AI Reality | How Token Costs Will Fall 10x And Usage Will Explode 100x | The Future Is Not One AGI; It's Millions of Specialised Models with Lin Qiao, Founder and CEO @ Fireworks

Arthur Mensch · Apr 29, 2024 · hedged

Cloud-provider investments into AI labs that get spent back on compute make sense for both sides even though they look like round-tripping

Scope: says he doesn't know that particular deal

24:37 20VC: Mistral's Arthur Mensch: Are Foundation Models Commoditising | How Do We Solve the Problem of Compute | Is There Value in the Application Layer | Open vs Closed: Who Wins and Mistral's Position

Jonathan Ross · Sep 29, 2025 · hedged

NVIDIA's $100B investment into OpenAI is not an infinite money loop, at most a partial one, because roughly 40% of the dollars flow out to suppliers building real infrastructure

It isn't round-tripping if actual productive outcomes are occurring downstream

Scope: ~40% share of spend estimated

11:11 20VC: OpenAI and Anthropic Will Build Their Own Chips | NVIDIA Will Be Worth $10TRN | How to Solve the Energy Required for AI... Nuclear | Why China is Behind the US in the Race for AGI with Jonathan Ross, Groq Founder

Byron Deeter · Aug 25, 2025

The trillion-plus dollars of private market cap in the top 100 cloud/AI companies is entirely legitimate rather than synthetic hype

The leading names are effectively tradable public-quality entities in a private wrapper, and there are real buyers and sellers at or above the last-round marks all the way down the top 100

Scope: value is skewed toward the top of the list; companies ranked 101-300 may still include walking wounded and artificial last-round prices

40:45 20VC: Do Margins Matter in AI? | Is Defensibility Gone For Good? | Is Vertical SaaS Dead in a World of AI | What SaaS Rules Are BS and No Longer Apply in a World of AI | The Future of Venture: Why Chanel vs Walmart is BS with Byron Deeter

Andrew Feldman · May 26, 2026

CoreWeave has been an extraordinarily innovative company whose valuation is earned through genuine financial innovation and rapid deployment skill, not merely hype

They were first to use debt in a very innovative way and are extremely good at rapid deployment, an important skill in this environment

Scope: speaking only about CoreWeave, not the other neoclouds; all companies face challenges as they grow

18:49 20VC: Cerebras CEO on the Future of Data Centres, Token Costs and Memory | We are Not in an Infra Bubble & Dario Got a Bad Deal with Elon for Compute | Should US Companies Sell to China & Why Most Layoffs are AI Washed with Andrew Feldman

David Cahn · Aug 5, 2024

The 'compute is the future' thesis and the compute-overproduction thesis are somewhat at odds, because compute is not an abstract commodity but a physical asset that can be built wrong and become obsolete.

Compute is a two-year, ~$2BN data centre build in a physical location with specific chips and cooling; nobody yet knows how to build GPU data centres well, and when better chips and cooling arrive the physical build has to be redone.

Scope: he still fully agrees compute is the future

14:47 20VC: Sequoia's David Cahn on AI's $600BN Question | Why the Data Centre is the Most Important Asset | Servers, Steel and Power: The Core Pillars Powering the Future of AI

Anjney Midha · Apr 14, 2026

We are not in an AI bubble; we are in a GPU wastage bubble, with billions of dollars of stranded compute sitting unutilized — and we are deeply underinvested in secure compute.

Pockets of compute sit idle and could be pooled onto a grid across independent operators.

32:17 20VC: Anj Midha on Investing $300M into Anthropic | The Early Days of Anthropic & How 21 of 22 VCs Turned it Down | The Four Bottlenecks to Compute | What the China Has Smashed and Why We Should Be Worried

Aravind Srinivas · Jun 15, 2026 · speculative

There is roughly a 20-30% chance of another DeepSeek moment producing a vastly more efficient, vertically integrated model that runs on local devices and strands built-out capacity

Export controls on Nvidia GPUs and HBM are forcing DeepSeek to build on the Huawei stack with memory-efficient innovations — small KV caches hostable on SSDs, attention layer changes, training algorithms that use less interconnect — producing a fully vertically integrated architecture very different from America's bet

42:52 20VC: Micron Will Be More Valuable Than Meta | How Export Controls Helped Not Hurt China | Power is the Bottleneck to AI | Why Dario Has Done a Disservice to AI with his Labour Replacement Messaging with Aravind Srinivas, Founder @ Perplexity

Harry Stebbings · Jul 11, 2026

If AI gets much cheaper, the already loss-making model businesses that underpin the global economy become very threatened.

