All-In PodcastDeepSeek Panic, US vs China, OpenAI $40B?, and Doge Delivers with Travis Kalanick and David Sacks
At a glance
WHAT IT’S REALLY ABOUT
China’s DeepSeek Shocks AI World; DOGE Slashes U.S. Spending Fast
- This All-In episode weaves together three big threads: Travis Kalanick’s vision for automated food infrastructure, the shockwave from China’s DeepSeek R1 AI model, and the Trump administration’s early DOGE-driven spending cuts. Kalanick details how CloudKitchens is building real estate, software, and robotics to industrialize and personalize food at grocery-level prices, while drawing lessons from China’s hyper-innovative food and logistics ecosystem.
- David Sacks breaks down why DeepSeek’s open-sourced reasoning model matters geopolitically, why its $6M training-cost claim is misleading, and how likely model distillation from OpenAI’s systems raises both IP and cloud-security questions. The besties debate whether foundation models will commoditize, shifting value to applications, data moats, and specialized ‘mixture of experts’ architectures.
- They then zoom out to U.S.–China competition, export controls, and how constrained Chinese engineers innovated around CUDA and compute limits. Finally, they examine DOGE’s first 10 days—federal buyouts, RTO mandates, lease cancellations—and connect government austerity, interest rates, and AI-driven productivity to America’s fiscal survival.
- The episode closes with concerns about outdated aviation safety systems after the DC crash, arguing for automation and software upgrades in air traffic control and cockpit systems as another frontier where AI and modern engineering can save lives.
IDEAS WORTH REMEMBERING
5 ideasFood production is being re-architected as infrastructure: real estate + software + robotics.
Kalanick positions CloudKitchens as ‘AWS/NVIDIA for food’—owning delivery-only kitchen real estate, proprietary automation like the Bowl Builder, and software that lets brands run virtual restaurants. Restaurants prep ingredients in the morning, then leave; robots handle final assembly, labeling, bagging, and handoff via lockers. This decouples labor-intensive, on-demand cooking from actual order timing, cutting costs and enabling hyper-personalized, nutrition-tracked meals.
DeepSeek’s real innovation is less the $6M headline and more its technical workarounds under constraint.
Sacks and Chamath argue the $6M figure is just the final training run, not total R&D or hardware, which likely exceeds $1B in GPUs. The real breakthroughs: a new RL algorithm (GRPO) instead of industry-standard PPO and coding directly in NVIDIA PTX (low-level) instead of relying on CUDA. These choices likely emerged from compute and export constraints—and demonstrate how scarcity can force more radical optimization than Western teams awash in capital.
Model distillation is probably widespread, blurring IP lines and forcing cloud providers into a policing role.
Multiple top AI people Sacks spoke with believe DeepSeek heavily trained on OpenAI outputs. DeepSeek V3 sometimes self-identified as ChatGPT-4, strongly suggesting training on ChatGPT data via web scrape or API abuse. OpenAI has told the FT it found evidence of ‘improper’ use. This raises tough questions: How can Azure or other clouds detect and block abusive distillation at scale? How do you enforce IP in a world where any open-sourced frontier model (like o1) can be cloned, hosted, and undercut by the same partners who profit from you?
Foundation models are rapidly commoditizing; durable value likely shifts to data, distribution, and domain-specific systems.
The group converges on the view that large general models depreciate fast—Gavin Baker’s ‘fastest-depreciating asset’ line is cited. As performance equalizes and prices plummet, moats move to: (1) ‘shims’ that abstract away any single model and allow hot-swapping; (2) proprietary or structurally advantaged data (e.g., YouTube video, Tesla driving data, domain-specific corpora); (3) application-layer products with deep user integration, workflows, and trust. Chamath expects mixture-of-experts and many small specialized models to further erode the advantage of a single giant model.
China’s trajectory shows how copycatting evolves into genuine innovation and market leadership.
Kalanick recounts Uber’s China battles, where Didi copied every product feature in weeks, forcing Uber to build a 400-person China team in SF. Over time, once there’s ‘nothing left to copy,’ Chinese companies start leading in innovation—especially visible in food delivery, logistics, and locker-based last-mile systems. DeepSeek, Meituan, and BYD are presented as examples of this transition from imitators to frontier innovators, especially when pushed by export controls and resource constraints.
WORDS WORTH SAVING
5 quotesIf you can get that cost down to the cost of going to the grocery store, you do to the kitchen what Uber did to the car.
— Travis Kalanick
Constraint makes for great art. DeepSeek had to invent their way around compute limits, and they did things the West didn’t because we weren’t forced to.
— Chamath Palihapitiya
The fastest depreciating asset in the world is a large language model.
— Referenced by Jason Calacanis (attributing Gavin Baker)
Cheap AI makes cheap autonomy. As AI costs drop, autonomy just gets easier and easier.
— Travis Kalanick
If we took 2019 spend and put it up against 2024 revenues, we’d have a $500 billion surplus. Versus the $1.5 trillion deficit. That’s all waste.
— Chamath Palihapitiya
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome