Overview of 20VC: Anthropic's $6BN Revenue Month | OpenAI Kills Sora & Hits $100M ARR on Ads | Oura Going Public & Whoop Raises at $10BN | Manus Founders Trapped in China & The Billionaire Tax: Anyone Left in California?
Harry Stebbings hosts a 50th-episode panel with Rory O'Driscoll and Jason Lemkin discussing the biggest tech/AI stories of the week. Major themes: Anthropic’s explosive revenue and leaked Mythos model, OpenAI product pivots and early ad traction, security risks in the agentic AI era, questionable revenue accounting and valuation playbooks, geopolitical risks for China-origin startups, and the effects of billionaire taxes and migration. The conversation mixes product, go‑to‑market, accounting, and macro/capital-structure angles — plus investor/founder advice.
News roundup (concise)
- Anthropic: Reported an extraordinary 28-day February with ~$6B revenue; accidental leak of Claude Mythos (reported ~10T-parameter model; ~3,000 unpublished assets) focused on cybersecurity. Leak caused short-term pullback in cybersecurity stocks.
- OpenAI: Killed Sora (video product) after low traction; early ChatGPT ads business reportedly ~$100M ARR — still nascent and needs to scale massively to compete with Google/Facebook ad revenue levels.
- SoftBank / Masa: SoftBank secured a ~$40B bridge loan to buy more OpenAI stock — large leverage, high risk if markets correct.
- Cybersecurity market: Stocks (CrowdStrike, Palo Alto, Zscaler, etc.) fell on Mythos news; panel argues sell-off was mostly knee-jerk.
- Emergent Labs & ARR questions: Fastest-to-$100M claims scrutinized; concerns about ARR recognition via trial billing and potential double counting of token-based revenue.
- Manus (acquired by Meta): Founders reportedly unable to leave China post-acquisition — raises geopolitical/business risk for China-origin startups and acquirers.
- Wearables / Consumer health: Oura IPO and Whoop raising at ~$10B discussed — consumer hardware/subscription durability and competitive dynamics compared to enterprise software.
- Layoffs/Entertainment: Epic Games layoffs highlight structural declines in some entertainment labor markets (and broader industry shifts).
- Migration & taxes: High-net-worth departures (e.g., Steve Jurvetson selling CA home) discussed in context of proposed “billionaire tax” and state-level moves (WA tax).
Key takeaways
- Agentic AI accelerates both capability and risk. Autonomous agents (always-on, goal-directed) will consume far more compute and create new, fast-moving security exposures (leaks, rogue agents) — speed amplifies mistakes.
- Anthropic is scaling revenue aggressively but accidental leaks expose operational security trade-offs; Mythos leak shows irony: cybersecurity model leaked via a security lapse.
- OpenAI’s consumer strategy is consolidating into two existential bets: (1) make ads work at large scale, or (2) double down on high-value coding/enterprise use cases. Killing Sora matches scarce-compute economics: video generation is very compute intensive with low direct revenue.
- Revenue/ARR reporting is messy across AI companies. Different recognition methods (trailing 4-week extrapolation, net vs gross reporting through partners, trial-billing recognition) can inflate comparables and create confusion—watch methodology, not headlines.
- Tranches and headline valuations: Tranched rounds (different prices to different investors) and gamified headline valuations are increasingly common; they provide optics but can create misaligned signals and future down‑round risk.
- Cybersecurity is entering a “golden age” of demand: agentic workflows increase attack surface and urgency for new defenses; short-term stock moves may be overreactions, but long-term opportunity looks solid.
- Geopolitics matters for startup exits and talent mobility. Deals involving China-origin teams raise new post-acquisition risks; acquirers must account for legal/regulatory and human‑capital constraints.
- Policy & mobility: State-level taxes targeting the wealthy (CA, WA) can change behavior of mobile ultra-high-net-worth individuals and affect state revenue and long-term fiscal dynamics.
Notable insights & quotes
- “We may be at the stage where we throw the humans under the bus, not the AI anymore.” — on shifting blame and risk attribution.
- “It’s a Ferrari engine in a shopping cart.” — on how great models are hampered by messy enterprise implementation.
- “Get the fuck over it.” — blunt take on expecting every company to be SaaS-style, lock-in ARRs.
- Practical investor math: Anthropic/OpenAI ARR extrapolations use trailing-week averages × 13 (13 four-week periods/year) as one reasonable smoothing approach.
Practical recommendations (for founders, CISOs, investors)
- Founders
- Be transparent about how you calculate ARR and SaaS-like metrics; document methodology (gross vs net, trailing windows, trial recognition).
- Build security and deployment processes early — agentic apps will magnify errors; restrict production access and staging leak risk.
- Don’t design your company for VC optics (headline valuation or ARR type); design it around customers and unit economics.
- Branding tip: consider a .tech domain for clarity if building a tech startup (host ad).
- CISOs / security teams
- Prepare for internal agent risks: agents running with broad privileges will require new detection/mitigation approaches (agent security, internal code-review automation, runtime controls).
- Start evaluating agent-aware security vendors now; acquisitions and consolidation are likely.
- Investors
- Scrutinize ARR math and unit economics before valuing AI firms — ask for trailing-week metrics, churn behavior, and trial billing treatment.
- Be cautious with tranched rounds/headline valuation games; blended cost basis matters for long-term returns.
- Consider geopolitical and talent mobility risk (e.g., China-related founder teams) when evaluating cross-border deals.
Actionable red flags to watch
- Sudden, large “ARR” announcements without clear accounting methodology.
- Companies relying heavily on compute‑intensive features (video generation) with low near-term monetization.
- Funded deals where key technical employees/founders are in or subject to restrictive jurisdictions (post‑acquisition operational risk).
- Any product/service that requires agent root-level access without strong runtime governance.
Bottom line
This episode is a wide-ranging checkup on where AI and tech markets are: rapid capability growth (Anthropic and agents), messy operational/security consequences (leaks, agent risks), consolidation of strategic bets (OpenAI focusing on ads and enterprise), and market behaviors (valuation optics, ARR accounting) that require careful due diligence. For founders and investors the core advice is practical: focus on customers and durable units economics, get security right early, and beware headline metrics and political/geographic tail risks.