Anthropic + SpaceX, Codex 100x'd Claude Code & Half Cash Half Stock | GMP EP12
Codex has 46 million downloads vs Claude Code's 491K — a 100x gap — and Anthropic just had to partner with SpaceX to avoid being rate-limited into irrelevance.
God Mode Podcast
Anthropic + SpaceX, Codex 100x'd Claude Code & Half Cash Half Stock | GMP EP12
Codex has 46 million downloads vs Claude Code's 491K — a 100x gap — and Anthropic just had to partner with SpaceX to avoid being rate-limited into irrelevance.
TL;DR
Rick, Luca, Ben, and guest Alan from the Philippines break down the biggest AI week of 2026: Anthropic signs a compute deal with SpaceX doubling rate limits, Codex outpaces Claude Code by 100x in downloads, and Anthropic stacks $380B in total committed compute. The crew digs into Jevons Paradox, Claude's new 'dreaming' memory agents, the PrintingPress.dev CLI ecosystem, and Ryan Cohen's viral 'half cash half stock' eBay bid. Key takeaway: the agent orchestration layer is becoming more valuable than the frontier model itself.
Anthropic partners with SpaceX for compute, Codex dominates Claude Code in downloads 100x over, and the agent layer continues to win. The hosts break down Anthropic's $380B compute stack, Claude's 'dreaming' memory update, the Jevons Paradox in AI token economics, Ryan Cohen's viral GME/eBay 'half cash half stock' play, and more. Guest Alan from the Philippines shares real-world Hermes agent use cases.
- Jevons Paradox
- An economic phenomenon where increased efficiency in resource use leads to greater total consumption — here applied to AI tokens: as inference gets cheaper, total token spend rises dramatically.
- MCP (Model Context Protocol)
- Anthropic's open protocol that allows AI models to connect to external tools, databases, and services, enabling agents to read and act on real-world data.
- CLI (Command Line Interface)
- A text-based interface where users type commands directly, preferred by developers over graphical UIs; increasingly relevant as agents interact with software via terminal commands.
- OAuth
- An open authorization standard that lets users grant apps access to their accounts without sharing passwords; used here to enable free model access through Codex.
- Orbital AI compute
- A proposed concept of hosting AI inference infrastructure on satellites in space, leveraging solar power and natural cooling to avoid terrestrial data center constraints.
- Colossus
- XAI and SpaceX's supercomputing cluster; Colossus 1 has ~100,000–200,000 Hopper GPUs and was offered to Anthropic; Colossus 2 uses newer Blackwell chips.
- TTFT (Time to First Token)
- The latency between sending a prompt and receiving the first token of a response from a model; lower TTFT means a faster-feeling user experience.
- Harness
- In AI, the orchestration layer or wrapper around a base model that manages prompts, memory, tool calls, and agent coordination.
- ICP (Ideal Customer Profile)
- A description of the hypothetical perfect customer for a product, used to target marketing and tailor user experiences.
- Anapurna Labs
- An Israeli chip startup acquired by Amazon in 2015 that now designs custom silicon for AWS; Amazon claims its chip business would generate $50B in annual revenue if standalone.
- Generative UI
- A design approach where the user interface is dynamically generated or personalized by AI based on the individual user's data and preferences.
- Kanban board
- A visual project management tool using columns (e.g., To Do, In Progress, Done) to track tasks; Hermes agent ships a built-in version to show agent task status.
- Vibe coding
- Colloquial term for using AI coding assistants to build software through natural language prompts rather than writing code manually.
- Machiavellian
- Characterized by cunning, scheming, and unscrupulous methods to achieve goals; used here to describe Ryan Cohen's calculated media strategy.
- Dreaming (Claude feature)
- Anthropic's agentic memory feature where Claude agents automatically update and consolidate their memory based on past conversations, analogous to how humans consolidate memory during sleep.
- Multi-agent orchestration
- The coordination of multiple AI agents working in parallel or in sequence on sub-tasks, with one agent often validating another's output.
- ZURP era
- Slang for the Zero/Ultra-low Real interest Rate Policy period (roughly 2020–2022) when cheap money fueled speculative investment in assets like meme stocks and NFTs.
- Arbitrage
- Profiting from a price or information gap between two markets or states; used here to describe exploiting the gap between token costs and the value of CLI efficiency tools.
Chapter 1 · 00:00
Cold Open & Introductions
It's incredibly reckless in this world to not learn AI.
Chapter 2 · 02:05
Codex vs Claude Code: 100x Download Gap
This guy built an agent that turns his X feed into a Mandarin listening class in 10 seconds.
