Anthropic made Opus dumber (73% Drop), Claude Code Hits $2.5B ARR & Tesla AI5 | God Mode Pod EP09
Claude Code hit $2.5B ARR in just nine months — the fastest B2B SaaS growth ever recorded — while Anthropic quietly slashed Opus reasoning depth by 73% to save compute.
God Mode Podcast
Anthropic made Opus dumber (73% Drop), Claude Code Hits $2.5B ARR & Tesla AI5 | God Mode Pod EP09
Claude Code hit $2.5B ARR in just nine months — the fastest B2B SaaS growth ever recorded — while Anthropic quietly slashed Opus reasoning depth by 73% to save compute.
TL;DR
Three hosts — Rik, Ben, and Luca — break down a dense three-week Anthropic news cycle on the God Mode Pod. Claude Code crossed $2.5B ARR in nine months, the fastest B2B SaaS growth ever. Claude Opus was quietly "nerfed" with a 73% drop in reasoning depth. Tesla taped out its AI5 chip at 2,500 TOPS. The OpenAI CRO memo leak reveals enterprise ambitions, and Allbirds' 910% single-day AI-pivot pop signals peak AI market frothiness. Key takeaway: compute constraints are now the defining bottleneck shaping every major AI product decision.
Rik, Ben, and Luca break down three weeks of Anthropic news: Claude Code hitting $2.5B ARR in nine months (fastest B2B SaaS growth ever), the 73% drop in Claude Opus reasoning depth, the new Claude Code desktop redesign, Tesla AI5 chip tape-out (2,500 TOPS), the OpenAI CRO memo leak, Allbirds' 910% AI-pivot stock pop, and a 24-year-old ex-OpenAI researcher turning $254M into $5.5B.
- ARR
- Annual Recurring Revenue — a SaaS metric representing the annualized value of subscription contracts, used here to describe Claude Code's $2.5B revenue run rate.
- Tape out
- The final step in chip design when the completed layout is sent to a semiconductor foundry for fabrication; Tesla's AI5 chip reaching tape out means mass production can begin.
- TOPS
- Trillions of Operations Per Second — a measure of AI chip compute throughput; Tesla's AI5 has 2,500 TOPS vs. 144 TOPS for its 2019 hardware.
- MCP
- Model Context Protocol — an open standard for connecting AI models to external data sources and tools, used extensively in Claude Code integrations.
- DEX
- Decentralized Exchange — a peer-to-peer crypto trading platform without a central authority; Jupiter is a DEX on the Solana blockchain that also trades pre-IPO tokenized equities.
- FSD
- Full Self-Driving — Tesla's autonomous driving software suite, which the AI5 chip is designed to power.
- Hallucination
- When an AI model confidently generates factually incorrect or fabricated information; Grok was cited as having a 20.2% hallucination rate in Vectara's benchmark.
- Innovator's dilemma
- A theory by Clayton Christensen describing how successful companies struggle to adopt disruptive innovations that threaten their core business; applied here to Google protecting its ad-driven search revenue.
- LiDAR
- Light Detection and Ranging — a sensor technology that maps environments using laser pulses; Waymo uses LiDAR maps while Tesla deliberately avoids this approach.
- Nerfing
- Reducing a feature's power or performance, often quietly; used to describe Anthropic cutting Claude Opus's reasoning depth by 73% to conserve compute.
- Harness
- A software wrapper or orchestration layer that routes tasks to the best-suited AI model; Perplexity is cited as an example of a multi-model harness.
- Cron job
- A scheduled automated task that runs at set intervals; Anthropic's new Claude Code 'Routines' feature is described as cron jobs for AI agents.
- FUD
- Fear, Uncertainty, and Doubt — market sentiment term describing negative speculation that drives asset prices down; used here in context of recent stock market volatility.
- Pre-market shares
- Equity in a private company traded on secondary platforms before an IPO; Anthropic's pre-IPO shares were traded on the Jupiter Solana DEX.
- Effective altruism
- A philosophy prioritizing evidence-based approaches to doing the most good; Luca argues Anthropic uses EA framing as a marketing narrative to justify withholding Mythos.
- Compute-constrained
- A state where a company's growth or feature rollout is limited by the amount of available GPU/TPU computing power rather than software or demand.
- Frothiness
- Excessive speculative enthusiasm in financial markets leading to overvalued assets; used to describe AI stock mania exemplified by Allbirds' 910% single-day surge.
- Diminishing returns
- An economic principle where additional inputs yield progressively smaller outputs; discussed in the context of how much memory context improves Tesla FSD reliability.
