All-In: Datacenter Wars

Category: Expert Interviews · Duration: 91 min · ▶ Watch

Speakers: Travis Kalanick

Switch language → 中文

Segments (31)

  • 00:00:00 · Introduction & Guest Travis Kalanick
    • The hosts welcome guest Travis Kalanick and begin the podcast.
  • 00:00:25 · NYC’s Pied-à-Terre Tax
    • The hosts discuss the proposed pied-à-terre tax in New York City, its potential percentage, and its impact on homeowners.
  • 00:02:07 · Doxxing, Dog Whistles, and Political Rhetoric
    • The panel debates whether a politician pointing out Ken Griffin’s penthouse constitutes doxxing and discuss the dangers of such political tactics.
  • 00:03:34 · Economic Impact of the Pied-à-Terre Tax
    • The hosts analyze how the tax targets the most elastic part of the real estate market and will likely kill demand and crash the market.
  • 00:04:15 · Housing Policy: NYC vs. London vs. Austin
    • A comparison of housing policies, highlighting how London’s similar taxes hollowed out the city, while Austin’s pro-development stance lowered rents despite population growth.
  • 00:07:10 · London’s Stamp Tax as a Cautionary Tale
    • Chamath explains how London’s stamp tax caused the high-end real estate market to collapse, predicting a similar fate for NYC.
  • 00:08:35 · The Role of ‘Whale’ Buyers in Real Estate
    • Sacks argues that high-end, price-insensitive buyers help underwrite new construction projects, and taxing them away will stifle development.
  • 00:09:08 · Congressional Stock Trading Performance
    • The hosts marvel at a chart showing Ro Khanna and Nancy Pelosi significantly outperforming the S&P 500 with their stock trades.
  • 00:09:36 · The AI Compute Constraint
    • Chamath introduces the main theme: the entire AI industry is massively compute-constrained, affecting everything from power to data center construction.
  • 00:10:02 · Public Sentiment Shifts Against AI
    • The panel discusses the growing negative public and political sentiment towards AI and data centers, leading to development bans.
  • 00:10:30 · The Eric Swalwell Scandal
    • Friedberg and Sacks discuss the coordinated release of allegations against Eric Swalwell, suggesting it was a political move by the Democratic establishment to clear the field for the California governor’s race.
  • 00:11:24 · OpenAI’s Identity Crisis & Strategy Shift
    • Jason introduces the topic of OpenAI’s internal struggles, citing a leaked memo where they attack competitor Anthropic and pivot towards enterprise.
  • 00:13:50 · Debate: OpenAI’s Focus - Consumer vs. Enterprise
    • The hosts debate whether OpenAI should focus on its dominant consumer product (ChatGPT) or pivot to the more scalable enterprise and coding markets.
  • 00:15:23 · The AI Race: Valuation and Growth
    • Travis Kalanick argues that growth is the single most important factor in the AI race, and the company growing fastest will win due to network effects.
  • 00:17:37 · The AI Flywheel: Growth, Revenue, and Compute
    • Friedberg and Travis discuss the AI flywheel, where user growth and innovation are interconnected, and how Anthropic’s rapid release cadence gives them momentum.
  • 00:19:22 · Capital as a Weapon and the Efficiency vs. Subsidy Debate
    • Travis explains that while massive capital raises are impressive, efficiency will ultimately outstrip subsidy, and the company funding growth with revenue has the long-term advantage.
  • 00:21:28 · The Enterprise AI Opportunity
    • Sacks argues that the biggest and most scalable revenue opportunity in AI is in enterprise, particularly coding, which OpenAI should be more focused on.
  • 00:25:22 · Physical Limits to AI Growth
    • Sacks points out that Anthropic’s exponential growth will inevitably hit physical limits related to compute, power, and data center availability.
  • 00:26:42 · The Challenge of Building with AI Agents
    • Chamath and Jason discuss the difficulty of building real, scaled products using AI agents, noting that current implementations are often just ‘vibe-coded slop’.
  • 00:27:31 · Frontier Labs’ Dilemma: Infrastructure and Business Models
    • Chamath outlines the two major problems for frontier labs: their dependency on hyperscalers for compute and the challenge of finding a profitable enterprise business model.
  • 00:30:26 · The AI Infrastructure Arms Race
    • The panel discusses the massive scale of new GPU clusters being built by Tesla (Colossus) and Meta (Prometheus) and Elon Musk’s entry into the data center business.
  • 00:31:31 · The Strategy Behind Holding Back AI Models
    • Sacks explains the theory that Anthropic held back its most advanced model (Mythos) due to compute constraints, turning a limitation into a successful marketing event.
  • 00:32:32 · Game Show Interlude: The Price is Wrong
    • Jason hosts a game show segment called ‘The Price is Wrong,’ where the panel guesses overvalued startups from the ZIRP era.
  • 00:33:56 · Allbirds’ Pivot to AI
    • The hosts discuss the absurdity of shoe company Allbirds pivoting to AI, which caused its stock to surge, as a prime example of peak bubble behavior.
  • 00:39:27 · The Real Story: Compute Constraints Driving Markets
    • Chamath argues that seemingly strange market moves, like Allbirds’ pivot, are canaries in the coal mine indicating a massive, underlying compute constraint.
  • 00:45:45 · Populism and the War on Data Centers
    • Friedberg posits that data centers have become a physical manifestation and target for populist anger against the perceived ‘wealthy tech elite’.
  • 00:48:17 · The Rate Payer Protection Pledge and Utility Business Models
    • Chamath explains that while the Rate Payer Protection Pledge is a good step, it doesn’t change the fundamental business model of utilities, which are incentivized to keep building and raising rates.
  • 00:49:50 · The Economic Impact of Data Center Construction
    • Sacks counters the claim that data centers don’t create jobs, highlighting the boom in high-paying, blue-collar construction jobs.
  • 00:52:01 · Geopolitics of Data Centers
    • Sacks discusses the controversy around building US-backed data centers in the GCC, arguing it was a strategic move that was unfairly maligned as serving China.
  • 00:53:20 · The AI Industry’s Messaging Problem
    • Jason laments that the primary spokespeople for AI are perceived as a doomsayer (Dario Amodei) and a sociopath (Sam Altman), creating a negative public image for the industry.
  • 00:54:39 · Game Show: Name That Startup!
    • Jason launches a new game show segment, ‘Name That Startup,’ where the panel has to identify overvalued startups from the past.

