The Daily Receipts

A new open index that rated all 18,796 O*NET job tasks and weighted them by employment in 146 countries estimates how much work AI agents could take on: in 2026 the share of task time in world employment that is at least partly delegable to AI agents is 35.2%, but only 2.5% can be handed over in full

Anthropic did not need the cash to buy its own AI chips outright: Epoch AI found that investors led by Apollo-managed funds, with Blackstone and global banks, committed debt to finance more than 1 GW of Google TPU systems for the lab, totalling $34.5 billion, of which $30 billion benefits from Broadcom's support

Answering one medium-length prompt of approximately 1,000 input and 1,000 output tokens takes very different amounts of energy depending on the model: benchmarking by Jegham et al. cited in Stanford's 2026 AI Index found DeepSeek V3.2 the most energy-hungry of 2025's top 15 models, at 23 Wh

Marvell agreed on 2 November 2026 to buy laser and optical-transceiver maker Lumentum in an all-stock deal that combines electrical and optical AI connectivity, at an implied enterprise value for Lumentum of $115 billion, a premium of approximately 23.9%

In the third quarter of 2026 Palantir's US commercial revenue overtook its US government revenue for the first time in the company's history, as US commercial revenue grew 148% year-over-year to $984 million, against $953 million from US government

A shortage of high-bandwidth memory is pushing up Chinese AI chip prices: Huawei reportedly told some customers it was raising the suggested price of its Ascend 950DT, its most powerful AI chip with roughly half the raw computing performance of Nvidia's 2022 H100, to more than RMB 250,000 (approximately US$37,000), as much as Nvidia's newer B200