THE TAKE:
The real hardware story this week isn't Hadron's reactor video (vaporware positioning masquerading as vision) — it's that llama.cpp keeps shipping incremental builds while the actual inference optimization battle gets ignored in favor of flashy modular reactor press releases. Neko Health's $700M raise for body scanning is venture theater; what matters is whether anyone's actually solving the compute density problem for on-device AI, which these papers hint at (neural decoding with unlabeled data, efficient interaction methods) but the industry refuses to fund properly. Watch the foundation model infrastructure (Anthropic SDK updates, llama.cpp's steady work) — that's where the real constraints are being solved, not in billionaire energy fantasies or biotech fundraising rounds.