Decoder with Nilay Patel

@decoderwithnilaypatel · Podcast

Latest segments

• Data center backlash, 2026-10-29
  Pivots from backlash to a compute glut
  Touches the backlash only as a premise - projects are slowing - then pivots to whether the build-out ends in a compute glut. Brookhart's answer is that shells can sit chipless because the chips are most of the spend. Analysis, not a new development.
• Gartner's neocloud bankruptcy call, 2026-10-29
  Explains which neoclouds survive the split
  Puts the neocloud shakeout question to SemiAnalysis's Joey Brookhart, who argues the split is backing: CoreWeave and Nebius survive on lab, hyperscaler, Nvidia or Broadcom support, while ex-crypto miners without it struggle to finance. Stays inside the Gartner and Musk framing already on the record.
• Anthropic's $2T IPO slip, 2026-10-29
  Reports Anthropic's warning on margin leverage
  Brookhart walks through the S-1's margin curve and says Anthropic told roadshow investors not to expect the current margin leverage to continue, because it wants to shift more compute toward training. Frames it as a Bezos-style deferral of profit.
• Anthropic's $2T IPO slip, 2026-10-29
  Breaks down revenue per inference megawatt
  SemiAnalysis's Brookhart explains where Anthropic's margins come from without its own chips: token throughput gains across Trainium, TPU and GPU fleets, and a revenue-per-inference-megawatt figure that has multiplied since last year. Adds the compute split between serving and training.
• Anthropic's $2T IPO slip, 2026-10-29
  Illustrates Tokenomics with the margin swing
  Brookhart explains SemiAnalysis's Tokenomics practice and uses Anthropic's margin swing as the illustration: negative 94% on inference two years ago, positive 85% now. The 85% figure is already on the record from the S-1; the segment stays inside the filing's numbers.
• AI spend concentration, 2026-10-29
  Adds the top 1% spending $50,000 per employee
  Asks where Anthropic's trillion-dollar revenue path comes from and gets a narrow-market answer: the top 1% of customers spend $50,000 to $100,000 per employee, Fortune 500 IT budgets are barely tapped, and Anthropic's spend is coding-specific. Cyber and bio are named as the next verticals.
• Anthropic's $2T IPO slip, 2026-10-29
  Reacts to the S-1 and the neocloud rally
  First reaction to the public S-1: Brookhart calls it an awesome filing, praising capital efficiency, gross margins and free cash flow, and notes compute and neocloud stocks rallied in the two days after. A broad read that stays inside the numbers already on the record.
• Anthropic's $2T IPO slip, 2026-10-29
  Previews the filing, listing and valuation
  Introduces the episode: the S-1 filing, a Nasdaq listing, a $1.5 to $2 trillion valuation and trading possibly next month, then hands off to SemiAnalysis's Joey Brookhart. Setup only, no new fact about the filing.
• SpaceX spectrum deal, 2026-10-17
  Recalls Dish and Sprint as cautionary history
  Reaches back to Dish's ORAN build and Sprint's small-cell promises as cautionary history for Starlink's carrier ambitions. Piecyk argues Dish's network worked and the failure was customer acquisition, which he says is Musk's strength. Stays inside the existing debate over whether SpaceX can build a network.
• SpaceX spectrum deal, 2026-10-17
  Argues AT&T and Verizon dismissed Musk
  Piecyk walks through the chain that put the 800 MHz block on the market: the T-Mobile-Sprint conditions, Ergen's failed bid, the forced auction. He argues AT&T and Verizon should have bought it to foreclose SpaceX and simply did not take Musk seriously.
• SpaceX spectrum deal, 2026-10-17
  Traces the block back to Nextel and Grain
  Traces the 800 MHz block back to Nextel, divested by Sprint in the T-Mobile deal and parked at Grain Management for years. Reads the FCC's EchoStar approval as requiring a terrestrial build, which explains why the spectrum finally had a buyer.
• SpaceX spectrum deal, 2026-10-17
  Breaks down the spectrum price and size
  Piecyk walks through the spectrum math behind the deal: 800 megahertz for $8 billion on top of the $20 billion EchoStar purchase, and the detail that SpaceX went back to EchoStar for a second buy after positive investor feedback. Adds the price and size of the new block.
• SpaceX spectrum deal, 2026-10-17
  Sets up SpaceX's wireless ambitions
  Opens the episode by laying out SpaceX's wireless ambitions: the new low-band spectrum deal, the FCC's approval of 15,000 next-gen Starlink satellites, and last year's $20 billion EchoStar purchase. Sets up the carrier question rather than reporting a new development.
• Paramount-Warner Bros. merger closes, 2026-10-10
  Walks through the $80 billion debt load
  Kafka walks through the balance sheet behind the close: the deal carries $80 billion in debt, and Zaslav's debt-reduction tenure never produced growth. A financial read on the same event, adding the leverage figure the card did not have.
• Paramount-Warner Bros. merger closes, 2026-10-10
  Recaps the chain of failed Warner owners
  Recaps the chain of failed Warner owners - AOL, AT&T, Discovery - and the pattern of cost-cutting without a revenue plan, then turns to the debt, the Gulf money and the Oracle backstop. Background and framing, not a new development.
• Paramount-Warner Bros. merger closes, 2026-10-10
  Traces the Ellison family's backing of Skydance
  Kafka walks through the Ellison family's backing and control of Skydance, then recaps the debt math behind the Warner deal and Zaslav's failed tenure. Background and framing, not a new development.
• Paramount-Warner Bros. merger closes, 2026-10-10
  Speculates on how Skydance will merge the studios
  Speculates about how Skydance will fold Paramount and Warner Bros. together: one mega streaming service, one movie studio, all run by David Ellison. It stays inside the deal's aftermath, with no new development on the record.
• Paramount-Warner Bros. merger closes, 2026-10-10
  Frames the deal that closed that day
  Opens the episode by framing the Skydance-Warner Bros. Discovery deal, which closed the day of recording, and sets up the question of whether David Ellison can succeed where AOL, AT&T and Discovery failed.
• Meta's Muse agent launch, 2026-10-08
  Questions how long Meta can fund free tokens
  Turns to the money question: how long Meta can fund free tokens and cloud computers, with Zuckerberg's Metaverse losses and quarterly earnings as the test. Hayden Field compares it to Uber's VC-subsidized rides and reports that some Meta employees think the company may move away from Muse later.
• Meta's Muse agent launch, 2026-10-08
  Debates who can afford the agent land grab
  Turns to the economics of the agent race: Zuckerberg's cut of transactions, the free cloud Linux box and NFL ad spend behind Muse, and OpenAI's inability to match it. The transaction-cut claim is already on the record; the rest is argument about who can afford the land grab.