Anthropic’s $30B Ramp, Mythos Doomsday, OpenClaw Ankled, Iran War Ceasefire, Israel's Influence
Summary
Anthropic’s reported annualized revenue surged from $1 billion at the end of 2024 to $30 billion by late March or April 2026, which Jason framed as turning AI demand from a question mark into an exclamation point. More than 1,000 enterprises reportedly spend over $1 million annually, while Brad Gerstner sees an $80 billion-$100 billion exit run rate this year. His call: intelligence is a “near infinite TAM,” and Anthropic remains compute-constrained at roughly 1.5-2 GW.
That extraordinary ramp still does not establish durable profitability or even perfectly comparable revenue. Chamath Palihapitiya placed the industry “between gross revenue and net revenue,” while Brad argued distribution commissions affect only a single-digit share and could be normalized by 5%-10%. Brad sees rapidly rising gross margins and possible “accidental profitability”; Chamath’s counter is that the incumbents with fortress balance sheets—Meta, Google and potentially SpaceX—will deliberately make AI a capital-intensive compute contest.
Anthropic’s decision to withhold Mythos creates a credible one-time cyber-defense window, even if fear is part of its launch playbook. The model reportedly found thousands of vulnerabilities, including 27-year-old OpenBSD and 16-year-old FFmpeg bugs, and can chain three to five weaknesses into an exploit. David Sacks called prior Anthropic scare research embarrassing but judged this case legitimate: “We have no choice but to treat it that way,” especially if Chinese open-source models such as Kimi K2 are only six months behind.
The OpenClaw dispute is an early test of whether a leading model supplier will price first-party and third-party agents fairly. OpenClaw power users allegedly consumed $2,000-$20,000 of tokens through $200 subscriptions, prompting Anthropic to require them to move to metered API access before unveiling its own managed-agent product. Brad called that rational repricing—Anthropic had been “selling dollars for 10 cents”—while Sacks and Brad noted that retaining flat-rate pricing for Anthropic’s agent while metering OpenClaw could support a bundling or discrimination case.
Anthropic may supply 50%-60% of today’s coding tokens, but AI-generated code still represents only about 5% of the broader coding market and faces decades of enterprise tech debt. David Friedberg said long-horizon, enterprise-grade generation remains “shit,” citing a $100 billion-revenue company that still recruits 60-year-old pensioners to interpret COBOL. Brad expects AI-generated code to rise from roughly 5% to 95% over the next few years, creating a potentially powerful code-data-agent flywheel for the early leader.
Open source threatens model-training economics sooner than it displaces trusted enterprise systems. Jason Calacanis highlighted Bittensor subnet 62’s Ridges AI, which allegedly reached 80% of Claude 4 after 45 days and roughly $1 million in TAO rewards; Brad sees distributed pre-training as an orthogonal attack on models requiring $10 billion-$20 billion checks. Their disagreement is where adoption stops: Jason invoked Linux, Kubernetes and PostgreSQL, while Brad assigned “probability zero” to an important company outsourcing its production codebase to an open-source project.
Markets treated the Iran war as a bounded shock, but the bullish outcome depends on ceasefires becoming durable settlements. Brad contrasted a roughly 5%-7% S&P/Nasdaq drawdown with the Nasdaq’s 22% tariff-era intraday decline and said agreements involving Iran, Lebanon, Ukraine-Russia, Venezuela and possibly Cuba could put markets “off to the races” before July 4. The political risk is erosion of U.S. support for Israel: Chamath argued Israel needs an offramp more than America does, while Sacks pointed to Naftali Bennett’s concern over deteriorating U.S. polling.
Deep dive
Not yet available upstream; scheduled sync will retry.