33:20 20VC: Why OpenAI and Anthropic Won't Win the App Layer | Why Teams Will Get Bigger Not Smaller in a World of AI | Why AI Removes Incumbents Advantage of Bundling | China vs America: Who Wins the AI War with Arvind Jain, Co-Founder @ Glean

Brendan Foody · Sep 15, 2025 · hedged

Weak competitors are being funded to the tune of hundreds of millions that shouldn't be, and AI is being overestimated on a three-year horizon while being underestimated on a ten-year one

Technology is generally overestimated in the short term and underestimated in the long term; on a ten-year view the extraordinary businesses being built now will look like a discount

Scope: frothy on a 3-year horizon; a discount on a 10-year horizon; unclear whether we're at 1996 or 1997 analogue

34:01 20VC: Mercor: From $1M to $500M in 17 Months: The Fastest Growing Company in the World | How to Think About Margins and Revenue Sustainability in AI | Why Evaluation Benchmarks in AI are BS Today with Brendan Foody

Demis Hassabis · Apr 7, 2026

AI is currently overhyped on a one-year horizon while still being underappreciated on a ten-year horizon

The old adage that we overestimate one year and underestimate ten still holds for AI, even though both timescales are compressed relative to other technologies

Scope: both short and long timescales are nearer than for other technologies

23:30 20VC: DeepMind's Demis Hassabis on Why AGI is Bigger than the Industrial Revolution | Why LLMs Will Not Commoditise & We Have Not Hit Scaling Laws | Bottlenecks in AI & The Energy Crisis Caused By AI | Whether AI Will Do More to Harm or Help Inequality

Victor Riparbelli · Jan 15, 2025

We are in an AI bubble, but that is not necessarily bad — though a lot of money will be lost on products that either aren't valuable or become features of the big cloud providers

Throwing lots of money at a million different products to see what sticks is Darwinistic and the right way to innovate

Scope: bubble framed as a healthy feature of capitalism

21:05 20VC: Why Scaling Laws Will Not Continue | OpenAI vs Anthropic vs X.ai: Who Wins and Why | How Far Will Model Providers Go Into the Application Layer | The End State for Models: Many Specialised or Few Generalised with Victor Riparbelli @ Synthesia

Anastasios Angelopoulos · Aug 3, 2026

The AI industry is over-hyped and would benefit from a sobering-up and consolidation

There's too much crap happening; consolidation would reveal what actually shakes out and let winners absorb good people

58:15 20VC: 70% of Neolabs Will Die | There Will be a $100BN US Open-Source Model | Data is a Trillion $ Market | Governments Cannot Regulate Models: It is Too Late | The Cyber Attacks to Come Will be Insane with Anastasios Angelopoulos @ Arena

Winston Weinberg · Jan 19, 2026

We will see repeated short, self-fulfilling AI panics like the DeepSeek moment, because sentiment is close enough to the edge that one trigger can make everyone freak out

If enough people are predicting a bust, a single event can make the bust happen as a self-fulfilling prophecy — but such episodes are usually short

Scope: episodes expected to be short-lived

42:58 20VC: How Model Performance is Plateauing | Two Key Rules for Effective Deal-Making | Company Building Lessons from Keith Rabois, Brian Halligan and Pat Grady | Why Enterprise AI Adoption is Years Off with Harvey CEO Winston Weinberg

Jerry Murdock · Aug 22, 2026

There are multiple possible triggers for a dislocation, not just one, because the AI buildout involves enormous global capital flows

The AI revolution is incredibly complex with massive amounts of dollars being spent globally

6:15 20VC: The AI Bubble Will Burst: Half the Neoclouds Will Die | China: Should We Ban Chip Exports & Be Fearful of Chinese Open-Source | Mag7: Who Dies and Who Thrives: Why Meta is Meh and Microsoft is Mega

Harry Stebbings · Dec 1, 2025

Value generation from AI depends fundamentally on whether budget transfers from human labor to AI technology; without that transfer, current valuations and capital inflows are unjustified

With the transfer you get the $30 trillion opportunity; without it you merely get a slightly larger software technology budget

21:35 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Gokul Rajaram · Mar 16, 2026

The valuations only make sense if you believe these companies take over huge parts of customers' services and payroll budgets and become multi-billion rather than billion-dollar companies.

