Approximately 5% of top Anthropic users account for roughly 80% of all token usage on the platform.
Claude Opus 4.8, browser in Codex, tokenmaxxing, and the upcoming AI IPO bear case nobody wants to hear
SpaceX is filing for an IPO at 100x revenue with a claimed TAM larger than China's entire GDP — and three 2026 AI IPOs together would dwarf every US IPO combined from 2015–2025.
May 29, 2026
1:27:25
Difficulty: Intermediate
Played
God Mode Podcast
Claude Opus 4.8, browser in Codex, tokenmaxxing, and the upcoming AI IPO bear case nobody wants to hear
SpaceX is filing for an IPO at 100x revenue with a claimed TAM larger than China's entire GDP — and three 2026 AI IPOs together would dwarf every US IPO combined from 2015–2025.
May 29, 2026
1:27:25
Difficulty: Intermediate
Played
Chapter
No indexed bits in this chapter.
Snapshots ()
Stats
Episode stats
Insight Overview
insights
chapters
Insight distribution
Sub-Categories
Speaker breakdown
Talk Time
Key Quotes ()
This episode
Cast
-
Discussed as the key factor behind SpaceX's premium valuation and retail investor enthusiasm, with debate on whether the 'Elon premium' justifies 100x revenue.
-
Referenced as the source of the 'browser inside Codex' prediction and the insight about Claude behaving like a junior dev vs. Codex's senior engineer approach.
-
Indie developer who went viral after being banned from Google Maps for leaving honest reviews; discussed in the context of fake reviews and AI replacing review platforms.
-
Cited as representative of institutional skepticism toward Elon Musk's ventures despite his repeated success.
-
Central to episode's IPO bear case analysis; discussed as Anthropic's GPU compute provider and upcoming massive IPO candidate.
-
AI lab discussed in the context of raising funds at a valuation higher than OpenAI and signing a compute deal with SpaceX.
-
Compared to Anthropic on valuation and model performance; mentioned as a future IPO candidate alongside SpaceX and Anthropic.
-
Referenced as benchmark for GPU hardware (H200, Blackwell) and as a valuation comparison at 25x revenue vs. SpaceX's 100x.
-
Used as a valuation benchmark at 16x revenue to contrast with SpaceX's 100x multiple and illustrate the scale of the Elon premium.
-
Cited as the company that reportedly accidentally spent $500M on AI tokens in one month and mentioned as a beneficiary of Blue Origin's progress.
-
SpaceX's satellite internet subsidiary; discussed as one component of SpaceX's fundamental valuation at approximately $150B.
-
Jeff Bezos's rocket company; its recent explosion was played on-air and used to discuss the difficulty of space launch and the bull case for SpaceX.
-
YC's Paperclub event in Mountain View was Luca's content of the week recommendation, featuring top researchers discussing inference scaling.
-
OpenAI's coding agent environment discussed as a competitor to Claude Code, with browser integration seen as a breakthrough feature.
-
New Anthropic model release discussed as a modest post-training update whose main improvement is a reduced dishonesty rate.
-
Cloud-based coding environment used by Luca, compared to Claude Code and Codex on cost, convenience, and hosting model.
-
xAI's model discussed as having one of the lowest hallucination rates and used by Luca for fast research alongside Gemini.
-
Google's AI model used by Luca for fast, top-level research as a cheaper and faster alternative to Claude Opus.
-
Database service used as the key example of how browser-in-agent environments will automate login and API key management.
This episode
Claims & Sources
2 / 18 cited (11%)
Factual claims made this episode, and whether a source was named.
Anthropic raised money at a higher valuation than OpenAI in its most recent funding round.
Claude Opus 4.8 reduced its dishonesty rate from 19.7% to 3.7% compared to Opus 4.7.
H200 GPU rental spot prices peaked at approximately $7 per hour and have since fallen to $4.11 per hour.
The underlying provider cost to run an H200 GPU is roughly $2–$3 per GPU per hour.
Anthropic's compute deal with SpaceX costs approximately $1.25 billion per month.
Back-of-envelope math implies Anthropic pays approximately $8 per GPU per hour to SpaceX under their compute deal.
SpaceX early employees who joined before the first successful launch stand to make over $100 million each at the $1.75T IPO valuation.
Three upcoming 2026 AI IPOs (SpaceX, OpenAI, Anthropic) collectively aim to raise $550 billion, exceeding all US IPOs combined from 2015 to 2025.
SpaceX's $1.75 trillion IPO valuation implies approximately 100x revenue, compared to NVIDIA at 25x and Tesla at 16x.
SpaceX claims a total addressable market of $28.5 trillion in its S-1 filing, larger than China's $19 trillion GDP.
SpaceX is allocating 30% of its IPO to retail investors, approximately six times the typical 5% retail allocation.
New US regulations have shortened the institutional investment window for IPOs from approximately three months to fifteen days.
Institutions typically take profit at 40% gains on hot IPOs and 17-20% on average IPOs.
Amazon reportedly accidentally spent $500 million in a single month on AI tokens after failing to set usage limits on employee cloud accounts.
