Tag
Briefings and ideas about Inference costs from AI podcast discussions and the research they cite, with links to the original sources.
Ramez Naam passed on Panthalassa's early Bitcoin pitch but has since invested in its wave-powered, seawater-cooled ocean AI computers. Its 2-cent/kWh power cost is a target; northern-Pacific pilots are planned for 2026.
Ramez Naam predicts AI-written kernels could erode NVIDIA's CUDA lock-in in 2026–2027, easing switches to rival chips. Panelists noted NVIDIA's fast chip-to-chip links still matter, even as more work shifts to inference.
BDH-CQ's authors report solving 118 of 400 public ARC-AGI-1 tasks, with up to two answers each, at an estimated $0.00070 inference cost per task. Emad Mostaque welcomed the experiment; Alex doubted it shows progress beyond one benchmark.
Higgsfield's AI feature The Cully Hill Boys reportedly cost about $2 million, half on compute; on Moonshots, Emad Mostaque forecast compute for a similar film could probably fall to $100,000. Its production files are licensed for study only.
NVIDIA signed memorandums with six financial firms aiming to mobilize over $500 billion for customers' AI infrastructure. On Moonshots, panelists debated whether GPUs can back long-term financing before newer chips undercut their earnings.