Previously these products were limited to one part of the stack sitting on top of other systems; now they can take over the entire software stack, so the right market sizing is how much of the customer's digital labor payroll and payments flow they can replace.

Scope: requires founders ambitious enough to go after the entire stack

31:18 20VC: The 8 Moats of Enduring Software Companies: How to Analyse for Durability and Defensibility in a World of AI | Why Dropouts are "AI Maxing" the World & Remote Early-Stage Companies are Dying with Gokul Rajaram

David Frankel · Oct 14, 2024

In the short term AI will disappoint, because the capex being deployed cannot be justified by near-term earnings — and the disappointment will run through the whole ecosystem, from hyperscalers to data centers, steel and chips

The gap between the scale of capital expenditure and actual earnings generated in the short term simply doesn't add up

Scope: short-term horizon only

51:40 20VC: Investing Lessons from FC Seeding Uber, Airtable and Coupang | Why Pro Rata is the Original Sin in VC | Why Liquidity Has Died in 2024 | Why LPs are Pissed with VCs | The Hard Truth About Seed Fund Economics with David Frankel @ Founder Collective

Harry Stebbings · Dec 1, 2025 · hedged

AI is likely to undershoot the revenues that have been promised, leading to a cooling period in which the AI bubble slowly deflates

Scope: framed as a personal worry; echoing what many expect

39:32 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing

Emad Mostaque · May 17, 2023

The AI funding bubble will be the biggest bubble and biggest shit show yet, because the amount of money being allocated is completely misaligned with the actual opportunity and capacity in the sector

His TAM analysis is roughly a thousand companies spending $10M, a hundred spending $100M, ten spending $1B — while a multiple of that will be allocated because AI is the only growth theme against rising rates and crashing real estate; PwC alone is committing $1B without knowing where it goes, Chinese firms pay $1.2M PhD salaries, and GitHub stars are leading $200M rounds with zero traction and no business model

16:18 20VC: Why the AI Bubble Will Be Bigger Than The Dot Com Bubble, Why AI Will Have a Bigger Impact Than COVID, Why No Models Used Today Will Be Used in a Year, Why All Models are Biased and How AI Kills Traditional Media with Emad Mostaque, Founder & CEO @

Kevin Hartz · Jul 22, 2024

We are in the early years of the biggest bubble ever — an AI bubble — with compounding add-on bubbles in crypto, fintech and clean energy, which will create enormous value alongside enormous foolishness

NVIDIA becoming the most valuable company on the planet is a precursor; transformers and ChatGPT proved the platform is here, giving founders many platforms to build apps on top of

Scope: 'bubble' framed in the context of capitalism; acknowledges much foolishness will accompany the value created

57:12 20VC: How I Lost Airbnb at Seed Because of an Exploding Term Sheet | Investing Lessons from Roelof Botha & Peter Thiel | Why VC is Less Collaborative Than Ever and Great Companies Are Being Destroyed by Too Much Cash with Kevin Hartz @ A*

Saam Motamedi · Jul 15, 2024

We are in an exuberant AI bubble that is arguably crazier than the 2021-2022 peak, and its size may not be fully appreciated.

Seed rounds for just-started companies price in the many tens of millions up to $100M+ post-money, and companies with a little revenue raise at 100-200x revenue, while the best public software names trade at only 15-20x forward revenue — a far worse dislocation than 2021 when 100x private pricing sat against 50-80x public comps.