Specific Prices (25)

Timestamp Item Value Context
00:00:35 Proposed NYC pied-à-terre tax 3.9% Speculated annual tax rate on second homes in NYC.
00:00:59 NYC pied-à-terre tax threshold $5,000,000 The tax applies to any home valued over $5 million.
00:04:03 Cost of a $10M unit with tax $20,000,000 Hypothetical total cost to break even on a $10M unit after about a decade with the new tax.
00:11:40 Anthropic’s alleged revenue inflation $8,000,000,000 OpenAI’s memo claimed Anthropic’s $30B run rate was inflated by $8B due to revenue sharing.
00:11:41 Anthropic’s stated run-rate revenue $30,000,000,000 The revenue figure that OpenAI’s memo disputed.
00:12:59 OpenAI’s valuation questioned by investors $852,000,000,000 The valuation of OpenAI that some investors are now questioning due to strategy shifts.
00:15:42 Hypothetical OpenAI IPO valuation needed $1,200,000,000,000 An investor stated OpenAI would need a $1.2T IPO for the last round to make sense.
00:15:49 OpenAI’s last private valuation $850,000,000,000 The valuation at which OpenAI’s last round closed, for which there are currently no buyers.
00:18:23 Sam Altman’s reported fundraising round $122,000,000,000 The massive amount of capital Sam Altman reportedly closed for a new venture.
00:23:09 Anthropic’s ARR growth $1B to $10B Anthropic’s annual recurring revenue grew from $1B to $10B in the last year.
00:23:16 Anthropic’s revenue by end of Q1 $30,000,000,000 Anthropic’s reported revenue run-rate by the end of Q1.
00:23:27 Anthropic’s projected year-end revenue $80B to $100B The projected revenue for Anthropic by the end of the current year.
00:24:28 Consumer AI subscription price $20/month The typical ‘all-you-can-eat’ subscription price for consumer AI products like ChatGPT.
00:34:28 Allbirds IPO fundraising $350,000,000 The amount Allbirds raised in its IPO.
00:35:16 Allbirds peak valuation $4,000,000,000 The peak market capitalization of Allbirds.
00:39:05 Newbird AI convertible note $50,000,000 The size of the convertible note for Allbirds’ pivot into ‘Newbird AI’.
00:39:40 Jane Street investment $1,000,000,000 A recent investment by Jane Street into a neoscaler company.
00:39:47 Jane Street compute deal $6,000,000,000 A compute deal associated with the Jane Street investment.
00:41:39 Missouri AI data center value $6,000,000,000 The value of an AI data center build that was approved in a small Missouri town, leading to a political backlash.
00:50:57 Economic value of contested data centers $162,000,000,000 The total economic value of 100 data centers currently being contested across the country.
00:55:20 OpenSea peak valuation $13,000,000,000 The peak valuation of NFT marketplace OpenSea.
00:56:51 Clubhouse peak valuation $4,000,000,000 The peak valuation of audio social network Clubhouse.
00:57:35 Theranos fundraising $900,000,000 The amount of capital raised by Theranos.
00:57:38 Theranos peak valuation $9,000,000,000 The peak valuation of the fraudulent blood-testing company Theranos.
01:28:37 Quibi fundraising $1,700,000,000 The amount of capital raised by short-form streaming platform Quibi.