Alan built an agent that pulls Chinese-language posts from X, generates a summary with vocabulary, converts it to audio via Google's TTS, and produces a structured Mandarin listening class — all in under 10 seconds. This is what useful AI agents actually look like.
Codex has 46 million downloads. Claude Code has 491,000. That's a 100x gap, and it's driving real behavioral change. Multiple users are cancelling Claude subscriptions mid-cycle after discovering Codex's speed and token efficiency.
Codex reached 46 million downloads while Claude Code had only 491,000 — roughly a 100x difference at the time of recording.
Chapter 3 · 06:55
Anthropic × SpaceX Partnership & Compute Deals
Anthropic peaked Monday. By Wednesday, the SpaceX deal made them unstoppable.
Alan estimated that a $20 Codex subscription delivers as much effective value as $100 spent on Claude Code, due to better token efficiency.
Following the SpaceX partnership, Anthropic doubled the five-hour rate limits for Claude Code Pro, Max, and Team plans and removed peak hour limits.
Anthropic looked like it peaked on Monday when Codex was eating its lunch. By Wednesday, the SpaceX compute deal dropped and doubled Claude's rate limits overnight. The fastest reversal in AI's short history.
Chapter 4 · 13:20
Orbital AI Compute & Anthropic's $380B Stack
Anthropic was dead on Monday or peaked on Monday. On Wednesday they're completely back.
Chapter 5 · 16:40
Claude 'Dreaming' & Multi-Agent Orchestration
Why build data centers on Earth when space has free solar power and zero cooling costs?
Space has no clouds, near-infinite solar power, and natural cooling. Elon isn't just selling GPUs to Anthropic — he's positioning SpaceX as the only entity that can build next-generation AI infrastructure without NIMBYism or energy grid constraints.
Chapter 6 · 22:10
Jevons Paradox & The Token Economy
Anthropic just committed $380 billion in compute deals. They now control 10% of US data center capacity.
Rick noted Anthropic's revenue grew approximately 90x in a single year, making it one of the fastest-growing companies in history.
Anthropic has stacked $380 billion in committed compute spend from nearly every major tech company. That represents 11.3 gigawatts — roughly 10% of all US data center capacity. When everyone holds equity, everyone roots for you to win.
Anthropic secured $380 billion in total committed spend across AWS, Google, Microsoft, Nvidia, FluidStack, SpaceX, and CoreWeave.
Anthropic's combined compute deals totalled 11.3 gigawatts of committed power, representing roughly 10% of the entire US data center capacity.
Chapter 7 · 28:20
SaaS Is Not Dead: The Bread-Baking Analogy
AI got 100x cheaper. The compute bill went up. This is Jevons Paradox in real time.
When inference dropped 100x in price, people didn't spend less — they used 10,000x more tokens. Jevons Paradox means every price drop in AI unlocks a new wave of demand, making the total compute bill explode upward.
Jevons Paradox in AI: inference got 100x cheaper but token consumption grew 10,000x, meaning compute bills actually went up.
Chapter 8 · 33:30
Amazon's Chip Business & Supply Chain Open Source
ATMs didn't kill bank tellers — they created more banks. AI won't kill developers either.
When ATMs arrived, the number of bank tellers went up — not down. Because the cost to run a bank fell, more banks opened and provided better service. The same dynamic is playing out in software: AI makes coding cheaper, so demand for software — and engineers — explodes.
Chapter 9 · 40:30
Agent Deep Dive: Hermes vs OpenClaw with Alan
Amazon has a secret $50B chip business most people have never heard of.
Amazon bought Israeli chip startup Annapurna Labs in 2015 and spent 11 years quietly building. The CEO now says that chip business would generate $50B in annual revenue if standalone — making it effectively the second-largest chip company in the world. US export rules keep it internal-only.
Amazon's CEO stated that if its Annapurna Labs chip business were standalone, it would have an annual revenue run rate of $50 billion.
OpenClaw is a Ferrari: blindingly fast, but breaks constantly and requires expert tuning. Hermes is a Honda: reliable, observable, and keeps chugging to the finish line. For most builders, reliability beats peak performance.
Chapter 10 · 50:00
PrintingPress.dev & The CLI Ecosystem for Agents
PrintingPress.dev is the new WordPress — but for AI agents plugging into the real world.
Most APIs, MCPs, and official CLIs waste agent tokens. PrintingPress.dev is building a library of efficient CLI tools that plug directly into agents — covering everything from Domino's pizza orders to Docker Hub. Luca called it the new WordPress.
