Long-time Google DeepMind researcher David Silver leaves to found his own AI startup

David Silver, a long-time DeepMind reinforcement‑learning lead and co-author on key projects including AlphaGo, AlphaZero and Gemini, has left Google DeepMind to found London-based startup Ineffable Intelligence (formed Nov 2025; Silver appointed director Jan 16 per Companies House). The company is actively recruiting AI researchers and seeking venture capital, positioning itself to pursue “endlessly learning” superintelligence via reinforcement learning — a potentially high‑value, long‑term technological bet in an environment where peers (e.g., Safe Superintelligence) have raised multibillion dollar funding. Near term this is primarily a talent/competitive development with limited direct market impact, but it signals continued VC interest and potential future strategic disruption in advanced AI research.

Analysis

Market structure: Silver’s departure is a positive shock to AI compute and cloud infra demand, not to Google’s commercial moat. Winners: NVIDIA (NVDA), AMD, TSMC supply chain, and hyperscalers (GOOGL, MSFT, AMZN) that sell training cycles; losers are talent‑starved incumbents and low‑margin AI consultancies. Expect upward pricing power for high‑end GPUs and cloud spot instances over 6–18 months as startups scale RL experiments requiring simulators and persistent online training.

Risk assessment: Tail risks include a regulatory hit (EU/US AI safety rules or export controls) or an operational failure from a superintelligence prototype that triggers funding freezes—each could force >20% drawdowns in AI equities within 1–6 months. Hidden dependencies: compute availability (NVDA wafer/capacity), memory DRAM/GPU supply, and immigration/patent constraints; a micro‑node shortage can amplify valuation dispersion. Key catalysts: VC raises (next 3–12 months), marquee research demos, and UK policy incentives for AI hubs.

Trade implications: Direct play is long infra (NVDA 2–3% NAV, GOOGL/MSFT 1–2% NAV) with 6–12 month horizons; pair trade long NVDA vs short C3.ai (AI) to capture value shift from software multiples to hardware/cloud. Options: buy 9–12 month NVDA call spreads (15% ITM buy / 35% OTM sell) to express directional demand with capped cost; size entry within 2–8 weeks as hiring and funding announcements cluster.

Contrarian angle: Consensus treats departures as weakening incumbents; miss is that talent dispersion increases aggregate compute spend and hyperscaler revenue. Reaction is likely underdone for NVDA and overdone for pure‑SaaS AI names lacking proprietary models. Historical parallel: post‑2012 ML researcher exodus boosted cloud/GPU cycles and rewarded infra providers; unintended consequence is fragmented standards raising integration costs and regulatory attention, which could invert winners into short candidates over 12–24 months.

Key Decisions for Investors

• Establish a 2–3% NAV long position in NVIDIA (NVDA) over the next 2 weeks; express via 9–12 month call spread (buy ~15% ITM LEAP, sell ~35% OTM) to capture elevated compute demand from RL startups while capping premium.
• Add 1–2% NAV split long between Alphabet (GOOGL) and Microsoft (MSFT) over 4–12 months to play increased cloud revenue from private RL training; trim if cloud spot instance rates decline >15% quarter‑over‑quarter.
• Implement a pair trade: long NVDA (1.5–2% NAV) and short C3.ai (AI) (1–1.5% NAV) for a 3–9 month horizon, targeting capture of infra vs software multiple rotation; exit if C3.ai outperforms NVDA by >20% in two consecutive months.
• Allocate up to 0.25–0.5% NAV to participate in a seed/Series A round for Ineffable Intelligence if available, contingent on founder lockups, IP assignment clarity, and a post‑money valuation implying >3x return potential in 3–7 years.
• Buy a protective hedge: 6–12 month put (or put spread) on NVDA sized to 0.5% NAV to guard against a regulatory or safety shock that causes >15% downside within 3 months.