Scope: specific to pure AI companies

6:16 20VC: Why We Are in a Bubble & Now is Frothier Than 2021 | Why $1M ARR is a BS Milestone for Series A | Why Seed Pricing is Rational & Large Seed Rounds Have Less Risk | Why Many AI Apps Have BS Revenue & Are Not Sustainable with Saam Motamedi @ Greylock

Harry Stebbings · Nov 20, 2024

It is a really hard time to be investing right now because pricing is insane and the AI bubble is more prolific than 2021 was

Scope: framed as his own worry and perspective

54:39 20VC: Turning $16.5M into $2.1BN; Lessons from the Greatest Venture Investment in European History: UiPath | Why VC is Not Being Commoditised | Why Price Does Not Matter | Lessons on Loss Ratio, Selling and Signalling with Cem Sertoglu

Harry Stebbings · Oct 14, 2024

AI capex implies roughly $600B of revenue that does not yet exist

51:52 20VC: Investing Lessons from FC Seeding Uber, Airtable and Coupang | Why Pro Rata is the Original Sin in VC | Why Liquidity Has Died in 2024 | Why LPs are Pissed with VCs | The Hard Truth About Seed Fund Economics with David Frankel @ Founder Collective

Mitchell Green · Mar 7, 2026

Many companies currently valued in the trillions are way ahead of themselves relative to the profits they will ultimately generate

Very few companies ever reach the profit levels of Facebook, Google, Microsoft, Amazon and Nvidia, and we haven't yet seen what these newly-valued companies actually exit at

33:40 20VC: Why the SaaS Apocalypse is BS | Why China Will Win the AI War | Why 50% of VCs Should Not Exist and are Tourists | Why Stock-Based Comp is the Hidden Sin of the Valley with Mitchell Green, Lead Edge Capital

Joelle Pineau · Nov 3, 2025

The biggest economic challenge in AI today is the absence of predictability — no one can say when breakthroughs land, how many GPUs are needed, or what return to expect.

There is a great deal we still do not know about the technology, so building data centers, workforces or data curation programs all require absorbing substantial risk, unlike other industries where people can get answers.

10:07 Unpredictable breakthrough timing makes capex and returns unknowable

Joelle Pineau · Nov 3, 2025

The current AI capital cycle is best understood as a bubble with unusually high variance — bigger upswings and bigger downswings — rather than as simply a good or bad bubble

There is a lot of variance in the system right now, so the outcome depends on the investor's risk tolerance

44:04 A bubble with unusually high variance in both directions

Ed Sim · Oct 27, 2023

Chasing companies purely because they have AI in the name is insane, and building a portfolio of AI deals at $100M post-money will not work even though massive companies will be built in AI

It mirrors the 1996 internet bubble when everything had a .com in its name; he lived through that

26:37 Chasing ai labeled deals blindly echoes the 1996 dot com naming bubble

Lucas Swisher · Feb 23, 2026

Many investors are crowding into pre-revenue AI companies mainly because they are locked out of the great platform companies and can only make the decisions available to them

If you can't get into OpenAI, Anthropic, Revolut, SpaceX or Canva and you're constrained to one part of the ecosystem, you invest where you can

35:50 Pre revenue crowding is access constrained not conviction

Douwe Kiela · Jun 30, 2023

The large AI venture rounds are justified rather than crazy, though some companies now carry expectations they must convert into real revenue before an eventual disillusionment dries up funding.

The technology is really going to change the world, so a correct bet has massive payoff.

31:49 Large ai rounds are justified but risk disillusionment once revenue expectations arent met

David Cahn · Aug 5, 2024

Whether AI will change the world and whether current capex levels are too high are two separate questions

You can believe AI will transform society while also believing the amount of capex spent in the next twenty four months may be difficult to pay back

7:57 Ai transformative potential and current capex payback risk are separate questions

David Cahn · Aug 5, 2024

Mark Zuckerberg and Sundar Pichai have acknowledged that AI infrastructure is being overbuilt and that the spend is a risk rather than risk-free

Zuckerberg's framing is that if AGI comes it's a great investment and if it doesn't it may be a difficult one, but it's a risk they have to take

9:58 Lab leaders publicly acknowledge overbuild risk rather than treating spend as risk free

David Cahn · Aug 5, 2024

The hyperscalers constitute one of the most powerful oligopolies in business history and will therefore spend aggressively to defend it

Microsoft, Azure and Google represent $7TN of market cap, roughly 10% of global market cap, so defending that position justifies enormous spend

11:02 Oligopoly defense of market cap justifies aggressive hyperscaler capex

Brendan Foody · Sep 15, 2025

The AI market is much more like 1996 or 1997 than a late-stage bubble, so remaining bullish while focusing on fundamentals buys durability

