20VC: Anthropic Unveils Mythos | SpaceX's Financials Leaked: Is it Worth $2TRN | Meta Debuts Muse Spark: Are They Back in the AI Race | Jason's Critique of Dario Amodei & How OpenAI Could Win the Enterprise Game

What It Covers

Harry Stebbings, Rory O'Driscoll, and Jason Lemkin analyze five major tech stories: Anthropic's withheld Mythos model and its cybersecurity implications, SpaceX's leaked financials at a 108x revenue multiple, Meta's Muse Spark debut, OpenAI's $2.5B ad revenue projection, and why legacy SaaS companies are failing the only test that matters in the AI era.

Key Questions Answered

• •The 60% Agent Death Spiral: Legacy SaaS companies building AI agents that are only 60% as capable as standalone solutions cannot charge for them — customers will use them but refuse to pay a premium. A 60% product must be bundled free, meaning no revenue reacceleration. Companies like Salesforce and ServiceNow face slow decline unless their agents match or exceed the quality of Claude or purpose-built competitors. The test is simple: can you charge for it independently?
• •Cybersecurity Machine Gun Effect: Anthropic's Mythos model autonomously scans entire codebases and discovers zero-day vulnerabilities without human steering — the same task older models can do only with repeated manual prompting. The difference is speed and scale, analogous to a rifle versus a machine gun. Within days of MyFitnessPal acquiring Cal AI, 3.2 million user records were stolen due to missing Firebase authentication, illustrating how AI accelerates breach discovery across every site with PII.
• •Enterprise Is Two-Thirds of the AI Revenue Game: The conventional assumption that consumer AI drives the most value is inverted in this cycle. Consumers want Netflix at home; enterprises want intelligence at work. Enterprise likely represents two-thirds of total AI revenue opportunity versus one-third for consumer — the mirror opposite of the internet era. This reframes OpenAI's ad business: even a $100B ad business may be insufficient to support burn without a parallel, scaled enterprise revenue line.
• •OpenAI's Enterprise Path Requires Microsoft Reconciliation: OpenAI's traditional enterprise sales DNA — exemplified by a leaked internal memo from CRO Denise Dresser pushing direct Fortune 500 sales — positions it well as CIOs consolidate AI budgets top-down rather than letting developer teams choose tools organically. However, Microsoft remains the dominant enterprise distribution channel globally. Without repairing the OpenAI-Microsoft relationship, OpenAI cedes the most efficient path to standardized enterprise deployment across large organizations.
• •SpaceX $2T Valuation Requires Zero Discount Rate: SpaceX's leaked financials show $18.5B revenue with a $5B loss, placing the potential IPO at roughly 108x revenue — the highest revenue multiple at scale in IPO history. Reaching a $2T valuation mathematically requires assigning near-100% probability to future initiatives like space-based data centers and direct-to-cellular, with zero time-value discounting. Investors applying standard probability adjustments and NPV calculations arrive at materially lower numbers regardless of long-term upside belief.
• •PE Software Portfolios Face a Bounded but Urgent Test: Private equity firms holding mature SaaS companies like Coupa and Anaplan — bought at 10x revenue with leverage, now trading at 2-4x equivalents in public markets — face a clear binary outcome. If portfolio companies can build AI agents customers will pay for independently, growth reaccelerates and debt gets serviced. If they deliver only incremental 60% solutions, enterprise value after debt deduction approaches zero. Hiring external AI consultants without empowering internal engineering teams will not close the gap.