Frontier Lab Research Team Learns Model Release Date From Compute Trading Desk Calendar

SAN FRANCISCO—A leading artificial intelligence research laboratory postponed Monday's scheduled release of its next-generation language model after its compute trading desk determined that current H100 forward contracts made an immediate launch "structurally inadvisable," according to internal communications reviewed by The Artifice.

The delay—the fourth in six months for a model the research team has called deployment-ready since February—came one trading day before 14,000 H100 futures contracts held by the lab's infrastructure hedging unit were set to settle at $1.71 per hour, a position projected to generate a $47 million mark-to-market loss if release-day inference demand failed to move spot prices.

"Our release schedule is governed entirely by research quality and safety evaluations," a company spokesperson said, in a statement that did not address the futures position.

The lab's compute trading division—eleven quantitative analysts hired last January from Citadel and Two Sigma—now holds combined notional exposure in H100, H200, and B200 forward contracts exceeding $900 million, a figure larger than the lab's entire 2024 research budget. Three researchers who have waited eighteen months for compute allocation to run ablation studies said they learned of the trading desk's existence from this article.

The model was rescheduled for Thursday, then moved again to the following Tuesday after the desk opened a long position in H200s expiring August 15.

"We are deeply committed to the responsible deployment of transformative AI," the CEO said in remarks at the Aspen Ideas Festival, where he headlined a panel titled 'Speed and Safety: False Dichotomy?'

A researcher who asked not to be named said her fine-tuning run had been allocated compute "after Q3 settlement."