Most important takeaway

Anthropic’s revenue run rate has exploded to $30 billion by end of March 2026 (up from $9B at end of 2025), representing the fastest revenue ramp in technology history, fueled primarily by enterprise adoption of coding and co-work tools. This validates the thesis that the TAM for intelligence is fundamentally different from any prior technology market, and investor Brad Gerstner suggests Anthropic could exit the year at $80-100B in revenue.

Summary

Key Themes

Anthropic’s Mythos Model and Cybersecurity Watershed Anthropic withheld its newest model, Mythos, citing extreme cybersecurity risks. The model autonomously discovered thousands of software vulnerabilities, including decades-old bugs in OpenBSD, FFmpeg, and the Linux kernel. Rather than calling for government regulation, Anthropic formed “Project Glasswing,” a 100-day coalition with 40 major companies (Apple, Microsoft, Google, Amazon, JPMorgan) to find and patch vulnerabilities before release. Sacks noted Anthropic has a pattern of using fear to market new products (citing the 2019 GPT-2 rollout and a prior blackmail study), but conceded this cybersecurity concern is more legitimate since better coding models logically find more bugs. Chamath was more skeptical, calling it “mostly theater” but agreeing there’s no choice but to take it seriously.

OpenClaw Controversy and Competitive Dynamics Anthropic effectively “ankled” OpenClaw — the massively popular open-source coding agent — by cutting off flat-rate subscription access and forcing API (pay-per-token) pricing. Power users who were consuming $2,000-$20,000 worth of tokens on $200 subscriptions got repriced. Days later, Anthropic announced its own competing agent technology (Claude Managed Agents). Sacks warned that dominant coding market share companies should “keep their nose clean” on antitrust, noting that if Anthropic bundles its own agent at a flat rate while metering third-party access, it could face bundling/price-dumping scrutiny. Jason argued open source will eventually win 90% of token usage and that giving your entire business to any single frontier model company is foolish.

Anthropic’s $30B Revenue Ramp From zero revenue in early 2023 to a $30B annualized run rate by March 2026, with over 1,000 enterprises paying $1M+ annually. Brad Gerstner noted they added the equivalent of Databricks plus Palantir combined in a single month. Gross margins are reportedly surging (from meaningfully negative 18 months ago to potentially 50-60%+), partly because compute is the main cost and they can’t stand up capacity fast enough. The company has only 2,500 employees versus Google’s 120,000 when it crossed similar revenue thresholds.

Iran War Ceasefire and U.S. Foreign Policy A two-week ceasefire was achieved after Trump’s aggressive social media pressure on Iran. JD Vance and Jared Kushner headed to Islamabad for talks. Markets largely shrugged off the conflict (5-7% drawdown vs. 15% during tariff chaos). The New York Times reported that Netanyahu pitched Trump on attacking Iran at a February White House meeting, with Vance warning it could fracture the MAGA 2.0 coalition. There’s growing bipartisan concern about Israel’s influence on U.S. foreign policy, with former Israeli PM Naftali Bennett publicly flagging Israel’s declining popularity in the U.S. as a problem.

AI Value Capture Across the Stack The discussion explored where AI value accrues: chips (NVIDIA), hyperscalers, model layer (Anthropic/OpenAI), and application layer. Enterprise software stocks (IGV index) are down 30% YTD as VCs crowd into frontier models and “out of harm’s way” bets. Sacks noted the baby may be getting thrown out with the bathwater on some software companies.

Actionable Insights

• CISOs and IT leaders should immediately use advanced AI models to audit existing codebases for dormant vulnerabilities during this 6-month window before capabilities are widely available.
• Enterprise software investors should look for value buys in beaten-down software stocks where moats aren’t purely about code generation.
• Startup founders building in AI should consider open-source coding tools (e.g., Ridges AI on BitTensor) as alternatives to locked-in frontier model subscriptions, especially given pricing volatility.
• Portfolio managers should watch for “accidental profitability” at frontier model companies as revenue scales faster than compute capacity can be deployed.
• Anyone using AI coding agents should avoid over-reliance on a single provider and consider the antitrust and pricing risks of platform lock-in, especially as companies bundle agent capabilities.

Chapter Summaries

Intro / Retard Maxing Discussion — Light banter about Eli Shalong’s philosophy of detachment and not ruminating. David Freeberg is out; Brad Gerstner fills in as guest bestie.

Anthropic’s Mythos Model Withholding — Deep discussion of Anthropic’s decision not to release Mythos due to cybersecurity risks. Brad credits the self-regulation approach. Sacks warns about Anthropic’s pattern of fear-based marketing but acknowledges the cyber concern is more legitimate. Chamath calls it mostly theater. Consensus: treat it seriously regardless, use the window to patch systems.

OpenClaw Gets Ankled — Anthropic cut off flat-rate access for OpenClaw users, forcing API pricing. Jason sees this as coordinated effort across the industry (Anthropic, OpenAI, Perplexity, Amazon, Apple) to neutralize the open-source coding agent threat. Sacks warns about clean antitrust behavior. Chamath argues the market is still early and small.

Anthropic’s $30B Revenue Ramp — The fastest revenue ramp in tech history. Brad details the trajectory and argues gross margins are exploding. Chamath pushes back on profitability metrics, noting we’re still in the “gross revenue vs. net revenue” discussion phase. Debate over whether Anthropic reports gross vs. OpenAI reporting net.

AI Market Competition: OpenAI vs. Anthropic — Brad credits Anthropic for coming from behind but warns against counting out OpenAI, citing their upcoming Spud model. Meta, Google, and xAI are all competitive. “This is not zero-sum” — the TAM for intelligence dwarfs anything seen before.

Iran War, Ceasefire, and Israel’s Influence — Two-week ceasefire achieved. Markets bounced back. NYT report on Netanyahu’s White House pitch to Trump. Vance and Rubio pushed back internally. Growing U.S. concern about Israeli influence on foreign policy. Bennett publicly flagged the problem. Brad remains optimistic about broader Gulf transformation.

X Auto-Translate Feature — Enthusiastic discussion of X/Twitter’s auto-translate feature powered by Grok, enabling cross-language engagement (Japanese, Hebrew, Chinese, Russian). Jason calls it potentially Nobel Peace Prize-worthy for cross-border understanding.

Liquidity Summit / All-In Summit Updates — Liquidity event sold out with hundreds on waitlist. Speakers include Bill Ackman, Andrej Karpathy, Dan Loeb, Brad Gerstner, Sarah Fryer. All-In Summit in LA September 13-15.