Story Compute Infrastructure

'Ghost Ancestors' in Every Human Genome — and a Warning Shot on AI Spending

A genome study published this week found DNA from so-called 'ghost ancestors' in all living humans — not isolated populations, but universally. The designation applies when researchers detect a genetic signal distinct from any known archaic human lineage: not Neanderthal, not Denisovan, but something older and separate that interbred with Homo sapiens ancestors tens of thousands of years ago and left no identified fossil record. Finding the signal universally, rather than in geographically confined populations, suggests multiple interbreeding events across different migration routes and pushes against a clean 'out of Africa' replacement model in favor of a far more reticulated picture of human origins.

Forecasters issued a separate deep-system warning this week: the 2026 El Niño event could break all previous intensity records. The economic costs would register primarily through agricultural disruption — altered rainfall patterns affect wheat production in Australia, coffee in Colombia, and rice across Southeast Asia — and through intensified wildfire and flooding risk across regions simultaneously. European economists published analysis arguing that the continent's heatwaves and wildfires are transitioning from weather events into macroeconomic threats, affecting insurance markets, agricultural output, and labor productivity in ways now visible in GDP calculations.

The 'What If We're Wrong?' challenge this week targeted the consensus view on Big Tech's $170 billion quarterly AI spending trajectory. The steelman counterargument runs in three parts. First, the assumption that AI capability gains translate to proportional revenue generation is unproven: monetization remains concentrated in enterprise productivity software, coding assistance, and search enhancement — real and growing, but not obviously at a pace justifying current expenditure rates indefinitely. Second, the winner-take-all framing may be historically off: IBM survived missing the PC era, and Microsoft remained dominant in enterprise despite badly missing mobile, suggesting AI competitive dynamics may ultimately resemble the cloud — a contestable market where multiple players coexist profitably — rather than the early internet's infrastructure-determines-everything logic.

The third counterargument concerns power. Data centers of the scale being announced — including DeepSeek's planned one-gigawatt facility in Inner Mongolia and Anthropic's $15 billion project — require electricity infrastructure that is itself constrained. If power capacity becomes the binding limit on compute expansion before revenue arrives, companies that over-committed to infrastructure could face stranded assets. The specific metrics to watch, according to this analysis: AI-attributed revenue growth rate versus capex growth rate — capex is currently growing faster, and that gap needs to close — and major power purchase agreements signed in the next six months. If those slow, the physical constraints may be biting before the business case fully materialized.