Chapter 1 · 00:00
Intro & Market Check: S&P All-Time High and AI Stocks
Coding an app is the new starting a podcast.
Chapter 2 · 03:20
Anthropic Pre-IPO Valuation: $900B on Jupiter DEX
SBF's 7.5% stake in Anthropic would be worth over $60 billion at the current $900B valuation — more than Coinbase's entire market cap — but was liquidated for ~$1.5B.
SBF's 7.5% stake in Anthropic would be worth over $60 billion at the current $900B valuation — more than Coinbase's entire market cap — but was liquidated for ~$1.5B.
Chapter 3 · 05:52
Claude Code Hits $2.5B ARR in Nine Months
Claude Code made $2.5B in 9 months. Snowflake took 9 YEARS to make $1B.
Claude Code hit $2.5B ARR in nine months — faster than Snowflake (9 years) and Figma (13 years) to their first billion. This isn't just a product win; it's proof that AI is eating the entire software stack, exactly as Naval predicted.
Claude Code reached $2.5 billion in annualized recurring revenue just nine months after launch, making it the fastest B2B SaaS product to reach that milestone ever.
Anthropic's pre-IPO shares on the Jupiter Solana DEX surged ~640% from $122 per share in October 2025 to a valuation of roughly $900 billion by April 2026.
Figma needed 13 years to reach $1 billion in ARR, while Snowflake needed 9 years — making Claude Code's 9-month trajectory unprecedented.
Anthropic operated with a single-person marketing team for its first three years, yet achieved $900B valuation — contrasting sharply with Lovable's heavy YouTube ad spend.
Chapter 4 · 10:50
Anthropic vs Lovable: Tesla Playbook vs Nvidia Playbook
Should Anthropic be the next Tesla or the next Nvidia? The answer changes everything.
Rik and Luca debate two competing visions for Anthropic's future. Rik argues Anthropic should expand like Tesla — best product, no marketing, now going direct to consumers. Luca says follow Nvidia instead: do one hard thing better than anyone and build an ecosystem of partners.
Chapter 5 · 17:30
Perplexity's Multi-Model Moat and IDE Future
Luca's bold prediction: Anthropic faces an antitrust case within 2 years. Here's why.
Luca predicts an Anthropic antitrust case within 1-2 years: subsidizing Claude Code for subscribers while charging competitors full API rates is the same playbook that triggered cases against Microsoft and Google. History shows companies usually navigate through — but the case will be made.
Three phases of AI value extraction: first Nvidia won, then the providers. Now everyone is asking — will the application layer also extract value? Ben argues providers still dominate because they can subsidize their own tools while application companies like Lovable pay full API rates to the same providers they compete with.
Perplexity doubled its ARR to $450M in a single month by being model-agnostic — routing to Claude for coding, ChatGPT for general tasks, Grok for real-time news. The CEO calls it 'no horse in this race.' Ben says it might be the strongest strategic position in the whole AI stack.
Perplexity doubled its ARR from approximately $200 million to $450 million in a single month, fueled by its multi-model 'harness' strategy.
Chapter 6 · 23:00
Claude Code Desktop Redesign Deep Dive
Anthropic's new desktop Claude Code app might finally be the IDE killer Boris Cherny promised.
Anthropic's new Claude Code desktop app organizes projects by folder with pinned threads, runs multiple agents in parallel, integrates MCPs locally, and auto-creates GitHub branches and PRs. It's more normie-friendly than Cursor but slightly behind on raw feature depth.
Chapter 7 · 25:40
Claude Opus 73% Reasoning Drop: Nerfed for Compute
Anthropic made Claude 73% dumber — and didn't tell anyone.
An AMD director analyzed 7,000 Claude Opus sessions and found a 73% drop in reasoning depth between January and March 2026. Anthropic didn't change the model — they throttled how much it thinks to save compute. Power users noticed.
An AMD director analysis of 7,000 Claude Opus sessions found a 73% reduction in reasoning depth between January and March — Anthropic quietly throttled thinking to conserve compute.
Chapter 9 · 33:40
OpenAI Codex vs Claude Code: Switching and Memory Portability
Open-router data showed OpenClaw token usage dropped roughly 60% from its peak of ~900 billion tokens to under 400 billion, likely as Anthropic's native Claude Code replaced it.
Open-router data showed OpenClaw token usage dropped roughly 60% from its peak of ~900 billion tokens to under 400 billion, likely as Anthropic's native Claude Code replaced it.
Chapter 11 · 41:00
Mythos Pricing Speculation and Anthropic's Marketing Strategy
'The world isn't ready for Mythos' is PR spin for 'we can't afford to give it to you.'
