Alibaba announced today it has open-sourced the Qwen3.8 model series; developers, research institutions and companies can freely download, deploy and use the models. The newly released Qwen3.8-27B is a native multimodal dense model with 27 billion parameters and is said to outperform Qwen3.7-Plus.

CME plans to launch H100 and B200 compute futures on Oct. 5, subject to regulatory approval. Contracts settle to Silicon Data’s GPU hourly rental rate; each contract represents one month of rental. BCG’s research arm says the near-term effect would be a public compute forward curve: AI firms could lock costs, cloud providers could hedge rental declines, and banks could better project data‑centre cash flows — potentially lowering finance costs and prompting more compute investment. Forward prices could become a near‑real‑time gauge of AI demand. H100 hourly rent fell from roughly $8 to $1.96, showing tangible hedging demand. Futures themselves transfer price risk rather than create credit risk, so they differ from 2008‑style mortgage securitization. The main systemic risk is if compute futures become the pricing backbone for GPU‑backed loans, data‑centre valuations or securitisations — that could lower lending thresholds, amplify leverage and, if rents drop, trigger collateral writedowns, margin calls and credit contraction. Standardisation also introduces basis risk: regional, network and service‑quality differences may leave significant spreads between the index and actual rents.

Fitch expects the U.S. debt ceiling to be reached around mid-2027.