Bottleneck Claims (5)

  • [00:39:55] The AI industry is massively compute-constrained.
    • Evidence: Market transactions like Allbirds pivoting to AI, Jane Street’s large compute deal, and the soaring stock of power solution companies like Bloom Energy all point to a desperate need for more compute.
  • [00:40:47] Public and political opposition (NIMBYism) is a major bottleneck to building new data centers.
    • Evidence: A Missouri town ousted its city council after approving a data center, and Maine passed a bill to ban all new data center construction, showing a growing tide of local resistance.
  • [00:48:17] The business model of regulated utilities is a bottleneck to expanding the power grid for AI.
    • Evidence: Utilities are incentivized to make capital investments to earn a guaranteed return, which means they will continue to build and pass costs to consumers, but this process is slow and doesn’t align with the rapid needs of AI.
  • [01:23:54] Change management is the biggest bottleneck to AI adoption in large enterprises.
    • Evidence: Travis Kalanick argues that getting the existing workforce, from middle managers to bureaucrats, to adopt new AI-driven processes is a very difficult human problem, often harder than the technology itself.
  • [01:27:31] Frontier AI labs are hitting a wall because they are dependent on hyperscalers for compute.
    • Evidence: Chamath argues that as these labs scale, renting compute becomes a strategic dependency and a huge liability, forcing them to get into the business of building their own infrastructure.

Predictions (3)

  • [00:01:37, Unspecified] The NYC pied-à-terre tax will crash the whole real estate market.
  • [00:47:26, Unspecified] Data centers will likely be banned in 30 states.
  • [00:12:50, Short-term (within a week)] GPT-5.5 will be released by the end of April.

Key Technologies (5)

  • Large Language Models (LLMs): The core AI technology behind companies like OpenAI and Anthropic, used for generating text, code, and other content.
  • AI for Coding (Codex, Claude): Specialized AI models designed to write, debug, and understand computer code, seen as a major enterprise use case.
  • AI Agents: Autonomous AI systems designed to perform tasks. The panel discusses their current limitations and potential.
  • GPU Clusters / Data Centers: The physical infrastructure required to train and run large-scale AI models, identified as a major industry bottleneck.
  • Fuel Cells (Bloom Energy): An on-site power generation technology that allows data centers to be built without relying on the existing, constrained power grid.

Companies Mentioned (27)

Uber · Amazon · Anthropic · OpenAI · Perplexity · xAI · Google (Gemini) · Meta · Apple · Open-Claw · Cursor · Tesla · SpaceX · Friendster · MySpace · Facebook · Stargate · Allbirds · Barnes & Noble · Bird · Bloom Energy · Crusoe · CoreWeave · OpenSea · Clubhouse · Theranos · Quibi

Notable Quotes (12)

The King of Atoms, yes, Captain Travis Kalanick is here. — Jason Calacanis @ 00:00:09

This is not a rich person tax. This is, within 15 miles of midtown Manhattan, you’re paying an extra tax. — Jason Calacanis @ 00:01:03

The most elastic part of the market is what they’re targeting for this tax. — David Sacks @ 00:01:18

It’s a dog whistle… to say that’s the next United Healthcare CEO. — Jason Calacanis @ 00:02:43

You have ChatGPT, a 1bn-user business growing 50-100 per cent a year, what are you doing talking about enterprise and code? It’s a deeply unfocused company. — Anonymous OpenAI Investor (quoted by Jason) @ 00:13:20

Growth is the whole damn thing. — Travis Kalanick @ 00:16:06

Efficiency will outstrip subsidy. — Travis Kalanick @ 00:19:53

You can plot their revenue on a logarithmic graph. I mean, again, no one’s ever seen anything like it before. — David Sacks @ 00:23:33

This is all a bunch of vibe-coded slop. — Chamath Palihapitiya @ 00:27:01

It is the temple of the wealthy. — David Friedberg @ 00:46:06

Key Topics

NYC Real Estate Taxes · AI Industry Competition (OpenAI vs. Anthropic) · AI Compute and Infrastructure Bottlenecks · Public and Political Sentiment towards AI · Data Center Development and Opposition · Enterprise vs. Consumer AI Strategy · Political Scandals and Establishment Power (Eric Swalwell, Joe Biden) · Market Bubbles and Startup Valuation (ZIRP era) · Congressional Stock Trading

Takeaways

  • The proposed ‘pied-à-terre’ tax in NYC is seen as a poorly designed policy that will likely harm the real estate market by deterring the most mobile and high-spending buyers.
  • The AI industry is in a fierce arms race, with growth and access to compute being the most critical factors for success. Anthropic is currently growing faster than OpenAI, causing concern among OpenAI investors.
  • A massive, underlying bottleneck in the AI revolution is the availability of compute, specifically power and physical data centers. This constraint is being exacerbated by growing political and public opposition (NIMBYism).
  • The public sentiment towards AI is becoming increasingly negative, fueled by fears of job loss and populist resentment against the tech elite. Data centers have become a physical target for this anger.
  • The business model for AI is still being figured out. While consumer AI has broad reach, enterprise applications, especially for coding, appear to have a more scalable and profitable, metered revenue model.
  • Political establishments, like the Democratic party, appear to wield significant power to coordinate messaging and clear the field for preferred candidates, as evidenced by the recent scandals involving Eric Swalwell and Joe Biden’s withdrawal.
  • The ZIRP (Zero Interest Rate Policy) era created market delusions where companies with poor underlying economics (like Allbirds and other physical goods companies) were valued like high-margin software businesses, leading to inevitable collapses.