In companies rushing to maximize AI token spend, 44% of spend goes to fixing AI-created bugs, 27% to rework, 11% to reviewing, and only 18% reaches a shipped product.
SpaceX's Anthropic deal is expected to generate approximately $15 billion in annual revenue for SpaceX.
Data center construction permits are being challenged in the US, with some campaigns allegedly funded by Chinese interests.
TL;DR
Rik, Ben, and Luca dissect Claude Opus 4.8 — concluding it's more PR for Anthropic's fundraise than a technical leap — then debate whether "token maxing" is a productivity strategy or a vanity metric. The crew runs the numbers on the SpaceX IPO bear case: a $1.75T valuation at 100x revenue, a TAM bigger than China's GDP, and looming insider unlocks that will flood the market with supply. Luca closes with a science corner on using AI to analyze personal DNA data. Key takeaway: low float pumps SpaceX early, but the selling pressure that follows is historically unprecedented.
#Claude Opus 4.8
#token maxing
#SpaceX IPO
#agentic onboarding
#browser-in-agent
#GPU rental prices
#AI compute costs
#Anthropic fundraise
#DNA sequencing AI
#frontier model selection
#OpenAI Codex
#AI productivity metrics
#IPO bear case
#SaaS future
#Anthropic
#GPU prices
#browser in Codex
#AI agents
#DNA sequencing
#OpenAI
#Codex
#Replit
#frontier models
#SaaS
#compute costs
2 minute taster
SpaceX IPO Bear Case: The Numbers Nobody Wants to Run
Three 2026 AI IPOs plan to raise $550B — more than all US IPOs from 2015 to 2025 combined. SpaceX enters at 100x revenue (NVIDIA trades at 25x), claims a TAM larger than China's GDP, and faces looming insider unlocks starting day 45. The emotional FOMO is real, but so is the math.
Browser Inside Codex: The Agentic Future of Computing
The next paradigm isn't using AI alongside your browser — it's your browser living inside your AI agent. Log into Supabase once and every future project can use it automatically. Your agent watches you work in Notion, answers questions about what you've done before, and turns one-time workflows into repeatable automations.
Look closer
Claude Opus 4.8 just dropped, Anthropic is renting SpaceX GPUs at $8/hour, and SpaceX filed toward the biggest IPO in history. Plus token maxing, agentic onboarding, and analyzing your own DNA with AI.
Chapter list
- Token maxing
- The practice of deliberately spending as many AI model tokens as possible within a subscription period, treating maximum usage as a productivity or status goal.
- Agentic onboarding
- A future UX pattern where users paste a prompt into an AI agent environment to set up their profile in an app, replacing traditional form-filling flows.
- Agent swarm
- A coordinated group of AI sub-agents launched in parallel to execute many tasks simultaneously, dramatically increasing token consumption speed.
- Colossus
- A massive GPU cluster operated by xAI/SpaceX (Colossus 1 has ~100,000 H100 GPUs); Anthropic signed a deal to access compute from both Colossus 1 and 2.
- H200
- NVIDIA's high-end GPU model used for AI inference and training; rental prices on the spot market are a key indicator of AI compute demand.
- Blackwell
- NVIDIA's next-generation GPU architecture succeeding the H100/H200 series, expected to provide substantially higher AI compute performance.
- TAM (Total Addressable Market)
- The total revenue opportunity for a product or company if it captured 100% of its target market; SpaceX claims a $28.5 trillion TAM in its S-1 filing.
- S-1
- The IPO registration statement filed with the SEC disclosing a company's financials, business model, and risk factors before going public.
- Float (stock market)
- The proportion of a company's shares available for public trading on the open market; a low float can amplify price movements on IPO day.
- EWS (Elon Web Services)
- A speculative narrative comparing SpaceX/xAI's emerging cloud and GPU services to Amazon Web Services, positioning Elon Musk's companies as a tech infrastructure provider.
- VCF (Variant Call Format)
- A genomics file format that records the specific differences between an individual's DNA sequence and a reference genome, used to identify genetic variants.
- BAM file
- A compressed binary genomics file containing aligned DNA sequencing reads with quality scores, produced after initial processing of raw sequencing data.
- FASTQ
- The raw output file from DNA sequencing machines, containing nucleotide sequences along with per-base quality scores before alignment and processing.
- Whole genome sequencing
- A laboratory process that reads the complete DNA sequence of an organism's genome, generating raw data files that require multi-step computational processing.
- Allele
- One of two or more variants of a gene at a specific chromosomal location; identifying your alleles helps predict genetic traits or disease predispositions.
- Dead Internet theory
- The idea that most online content is now AI-generated or bot-driven, making pre-2021 human-created content increasingly valuable as authentic signal.
- Metcalfe's law
- The principle that a network's value grows proportionally to the square of its users; referenced in the episode as context for why AI agent networks compound in value.
- Return on token spend
- A proposed metric for evaluating AI usage efficiency: measuring measurable business outcomes (e.g. SEO clicks, shipped features) per token consumed.
- SaaSpocalypse
- The feared mass collapse of SaaS businesses due to AI making software trivially easy to build; debated in the episode as overstated.