Inefficiency is a one-way door in which it's easy to lose the right values and culture

50:58 Comparable to 1996 97 internet not a late stage bubble

Chris Degnan · May 23, 2026

Many current AI companies are reselling large language models at negative margin, which is a serious problem

39:04 Negative margin model reselling makes it unsustainable

Anastasios Angelopoulos · Aug 3, 2026

Investors are underwriting neolab seed rounds as near-zero-risk because the team alone could be acquihired for around a billion dollars

If a couple hundred million goes in and the team alone is worth a billion, the investment looks safe

39:37 Acquihire floor makes neolab seed prices rational not frothy

Anastasios Angelopoulos · Aug 3, 2026 · hedged

The compute debt build-out is worrying because open source could make cost savings salient enough to cut OpenAI and Anthropic enterprise revenue and lead to insolvency

If open source makes the cost-saving opportunity salient to enterprises, lab revenue falls while debt obligations remain

57:29 Open source substitution could cut lab revenue and break debt financed buildouts

Harry Stebbings · Aug 3, 2026

The industry has never been so dependent on two companies hitting their targets — if OpenAI and Anthropic falter, the whole ecosystem stops

Every downstream company in the AI stack depends on those two companies' trajectory continuing

57:54 Whole ecosystem hinges on two labs hitting their numbers

Anastasios Angelopoulos · Aug 3, 2026

High bandwidth memory and GPUs are extremely over-hyped across all stages from early stage to public markets

59:04 Memory and gpu layer is where the froth is concentrated

Immad Akhund · May 12, 2025

The only AI seed deals worth doing today are ones where a second-time founder deeply understands the domain and applies AI to it, or where the traction is already real enough to justify a high valuation

The vast majority of AI at seed has no traction and is the fifth team funded for the same idea; domain depth or proven traction is what separates real companies from AI-for-the-sake-of-it

37:38 Only second time domain experts or proven traction justify ai seed investment

Martin Casado · Jul 28, 2025

Investor behavior should follow business fundamentals, not marks; 2021 was a mistake because behavior followed marks, whereas today's AI growth genuinely justifies aggressive investing

In 2021 public markets and Tiger's capital deployment set valuations that drove behavior; today you have some of the fastest growing companies ever by users and revenue, and enormous value has shifted into the space, so following that is the job

5:56 Todays aggressive investing is justified by fundamentals unlike 2021s mark driven mistake

Deven Parekh · Oct 11, 2023

AI is currently in a hype cycle where valuations are way ahead of where the companies actually are, so it's right to be cautious this year despite spending heavy time on the space

Companies that raised as recently as six or nine months ago are already being impacted by things developed six months later, so the durability of these businesses is unproven relative to price

42:27 Ai valuations are ahead of fundamentals warranting near term caution

Daniel Dines · Dec 18, 2024

Investing $9 trillion of CapEx to unlock $9 trillion a year in GDP gains would be an easy investment if the outcome were predictable

The payback ratio is trivially good — the spend is small relative to a recurring annual return

35:07 9 trillion capex for 9 trillion annual gdp gain is easily justified if the outcome were predictable

Micha Kaufman · Jun 9, 2025 · hedged

The AI market will go through an investor-triggered cleanup in which the vast majority of opportunistic companies disappear, leaving a few foundational players whose offering may reduce to a commodity cloud/GPU-cycle play plus application layers on top

technology that is not free or $20 or $200 but truly expensive means only those already large or well-backed can afford it, which creates differentiators

26:08 Investor triggered cleanup leaves few foundational players amid commodity compute plus app layers

Victor Riparbelli · Jan 15, 2025

AI is in a bubble and a lot of money invested in AI products will be lost

Enterprise buyers don't really know what they want, so contracts are not real signal

0:00 Enterprise buyers dont know what they want so contracts are fake demand signals causing losses

Steeve Morin · Feb 24, 2025 · hedged

There is an H100 bubble that will burst, because the economics of buying H100s for inference don't work

The financing model was inherited from the A100 (train on the new generation, do inference on the last), but the H100 cost five times the price while delivering only about twice the inference performance — and at launch ran inference at the same speed as the A100 — so the money gap has to be bridged

9:32 H100 inference economics dont justify the price signaling a looming hardware bubble

