Signal creator Moxie Marlinspike wants to do for AI what he did for messaging

Moxie Marlinspike has launched Confer, an open-source AI assistant designed to cryptographically verify its stack and keep user data unreadable to platform operators, attackers, and law enforcement by encrypting conversations inside a trusted execution environment with keys retained on users' devices. The move addresses legal risks underscored by court orders forcing platforms to preserve user logs and the practice of human chat review, and could spur demand for privacy-first LLM deployments even as scalability and the trade-offs between public-cloud efficiency and local models remain unresolved.

Analysis

Market structure: Privacy-first LLMs like Confer tilt demand toward confidential computing, TEEs, and on-device inference for regulated verticals (healthcare, legal, psychotherapy). Winners include cloud vendors that already sell "confidential computing" (MSFT Azure, GOOGL Cloud), TEE/hardware sellers (INTC/AMD for CPU TEEs, NVDA for inference GPUs) and cybersecurity vendors that integrate key management; ad-driven social platforms (META) and pure-play API-data monetizers face revenue pressure for sensitive use cases. The immediate pricing effect will be modest but could shift incremental enterprise procurement away from raw API spend toward higher-margin, bundled confidential-cloud services over 12–36 months.

Risk assessment: Tail risks include government mandates for retained logging or compelled key disclosure that could nullify privacy claims, and discovery of TEE side-channel exploits that destroy trust—both low probability but high impact within 6–24 months. Short-term implementation risks: slow enterprise adoption, compute-cost ceilings and developer inertia; hidden dependencies: open-source community velocity, availability of high-memory GPUs and cooperation from CPU vendors. Key catalysts: court rulings on data preservation (30–90 days), major TEE vulnerability disclosure, or a high-profile data leak driving corporate procurement.

Trade implications: Tactical positions: overweight NVDA (NVDA) for sustained GPU demand and MSFT (MSFT)/GOOGL (GOOGL) for confidential-cloud revenue; underweight META (META) given potential ad targeting erosion. Option plays: buy 3–9 month NVDA call spreads to capture upside from supply tightness; buy 3–6 month puts on META as a hedge if ad revenue misses. Rotate into cybersecurity names (CRWD) selectively if earnings show ARR tailwinds; size initial longs 2–3% portfolio each, trims at +25% gains or 90 days.

Contrarian angles: The market may overstate an exodus from cloud — enterprise will likely prefer confidential cloud over fragmented local stacks, which benefits large cloud incumbents (MSFT/GOOGL) not niche OSS projects. Privacy-first LLMs could raise GPU prices and consolidate compute power with incumbents, a pro-NVDA outcome. Unintended consequence: regulation forcing backdoors would make privacy plays binary and volatile; position sizing should assume a 20–40% drawdown scenario over 6–12 months.

Key Decisions for Investors

• Establish a 2–3% long position in NVDA (NVIDIA) within 2 weeks to capture higher GPU demand from on-device/TEE LLMs; hedge with a 3–6 month 20–30% OTM call spread instead of naked calls if implied vol > 60%.
• Add 2% each to MSFT and GOOGL as 6–18 month plays on confidential-cloud services; target exits or re-evaluation at +30% price moves or after major court rulings on data preservation (monitor next 30–90 days).
• Trim exposure to META by 5–10% (or buy 3–6 month puts sized to reduce net exposure) anticipating weaker ad targeting for privacy-shifted users; reassess after quarterly ad-revenue prints or regulatory actions within 90 days.
• Initiate a 1–2% position in CRWD or another large-cap cybersecurity name to capture enterprise demand for key management/TEE integrations; set stop-loss at -20% and take-profit at +30% over 6–12 months.