Chapter 11 · 56:40
MCP, Chat as UX & The Future of Apps
Rick runs his entire podcast prep through Claude via MCP — one chat window replaces five apps.
Rick's podcast prep is now fully agent-driven: topic ideas arrive via Telegram, get stored in a database, and Claude reads them via MCP to generate introductions, outlines, and a custom chat dashboard. The future of productivity is one chat window with MCP.
Chapter 12 · 1:05:00
Fireside Round: Grok Build, Voice Agents & Cursor Orchestrate
The future of SaaS isn't one app for all — it's a different app for every customer.
Ben's Sales Echo already tailors onboarding flows, email sequences, and UI labels to each user's industry. The trajectory is clear: SaaS moves from generic to ICP-specific to fully personalized per individual. AI makes this possible without 10x the engineering work.
GPT-4o was the only OpenAI model trained natively on voice — predicting the next sound, not converting speech to text. The new GPT-5.5 Real-Time brings that voice-native training to the GPT-5 class. Ben says this is the most exciting voice AI development in years.
Chapter 14 · 1:17:00
Mark Liu, DataFast & Generative UI
Apple is putting Claude and Gemini in the iPhone — and cutting OpenAI out entirely.
Apple's iOS 27 will let users choose Claude or Gemini as their AI backbone — but not OpenAI. Ben says it's a privacy alignment issue: Anthropic and Apple share values around data, while OpenAI's full-capitalism model is antithetical to Apple's brand.
Ben noted that Google pays Apple approximately $10 billion per year to remain the default search engine in Safari.
Chapter 15 · 1:24:30
Layoffs & Stock Prices: Coinbase & Cloudflare
Cloudflare's stock fell 16% after hours following an announcement that it was laying off 100 employees.
Cloudflare's stock fell 16% after hours following an announcement that it was laying off 100 employees.
Chapter 16 · 1:26:40
Google Fitbit Air vs Whoop
Google's new Fitbit Air is priced at $99 with no subscription, directly undercutting Whoop's $260-per-year subscription model.
Google's new Fitbit Air is priced at $99 with no subscription, directly undercutting Whoop's $260-per-year subscription model.
Chapter 17 · 1:31:40
Neuralink Robot Surgery & Brain-Computer Interfaces
21 people can now type 40 WPM with their minds. The cyborg era has quietly begun.
21 people in 4 countries have a Neuralink chip. They can type 40 words per minute with their minds. This is just the medical phase. Once consumer adoption begins, the competitive advantage will be enormous — Rick said he'd hire the Neuralink candidate every time.
As of the episode, 21 people across four countries have received a Neuralink chip implant, with subjects able to type at 40 words per minute.
Neuralink-implanted users can type at 40 words per minute with their minds — matching or exceeding average typing speeds for most people.
Ben estimated Neuralink's last valuation at approximately $43 billion, calling it an investment he would make.
Chapter 18 · 1:36:55
SpaceX Starship Static Fire Test
Is this proof that if you can think it, you can do it. And Elon is just the living example of thinking or believing in something as possible and then actually designing everything around it to make it happen.
Chapter 19 · 1:39:20
Half Cash Half Stock: Ryan Cohen's GME/eBay Play
Ryan Cohen's eBay bid isn't crazy — it's a no-lose trade backed by 25 million derivative contracts.
Ryan Cohen didn't just make a meme — he built a no-lose trade. He secretly accumulated 25 million eBay derivative contracts, then went viral with a confusing media interview. If eBay acquisition happens, GameStop transforms. If not, he pumps the stock he already owns through derivatives.
Ryan Cohen argued eBay spent $3.2 billion on marketing and only acquired one million new users — a cost of $3,200 per new user.
Ryan Cohen's team reportedly accumulated close to 25 million eBay derivative contracts while only purchasing 25,000 actual shares.
No indexed bits in this chapter.
Show stoppers
Codex has 46 million downloads. Claude Code has 491,000. That's a 100x gap, and it's driving real behavioral change. Multiple users are cancelling Claude subscriptions mid-cycle after discovering Codex's speed and token efficiency.
Ryan Cohen didn't just make a meme — he built a no-lose trade. He secretly accumulated 25 million eBay derivative contracts, then went viral with a confusing media interview. If eBay acquisition happens, GameStop transforms. If not, he pumps the stock he already owns through derivatives.
Snapshots ()
Key Quotes ()
This episode
Claims & Sources
Factual claims made this episode, and whether a source was named.