Steeve Morin · Feb 24, 2025 · hedged

A compute oversupply has already begun and will worsen, with distressed GPU data centres potentially selling at around 30 cents on the dollar

he started receiving discount cold emails from unknown providers around October-November; running a real cloud is much harder than building a cluster for a training run, so many who built for training runs will be left with CapEx they cannot use, and the chips are the loan collateral

24:05 Distressed gpu data centres already signal emerging compute oversupply

Guillermo Rauch · Oct 6, 2023 · speculative

It is false that 99% of money going into AI companies today goes to zero — roughly 20% of AI investments will be productive and the era will produce many successful IPOs, as web 2.0 did

The mobile and social platform shift produced countless successful outcomes like Uber, Airbnb and Facebook, and the AI platform shift should follow the same pattern

49:22 Roughly twenty percent of ai investments will be productive not 99 percent failure

Imran Khan · Aug 26, 2024 · hedged

The central risk for Amazon and Google in building AI capacity is that today's demand is not sustainable and they get stuck with stranded capacity in five years

Amazon did exactly this in 2020, building massive capacity assuming COVID buying patterns would persist, and suffered huge margin pressure when they didn't; the analysis to do is whether the customers demanding capacity can pay long-term

41:51 Demand may not be sustainable leaving hyperscalers with stranded ai capacity

Harry Stebbings · May 9, 2026

The insane revenue ramps of consumer AI companies are distorting investors' judgment

85:11 Consumer ai revenue ramps distort investor judgment

Emad Mostaque · May 17, 2023

The flood of capital starts out good for the space but turns bad, producing grifters, a race dynamic where everyone builds their own models, and massive economic waste

It distracts from what actually needs doing — standardizing and feeding models better data

17:31 Capital flood starts good then turns bad producing grifters and wasteful race dynamics

Emad Mostaque · May 17, 2023

Spending on supercompute is not comparable to burning money on marketing — GPUs are currently a scarce appreciating asset, so there is no harm in scaling compute.

Chips with interconnects are unavailable on the market and companies come asking to rent his 7,000 A100s and TPUs; unlike scooter companies' marketing spend, this is a real asset, and at worst he could run it as a foundation-model-as-a-service business at below market rates and still make $100M profit.

44:36 Gpu compute spend is a real appreciating asset unlike marketing burn

Jonathan Ross · Sep 29, 2025

'Is there a bubble?' is the wrong question; the better question is what the smart money is doing, and the smart money is doubling down on AI

Google, Microsoft, Amazon and nation states keep raising their announced AI spend each time they update it, which reveals their conviction

5:23 Hyperscaler spending escalation reveals conviction not bubble

Jonathan Ross · Sep 29, 2025

The NVIDIA-OpenAI arrangement makes economic sense and the resulting stock-price gain exceeding the revenue is justified, because the revenue is likely to persist

Stock price rising more than revenue reflects belief that the revenue continues; that belief holds because of lock-in and because there simply isn't enough compute in the world

11:40 Stock gain exceeding revenue justified by lockin and scarcity

Jonathan Ross · Sep 29, 2025

Despite appearances, there is a lot of fear in and around AI right now, not greed — the apparent greed is just a handful of fast-moving people making a lot of money

You should be greedy when others are fearful and fearful when others are greedy; the visible activity comes from a small number of smart, aggressive actors rather than broad exuberance

73:06 Market sentiment is fear not greed despite apparent frenzy

Mark Goldberg · Oct 25, 2024

The oxygen is starting to come out of the AI bubble: simply being a '.ai' company no longer attaches the premium it did eighteen months ago and proportionality is returning to parts of the market

The same cohort of founders who were '.xyz' during the crypto boom and migrated to .ai are recognizing they will be judged on building a sustainable business

39:21 Ai label premium fading as market normalizes outside infrastructure hotspots

David Cahn · Oct 27, 2025

There is no coordinating mechanism that will stop AI spending or pop the bubble; the behavior is uncoordinated and incentive-driven, and will not change until the incentives change

The players are big sophisticated companies acting out their own incentives via the invisible hand; people wrongly assume coordination because it is easier for the brain to grok

23:21 No coordinating mechanism exists so uncoordinated incentive driven spending continues