Luca argues Anthropic's refusal to release Mythos publicly — framed as waiting until 'the world is ready' — is just effective altruism marketing spin for 'we don't have enough compute.' The strategy works because it makes the model sound dangerous and desirable.
Google needs you to click links, see ads, and browse the open web. But when 90% of internet traffic flows through AI agents, Google's entire ad-revenue machine breaks. Google knows it — and that's why they keep implementing others' features but won't go full computer-use.
Chapter 13 · 47:20
Grok Takes #1 in Medicine — But Has the Highest Hallucination Rate
Grok is #1 in medicine AI benchmarks — and also hallucinates more than any other frontier model.
Grok topped the medicine text arena benchmark, but Vectara's separate test showed it has a 20.2% hallucination rate — the highest of any frontier model. Healthcare is the last place you want an AI that confidently makes things up.
Vectara's hallucination benchmark found Grok had a 20.2% hallucination rate — the highest of any frontier model tested — undermining its #1 medicine benchmark ranking.
Chapter 14 · 49:00
Tesla AI5 Chip Tape-Out: 2,500 TOPS and the Road to Autonomy
Tesla's new AI chip is 17x more powerful than 2019. Autonomy is now a hardware question.
Tesla's AI5 chip just taped out with 2,500 trillion operations per second — 17x the compute of its 2019 debut chip. One AI5 chip outperforms two AI4s combined, and it's the chip Tesla says will finally get them to full autonomy and power the Optimus robot.
A single Tesla AI5 chip is more powerful than two Tesla AI4 chips combined, enabling the next generation of Full Self Driving and Optimus robots.
Tesla's AI5 chip has 2,500 trillion operations per second, up from 144 TOPS in 2019's hardware 3 — a 17x improvement over six years.
Ben explains why Elon refuses to use pre-mapped LiDAR routes: what happens when a city changes its roads? Tesla's vision-only approach needs the AI5 chip's 2,500 TOPS to build real-time simulations of every nearby object rather than relying on a fixed map of the world.
From Tesla hardware 3 (144 TOPS in 2019) to AI5 (2,500 TOPS), Tesla's in-car AI compute has grown 17x, enabling higher-resolution real-world understanding for FSD.
Chapter 15 · 57:00
24-Year-Old Ex-OpenAI Researcher's $254M to $5.5B Fund
A 24-year-old with zero finance experience turned $254M into $5.5B in one year. His edge: he knew AI.
A 24-year-old ex-OpenAI researcher with no fund experience grew his portfolio 20x in one year by betting on AI compute and memory — SanDisk up 817%, CoreWeave up 670%. His edge wasn't financial expertise; it was knowing what makes AI models run.
A 24-year-old ex-OpenAI researcher's fund grew from $254 million in Q4 2024 to $5.5 billion in Q4 2025 — roughly a 20x return — betting on AI compute and memory plays.
SanDisk, a memory chip company, was up 817% in the young fund manager's portfolio and represented 6% of his fund holdings in Q4 2025.
Chapter 16 · 1:00:00
Allbirds 910% AI Pivot: Peak Frothiness and Meme Stock Warning
A near-bankrupt shoe company said 'AI' and went up 910% in one day. We've seen this before.
Allbirds stock surged 910% in one day on an AI pivot announcement — for a company that was selling all assets for $39M and was weeks from Nasdaq delisting. Ben draws a direct line to the 2017 Long Island Iced Tea blockchain pump. History rhymes.
Allbirds stock surged 910% in a single day, adding $143 million in market cap, after the near-bankrupt shoe company announced an AI pivot — echoing Long Island Iced Tea's 380% blockchain-name pump in 2017.
Chapter 17 · 1:04:00
OpenClaw Agent Selling Pool Installations: AI Marketing in the Physical World
An AI agent found your home, rendered a pool in your backyard, and mailed you a postcard. No humans.
Someone built an OpenClaw agent that finds $500K-$1.2M homes without pools, renders a pool into their backyard image using AI, and physically mails a before-and-after postcard. No humans involved. This is AI marketing crossing from digital into the physical world.
No indexed bits in this chapter.
Show stoppers
An AMD director analyzed 7,000 Claude Opus sessions and found a 73% drop in reasoning depth between January and March 2026. Anthropic didn't change the model — they throttled how much it thinks to save compute. Power users noticed.
Claude Code hit $2.5B ARR in nine months — faster than Snowflake (9 years) and Figma (13 years) to their first billion. This isn't just a product win; it's proof that AI is eating the entire software stack, exactly as Naval predicted.
Snapshots ()
Key Quotes ()
This episode
Claims & Sources
Factual claims made this episode, and whether a source was named.
