Sleep is the near-term neurotechnology wedge because users already accept physiological measurement, while an eye mask reportedly measures REM, induces sleep-promoting eye movements, and amplifies REM, with related technology expected in roughly seven to 12 months. Huberman predicts Meta, OpenAI, Anthropic, and Elon’s Neuralink effort will converge on neural read/write biotech, but missing models of memory and thought, plus a possible 2027-28 downturn in optimization content, remain key risks.

AI seed investing remains viable as an option strategy, but Menlo Ventures concentrates only after revenue and quantifiable evidence emerge, amid neo-labs and $10–20 million application rounds. With metrics vulnerable to gaming and a 10% seed stake potentially diluting to 3.5–4%, Venky says venture must beat the “Mag Seven” by roughly 1,000 basis points; vintage risk remains.

X says the cost of producing a graduate has fallen roughly threefold over 16 years, while its funnel yields only two graduates from up to 200 code-named ideas annually. Protected governance and early kill discipline make that portfolio economically plausible, but AI cannot replace human-led acceptance and distribution, while clean water still must reach roughly one cent per liter and industrialization remains 95% of the work.

Zuckerberg’s durable edge is methodical, founder-controlled compounding: Meta can pair 20 years of distribution across its apps with owned infrastructure and attention, even as critics call it behind in AI. Luckey’s costly early Trump support and Wu’s apparent refusal to sell point to conviction and independence as differentiators, while defense technology’s record in Israel and Ukraine remains unresolved.

Gemini Robotics 2 advances whole-body control and cross-embodiment transfer, but Keerthana still calls robotics its “GPT-2 moment”: one-shot demonstrations do not yet prove reliable generalization across bodies and changing scenes. Google’s three-model hierarchy separates embodied reasoning, action, and local execution, pointing to controlled factories, warehouses, and professional servicing as early wedges while latency, hand economics, safety, and compounding handoff errors remain unresolved.

Memory—not raw compute—is AI’s deepest systems bottleneck, making lower-memory architectures a potential lever across bandwidth and networking. As 144-GPU systems push power and cooling higher and copper fails at 400 gigabits, optics, DRAM-on-logic and specialized inference architectures may gain relevance, while Anthropic’s S1 points to more than $500 billion in obligated capital and compute and possible over-ordering.