Show Notes
Inference is becoming more important than pre-training in AI chips, including how pre-fill and decode work and why more compute is shifting toward serving models. Today we walk through Etched’s ASIC system for transformer inference, its claims around efficiency and throughput, and the tradeoff between specialization and general-purpose GPUs like NVIDIA’s. We also look at custom chip efforts from companies like OpenAI, Google, and Amazon, and argues that inference demand may keep growing as AI agents and long-running workloads expand. ------ 🌌 LIMITLESS HQ ⬇️ EMAIL US: [email protected] NEWSLETTER: https://limitlessft.substack.com/ FOLLOW ON X: https://x.com/LimitlessFT SPOTIFY: https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQ APPLE: https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890 RSS FEED: https://limitlessft.substack.com/ ------ TIMESTAMPS 0:00 Inference’s New Frontier 2:14 Training Versus Inference 5:19 Etched’s Bold Bet 7:58 Building the Whole Rack 10:48 TSMC and the Hard Problems 13:29 Why Inference Matters 14:59 The Transformer Risk 17:02 OpenAI’s Jalapeno Chip 18:59 Why Accelerators Keep Winning 22:28 The Market Is Underpricing It 23:10 NVIDIA Is Still in the Game 24:56 Vertical Integration Wins ------ RESOURCES Josh: https://x.com/JoshKale Ejaaz: https://x.com/cryptopunk7213 ------ Not financial or tax advice. See our investment disclosures here: https://www.bankless.com/disclosures Josh works with Anthropic as a contractor. All views expressed are his own and do not represent Anthropic, its leadership, or its affiliates. Nothing in this episode is investment advice.
More from In the Loop AI
Four months ago, half of Wall Street wrote off SpaceX's record IPO as massively overvalued. Last night it reported $14.6 billion in revenue in a single quarter and the first profit in its history. The surprise is where the money came from: AI. SpaceX has become the number one neocloud, renting compute to Anthropic and Google for billions of dollars a month, and its AI business now out-earns Starlink. Ejaaz breaks down the xAI pivot, how Elon brings 200,000 GPUs online a month, the power bottleneck, and Elon's projection of a $500 billion to $1 trillion annual run rate in 2027.
One chart and one five-letter reply ("Yikes") sent AI stocks into their longest losing streak since 2008, with CoreWeave's bond yield touching 15% and roughly $1.6 trillion knocked off chip stocks. Ejaaz breaks down what actually happened: how neoclouds like CoreWeave borrow to build data centers, why the market got spooked, and whether credit default swaps are really pricing a coin flip.
Anthropic just pledged 5% of its shares to a new Public Benefit Fund, and with its founders' and employees' earlier pledges, roughly 20% of the company is now committed to "the public good." The fund could make "direct distributions to specific populations," and the reason why is more unsettling than the gift: Anthropic expects AI to disrupt a lot of jobs. Today we break down what UBI actually is, how Sam Altman, Dario Amodei and OpenAI got here, how Anthropic's fund works, why David Sacks, Palmer Luckey and Travis Kalanick are skeptical, Elon's plan to give every American kid a SpaceX share, and what a world on AI money could look like. == Timestamps 00:00 Anthropic Gave Away 5% 1:50 What UBI Actually Is 2:22 Altman's Equity Fund 3:22 Dario's Jobs Warning 4:10 How Anthropic's Fund Works 8:40 The Backlash 11:53 Keep the Money? 12:28 Elon's Counteroffer 16:49 Nobody Knows Where This Ends 19:33 Universal High Income 22:09 Final Thoughts #anthropic #ubi #ai #elonmusk #claude == AlphaSpace by Yahoo Finance: A Professional-Grade Investment Platform Built for Everyday Investors. 👉 https://finance.yahoo.com/about/promos/gold/alphaspace/?ncid=100003571
Tesla finally unveiled the new Roadster, and it drove upside down on a ramp, flipped through the air and landed using SpaceX-developed thrusters. It's also the clearest look yet at what a Tesla–SpaceX merger could build, and Barclays says that merger could be announced before December. Today we break down the three tricks behind the flying Roadster, why Elon is really building it, and the rest of the week in AI: Anthropic's slowing growth ahead of its IPO, Meta open-sourcing Muse Spark, and Google's Gemini 4 Argon. == Timestamps 0:00 Tesla's Flying Roadster 1:19 Watching the Flip 2:12 Trick 1: Fan Suction 3:30 Trick 2: SpaceX Thrusters 6:04 Trick 3: Track Mode Wings 8:01 Price and Trims 8:40 The Real Reason: A Merger 10:35 How the Deal Could Work 11:07 The Critics vs. Elon 12:58 Anthropic's IPO Slowdown 16:10 AI That Improves Itself 17:49 Meta's Open-Source Comeback 21:12 Google's Gemini 4 Argon 24:12 Final Thoughts #tesla #spacex #elonmusk #roadster #ai == AlphaSpace by Yahoo Finance: A Professional-Grade Investment Platform Built for Everyday Investors. 👉 https://finance.yahoo.com/about/promos/gold/alphaspace/?ncid=100003571
OpenAI’s own AI helped catch a spy inside its headquarters. An employee allegedly hid a $24 recording device behind a TV in an executive conference room, and the AI tools he used to plan it became the trail that led the FBI to him. Today we break down how the device was found, how an internal agent and a chatbot allegedly helped build the plan, the charge he faces, and why the frontier labs are turning into national security sites. == Timestamps 0:00 ChatGPT Catches a Spy 1:21 The Bug Behind the TV 2:51 OpenAI’s Model Investigates 5:34 Who Planted It 6:10 Planning With AI 9:04 Caught on Camera 10:13 How AI Cracked the Case 12:20 Your Digital Footprint 13:26 The Charge and the Sentence 14:15 AI as National Security 16:32 Will Washington Step In? 18:33 OpenAI vs. Apple 19:08 Final Thoughts #openai #chatgpt #ai #cybersecurity #fbi #nationalsecurity == AlphaSpace by Yahoo Finance: A Professional-Grade Investment Platform Built for Everyday Investors. 👉 https://finance.yahoo.com/about/promos/gold/alphaspace/?ncid=100003571