David Cahn · Oct 27, 2025

Microsoft and Amazon have stepped back from absorbing AI demand risk, and the replacements — Oracle and CoreWeave — are too small to absorb as much risk

A year ago the hyperscalers explicitly fronted the risk with long leases and credit backing; Microsoft walking away from two data centers signalled it would no longer be the ecosystem's risk absorber, and Oracle and CoreWeave are much smaller balance sheets

24:16 Risk absorbing hyperscalers are stepping back and replacements are too small

Miles Dieffenbach · Aug 4, 2025

SpaceX cannot be killed by a capital markets downturn, unlike OpenAI, because it is self-funding

Starlink has reached escape velocity as a high-margin product and the company does secondary tenders rather than needing primary cash — the same way Google and Meta had 30-40% GAAP operating margins at IPO and were immune to markets

70:45 Self funding businesses are immune to capital markets downturns unlike cash burning labs

Bret Taylor · Oct 2, 2024 · hedged

With the benefit of hindsight, most of the excess of the dot-com bubble may have been justified

Many of today's top market-cap companies — Amazon, Google, PayPal, eBay, Salesforce — were founded in that period, and a huge share of thirty years of stock market gains came from those digital companies, even though people emotionally associate the era with Webvan and Pets.com

6:02 Dotcom bubble excess was justified in hindsight by the value created afterward

Bret Taylor · Oct 2, 2024

It is rational for large tech companies to keep pouring capex into AI, because no other option offers comparable ten-year revenue growth and the potential economic upside is enormous

The potential to create something this valuable for humanity and the economy justifies not being penny wise pound foolish; for a large company there is no better option for meaningful ten-year revenue growth

36:52 Capex is rational because no better option exists for decade scale revenue growth

Harry Stebbings · Mar 16, 2026

Paying $5B today for an agent business at $100M revenue prices in two years of tripling ahead of time, so the trade only works if public market multiples expand well beyond current levels.

At triple-triple growth the company reaches $900M revenue in two years, which is roughly what today's price already implies at public market pricing.

30:51 App layer prices already embed two years of tripling

Julien Bek · Aug 24, 2026

Rapid valuation step-ups are justified today because AI lets companies hit round-to-round milestones far faster than the old 18-month seed-to-A-to-B cadence

The milestones between rounds haven't changed, but the time to reach them has collapsed, so it's only normal to command a much higher valuation in a short time if you've proven yourself right

21:35 Compressed time to milestones justifies rapid step ups

Richard Socher · Apr 18, 2025

Cash can be a moat in AI but is not sufficient, and a correction is coming for companies trading at ~100x ARR with no real moat and near-zero switching costs

Some companies raised hundreds of millions and still died; if users can trivially switch to DeepSeek or another provider, the valuation isn't supported

45:54 High arr multiple companies without a moat face a coming correction

Harry Stebbings · Jul 27, 2026

Retail marketing of private-company SPVs, such as SpaceX SPVs advertised on Instagram Reels, is a signal of a dangerously heated market

Analogy to the London cab driver talking about Bitcoin price being the signal to sell

18:37 Retail marketed spvs are a top of market signal

Eiso Kant · Oct 7, 2024

Total spend and progress toward AGI are not the same thing and must be analyzed separately

Model creation is CapEx while running models is OpEx; if you spend $100 making a model that only ever returns $2-3 of value, the world will punish it and it won't exist

29:20 Capex must be judged against returned value not equated with agi progress

Jonathan Ross · Feb 17, 2025

Today's unprecedented AI capex is justified in a way past spending booms were not, because the value at the end is unusually clear to everyone

Unlike early search, where Google could hide the economics, everyone now knows how much money there is in AI, so everyone is racing for it

40:28 Capex is justified because the eventual value is unusually clear unlike past booms

Sridhar Ramaswamy · Feb 10, 2025 · hedged

Whether this is a 'good bubble' or a value-destroying one is genuinely unknowable right now, and hinges on whether the money goes into power and buildings (durable surplus) or into rapidly depreciating hardware (value gone in a poof)

The telecom bubble laid fiber the whole world later used; the dot-com grocery era just burnt money and we waited fifteen years for Instacart

38:00 Genuinely unknowable hinges on whether capital becomes durable infrastructure or depreciating hardware