Codex had 46 million downloads compared to Claude Code's 491,000 — a roughly 100x difference.
Anthropic doubled Claude Code's five-hour rate limits for Pro, Max, and Team plans and removed peak-hour limits following the SpaceX partnership.
Anthropic's total committed compute spend across all partners reached $380 billion, covering 11.3 gigawatts of power.
11.3 gigawatts of committed power represents approximately 10% of total US data center capacity.
Google has a deal to pay Apple approximately $10 billion per year to be the default search engine in Safari.
Amazon's Annapurna Labs chip business would have a $50 billion annual revenue run rate if it were a standalone company.
Amazon acquired Annapurna Labs in 2015.
Neuralink has implanted chips in 21 people across four countries.
Neuralink-implanted users can type at 40 words per minute using the neural interface.
Neuralink's last valuation was approximately $43 billion.
Google's Fitbit Air is priced at $99 with no subscription, compared to Whoop's $260-per-year subscription.
eBay spent $3.2 billion on marketing but only acquired 1 million new users, equating to $3,200 per new user.
Ryan Cohen's team accumulated approximately 25 million eBay derivative contracts while purchasing only 25,000 actual eBay shares.
Ryan Cohen was a co-founder of Chewy.com and made $5 billion from it.
Grok and XAI are approximately six months behind competitors when it comes to coding capabilities.
Colossus 1, the XAI/SpaceX cluster offered to Anthropic, contains approximately 100,000 to 200,000 Hopper-generation GPUs, while Colossus 2 uses 500,000 newer Blackwell chips.
This episode
Cast
-
Discussed as the architect behind SpaceX's Anthropic deal, orbital compute plans, and his long-term $10 trillion net worth ambition.
-
GameStop CEO who went viral with his 'half cash half stock' eBay acquisition offer; discussed as a viral marketing and derivatives trading genius.
-
Central topic of the episode — discussed regarding its SpaceX compute partnership, $380B capital stack, Claude 'dreaming' feature, and competitive position vs OpenAI.
-
Announced a compute partnership with Anthropic, providing Colossus 1 GPU access while exploring multi-gigawatt orbital AI compute together.
-
Discussed as Anthropic's main competitor, maker of Codex and GPT-5.5; described as competing alone without the partner network Anthropic has built.
-
Track
Target of Ryan Cohen's viral 'half cash half stock' acquisition offer; discussed for its poor marketing ROI and potential synergies with GameStop's physical locations.
-
Track
Ryan Cohen's company at the center of the viral eBay acquisition bid; discussed for its meme culture ties and potential strategic value merging physical retail with digital auctions.
-
Track
Major Anthropic investor with a 17% stake and $100B compute deal; also discussed for its Annapurna Labs chip business with a $50B implied annual run rate.
-
Track
Invested $40B equity in Anthropic as part of a $200B compute deal; also discussed for Gemini integration in Apple iOS 27 and the new Fitbit Air launch.
-
Brain-computer interface company with 21 implants across 4 countries; users can type 40 WPM; discussed as a future consumer computing platform valued at ~$43B.
-
Elon Musk's AI company; discussed as the provider of Colossus 1 compute to Anthropic, and as maker of Grok and GrokVoice; characterized as 6 months behind in coding.
-
Fitness tracker company discussed as under competitive threat from Google's Fitbit Air, which offers similar tracking at $99 with no subscription vs Whoop's $260/year.
-
Israeli chip startup acquired by Amazon in 2015 that now designs custom silicon for AWS with an implied $50B annual revenue run rate if standalone.
-
Track
Stock crashed 16% after hours when it announced laying off 100 employees, representing a reversal of the recent trend where layoffs boosted stock prices.
-
OpenAI's coding agent that achieved 46 million downloads versus Claude Code's 491K, discussed as the dominant AI coding tool of the moment.
-
AI agent platform by Nous Research compared to a reliable Honda; Alan uses it for language learning, co-living feedback, and other automations.
-
Anthropic's coding agent discussed as losing ground to Codex due to rate limits and token inefficiency, though Anthropic doubled its rate limits via the SpaceX deal.
-
AI coding IDE discussed for its new /orchestrate feature that recursively spawns agents; also referenced in SpaceX's partnership context as a major Anthropic customer.
-
A CLI tool library for agents that replaces token-inefficient APIs and MCPs; Luca called it 'the new WordPress' for connecting agents to real-world services.
-
Google's new no-subscription fitness tracker priced at $99, described as a direct Whoop competitor powered by Gemini.
Stats
