Back

THE DAILY RECEIPTS

The Daily Receipts

Edition of 29 September 2026

This edition's sourced facts, the last close and the stock of the day — exactly as they were mailed.

Today's facts

LabourPrimary

A Census Bureau working paper tracking college graduates by major since ChatGPT's 2022 debut found the most AI-exposed decile of majors saw their odds of landing a first job fall five percentage points, while their full-quarter starting earnings fell 13% -- a decline comparable to graduating into a large recession

See the receipt →
Supply chainEstimate

US export controls are meant to keep advanced AI chips out of China, so Epoch AI compared Malaysian and Chinese customs records for the same server shipments between April 2024 and June 2025, and found a gap consistent with about $3 billion of AI chips smuggled to China through Malaysia, roughly 150,000 H100-equivalent GPUs

SourceEpoch AI·Sep 2026
See the receipt →
ComputePrimary

IREN converted former bitcoin-mining sites into GPU data centres for AI workloads, and when it reported fiscal 2026 results on 27 August 2026 it disclosed $4 billion in contracted 2026 AI Cloud revenue, with $1 billion already operating

SourceIREN Ltd·27 Aug 2026
See the receipt →
EnergyPrimary

Not every AI-era power saving comes from a new power plant -- Berkeley Lab reported in December 2025 that a two-year cooling retrofit at its NERSC supercomputing facility cut non-IT power draw, the electricity a data centre spends on cooling rather than compute, by 42%, saving more than 2 million kilowatt-hours a year

SourceBerkeley Lab·Dec 2025
See the receipt →
CapitalEstimate

A new NBER paper modeling the AI investment boom finds the five largest US technology firms spent $380 billion on capital expenditure in 2025, and are forecast, on the authors' industry-forecast calibration, to spend roughly $760 billion in 2026

SourceNBER (Wachter & Wachter)·Jun 2026
See the receipt →
ModelsPrimary

Epoch AI had frontier models repeatedly play the board game Earthborne Rangers to test whether they improve with practice, and found GPT-5.5 and Claude Opus 4.8 both mismanaged its core fatigue mechanic at 2.1 fatigue per round, versus 0.6 for expert humans and 3.5 for purely random play

SourceEpoch AI·July 2026
See the receipt →

The last close

Broad decline — 59 of 72 AI names lower — session 2026-09-28

Stock of the day

NVDA Nvidia, +2.1% today, at 229.79