People Want Jobs, Not UBI

52% of ~1,000 respondents across 60 countries prefer guaranteed jobs vs 39% for guaranteed income; 40% believe their job is likely to be automated within 10 years while only 21% think it should be. Nvidia reports a staggeringly high profit margin of 55.8%, drawing competitors like Fractile, which claims an inference chip with ~100x more on‑chip memory to target long‑context AI models. Dozens of civil‑society groups urged Meta to drop facial‑recognition in RayBan glasses, citing significant privacy and safety risks.

Analysis

A memory-centric inference architecture emerging from startups creates a structural fork in the AI hardware roadmap: incumbents win on software ecosystem and scale, challengers win on cost/perf for long-context and edge inference. That bifurcation will manifest as a reallocation of capex at hyperscalers — fewer generic GPU hours, more domain-specific ASIC slots — shifting demand from DRAM & off-chip interconnects toward on-chip SRAM and advanced SoC integration in the 12–36 month window.

The largest near-term reversal risk is software portability: a small set of widely adopted runtimes or a hyperscaler endorsement would collapse adoption friction and re-center workloads on incumbent GPUs within 6–12 months. Regulatory and consumer-privacy headwinds around wearable AI features create a separate revenue and reputational drag for platform owners that try hardware-first consumer plays; expect ambiguity to persist and potentially depress product adoption and incremental ad monetization over the next 3–18 months.

For portfolio positioning, bifurcation favors (A) option structures that capture multi-year secular upside to incumbents while limiting near-term gamma; (B) idiosyncratic long exposure to foundries and memory suppliers that benefit from heterogeneous SoC demand; and (C) asymmetric short/put exposure to consumer platforms that accelerate privacy-sensitive hardware deployment without clear monetization. Monitor three triggers for re-pricing: 1) published independent inference benchmarks on cost-per-token, 2) hyperscaler platform endorsements or SDK commitments, 3) major privacy/regulatory actions or class-action suits tied to biometric/face-ID features.

Key Decisions for Investors

• Long NVDA via 12–24 month LEAP calls (size 3–5% NAV) to capture continued GPU demand for training; finance by selling 1–3 month call spreads to reduce premium. Risk/reward: limited premium loss if incumbents lose a small share; >2x payoff if NVDA retains platform dominance or captures new data-center wallet share within 18 months.
• Short META via 6–12 month put spread (buy puts / sell lower-strike puts) sized 2–4% NAV to express regulatory and monetization execution risk on hardware-led consumer initiatives. Risk/reward: capped loss (premium paid) vs 3–4x potential if regulatory/ad-revenue setbacks materialize over next 9–18 months.
• Long foundry/memory exposure (TSM or ASML via 12–24 month calls or equity overweight) to play structural SoC integration demand as specialized chips scale. Risk/reward: steady demand even if a single chip supplier fails; downside if fab capacity is reallocated or macro capex stalls — hedge with short cyclical semi suppliers.
• Allocate 1–2% NAV to private or secondary stakes in emerging inference-accelerator startups (or listed small-cap peers) with software compatibility roadmaps; tender to win design slots at hyperscalers. Risk/reward: high idiosyncratic risk but >5x upside on design wins within 18–36 months; liquidity risk and technical risk to mitigate with staged funding and milestones.