Claude Code reached $2.5 billion in annualized recurring revenue (ARR) after nine months on the market.
Figma took 13 years to reach $1 billion in ARR.
Snowflake took nine years to reach $1 billion in ARR.
There was a 73% drop in Claude Opus reasoning depth across 7,000 sessions between January and March 2026, measured by an AMD director.
Anthropic's pre-IPO share price on Jupiter DEX rose approximately 640% from $122 in October 2025 to imply a $900 billion valuation by April 2026.
SBF's approximately 7.5% stake in Anthropic would be worth over $60 billion at Anthropic's current valuation, more than Coinbase's entire market cap, but was liquidated for roughly $1.5 billion.
Tesla's AI5 chip delivers 2,500 TOPS, compared to 144 TOPS for the 2019 Hardware 3 chip — a 17x improvement over six years.
One Tesla AI5 chip is more powerful than two Tesla AI4 chips combined.
Grok had a 20.2% hallucination rate — the highest of any frontier AI model tested — according to Vectara's hallucination benchmark.
Perplexity doubled its ARR from approximately $200 million to $450 million in a single month.
Allbirds stock surged 910% in a single day after announcing an AI pivot, adding $143 million to its market cap, despite the company having agreed to sell all assets for $39 million.
Long Island Iced Tea company's stock rose 380% overnight in 2017 after adding 'blockchain' to its name; the company was subsequently delisted after an insider trading investigation.
A 24-year-old ex-OpenAI researcher's fund grew from $254 million in equity positions in Q4 2024 to $5.5 billion in Q4 2025 — approximately a 20x return — through concentrated AI compute and memory bets.
SanDisk was up 817% in the 24-year-old fund manager's portfolio and represented 6% of his Q4 2025 fund holdings.
OpenClaw token usage on OpenRouter dropped approximately 60% from its peak of ~900 billion tokens to around 377 billion tokens, correlating with Anthropic's Claude Code feature releases.
This episode
Cast
-
Central subject of the episode — discussed for Claude Code ARR, Opus reasoning nerf, Mythos withholding, desktop redesign, and pre-IPO valuation.
-
Track
Discussed for taping out the AI5 chip (2,500 TOPS), FSD progress, and the Optimus robot; also used as a marketing strategy analogy for Anthropic.
-
Discussed for the leaked CRO memo, Codex as a Claude competitor, Mythos/o3 model speculation, and enterprise strategy pivot.
-
App-layer coding tool discussed as both a partner-dependent-on-Anthropic and potential acquisition/kill target if Anthropic builds a competing product.
-
Highlighted for doubling ARR to $450M in a month via a model-agnostic 'harness' strategy that routes to the best AI for each task.
-
Track
Discussed for its new Gemini Skills browser feature, its innovator's dilemma around AI agents threatening ad revenue, and as an antitrust history precedent.
-
Track
Used as a strategic template for Anthropic (do one thing well, build ecosystem) and cited as first-phase AI value extractor; also historically outsourced Tesla's early AI chips.
-
App-layer coding platform Luca uses; discussed as a Lovable competitor that benefits from Anthropic models and has strong user stickiness.
-
Near-bankrupt shoe company whose stock surged 910% in one day after announcing an AI pivot, used as a peak-frothiness warning sign.
-
Solana-based decentralized exchange where Anthropic's pre-IPO shares are tokenized and traded, showing a ~640% valuation rise since October 2025.
-
Memory chip company cited as the best-performing S&P 500 stock two years running and an 817% winner in the 24-year-old fund manager's AI compute portfolio.
-
Contrasted with Tesla's vision-only FSD approach; Waymo uses LiDAR pre-mapped routes which Tesla's Elon Musk argues is the wrong long-term strategy.
-
Track
AI cloud compute company cited as up 670% in the 24-year-old fund manager's AI infrastructure portfolio.
-
Company that runs a hallucination benchmark for AI models; their data showed Grok had the highest hallucination rate at 20.2% of any frontier model tested.
-
Anthropic's AI coding tool that hit $2.5B ARR in nine months, also newly redesigned as a chat-first local desktop IDE.
-
xAI's model discussed for taking #1 in the medical text arena while simultaneously having the highest hallucination rate (20.2%) of any frontier model per Vectara.
-
OpenAI's coding agent discussed as a potential alternative to Claude Code, especially for execution and testing tasks where Opus handles planning.
-
Anthropic's next frontier model, withheld from general release — discussed for its pricing speculation, exclusive access controversy, and effective altruism framing.
Stats
Sponsors
Related episodes
No links parsed
We scan show notes for social handles, websites and apps. Nothing matched on this episode.
